Introduction
This book is the primary reference for the Rust programming language. It provides three kinds of material:
- Chapters that informally describe each language construct and their use.
- Chapters that informally describe the memory model, concurrency model, runtime services, linkage model, and debugging facilities.
- Appendix chapters providing rationale and references to languages that influenced the design.
Warning: This book is incomplete. Documenting everything takes a while. See the GitHub issues for what is not documented in this book.
Rust releases
Rust has a new language release every six weeks.
The first stable release of the language was Rust 1.0.0, followed by Rust 1.1.0 and so on.
Tools (rustc
, cargo
, etc.) and documentation (Standard library, this book, etc.) are released with the language release.
The latest release of this book, matching the latest Rust version, can always be found at https://doc.rust-lang.org/reference/. Prior versions can be found by adding the Rust version before the “reference” directory. For example, the Reference for Rust 1.49.0 is located at https://doc.rust-lang.org/1.49.0/reference/.
What The Reference is not
This book does not serve as an introduction to the language. Background familiarity with the language is assumed. A separate book is available to help acquire such background familiarity.
This book also does not serve as a reference to the standard library included in the language distribution. Those libraries are documented separately by extracting documentation attributes from their source code. Many of the features that one might expect to be language features are library features in Rust, so what you’re looking for may be there, not here.
Similarly, this book does not usually document the specifics of rustc
as a tool or of Cargo.
rustc
has its own book.
Cargo has a book that contains a reference.
There are a few pages such as linkage that still describe how rustc
works.
This book also only serves as a reference to what is available in stable Rust. For unstable features being worked on, see the Unstable Book.
Rust compilers, including rustc
, will perform optimizations.
The reference does not specify what optimizations are allowed or disallowed.
Instead, think of the compiled program as a black box.
You can only probe by running it, feeding it input and observing its output.
Everything that happens that way must conform to what the reference says.
Finally, this book is not normative.
It may include details that are specific to rustc
itself, and should not be taken as a specification for the Rust language.
We intend to produce such a book someday, and until then, the reference is the closest thing we have to one.
How to use this book
This book does not assume you are reading this book sequentially. Each chapter generally can be read standalone, but will cross-link to other chapters for facets of the language they refer to, but do not discuss.
There are two main ways to read this document.
The first is to answer a specific question.
If you know which chapter answers that question, you can jump to that chapter in the table of contents.
Otherwise, you can press s
or click the magnifying glass on the top bar to search for keywords related to your question.
For example, say you wanted to know when a temporary value created in a let statement is dropped.
If you didn’t already know that the lifetime of temporaries is defined in the expressions chapter, you could search “temporary let” and the first search result will take you to that section.
The second is to generally improve your knowledge of a facet of the language. In that case, just browse the table of contents until you see something you want to know more about, and just start reading. If a link looks interesting, click it, and read about that section.
That said, there is no wrong way to read this book. Read it however you feel helps you best.
Conventions
Like all technical books, this book has certain conventions in how it displays information. These conventions are documented here.
-
Statements that define a term contain that term in italics. Whenever that term is used outside of that chapter, it is usually a link to the section that has this definition.
An example term is an example of a term being defined.
-
Differences in the language by which edition the crate is compiled under are in a blockquote that start with the words “Edition differences:” in bold.
Edition differences: In the 2015 edition, this syntax is valid that is disallowed as of the 2018 edition.
-
Notes that contain useful information about the state of the book or point out useful, but mostly out of scope, information are in blockquotes that start with the word “Note:” in bold.
Note: This is an example note.
-
Warnings that show unsound behavior in the language or possibly confusing interactions of language features are in a special warning box.
Warning: This is an example warning.
-
Code snippets inline in the text are inside
<code>
tags.Longer code examples are in a syntax highlighted box that has controls for copying, executing, and showing hidden lines in the top right corner.
// This is a hidden line. fn main() { println!("This is a code example"); }
All examples are written for the latest edition unless otherwise stated.
-
The grammar and lexical structure is in blockquotes with either “Lexer” or “Syntax” in bold superscript as the first line.
Syntax
ExampleGrammar:
~
Expression
|box
ExpressionSee Notation for more detail.
Contributing
We welcome contributions of all kinds.
You can contribute to this book by opening an issue or sending a pull request to the Rust Reference repository.
If this book does not answer your question, and you think its answer is in scope of it, please do not hesitate to file an issue or ask about it in the t-lang/doc
stream on Zulip.
Knowing what people use this book for the most helps direct our attention to making those sections the best that they can be.
We also want the reference to be as normative as possible, so if you see anything that is wrong or is non-normative but not specifically called out, please also file an issue.
Notation
Grammar
The following notations are used by the Lexer and Syntax grammar snippets:
Notation | Examples | Meaning |
---|---|---|
CAPITAL | KW_IF, INTEGER_LITERAL | A token produced by the lexer |
ItalicCamelCase | LetStatement, Item | A syntactical production |
string | x , while , * | The exact character(s) |
\x | \n, \r, \t, \0 | The character represented by this escape |
x? | pub ? | An optional item |
x* | OuterAttribute* | 0 or more of x |
x+ | MacroMatch+ | 1 or more of x |
xa..b | HEX_DIGIT1..6 | a to b repetitions of x |
| | u8 | u16 , Block | Item | Either one or another |
[ ] | [b B ] | Any of the characters listed |
[ - ] | [a -z ] | Any of the characters in the range |
~[ ] | ~[b B ] | Any characters, except those listed |
~string | ~\n , ~*/ | Any characters, except this sequence |
( ) | (, Parameter)? | Groups items |
String table productions
Some rules in the grammar — notably unary operators, binary operators, and keywords — are given in a simplified form: as a listing of printable strings. These cases form a subset of the rules regarding the token rule, and are assumed to be the result of a lexical-analysis phase feeding the parser, driven by a DFA, operating over the disjunction of all such string table entries.
When such a string in monospace
font occurs inside the grammar,
it is an implicit reference to a single member of such a string table
production. See tokens for more information.
Lexical structure
Input format
This chapter describes how a source file is interpreted as a sequence of tokens.
See Crates and source files for a description of how programs are organised into files.
Source encoding
Each source file is interpreted as a sequence of Unicode characters encoded in UTF-8. It is an error if the file is not valid UTF-8.
Byte order mark removal
If the first character in the sequence is U+FEFF
(BYTE ORDER MARK), it is removed.
CRLF normalization
Each pair of characters U+000D
(CR) immediately followed by U+000A
(LF) is replaced by a single U+000A
(LF).
Other occurrences of the character U+000D
(CR) are left in place (they are treated as whitespace).
Shebang removal
If the remaining sequence begins with the characters #!
, the characters up to and including the first U+000A
(LF) are removed from the sequence.
For example, the first line of the following file would be ignored:
#!/usr/bin/env rustx
fn main() {
println!("Hello!");
}
As an exception, if the #!
characters are followed (ignoring intervening comments or whitespace) by a [
token, nothing is removed.
This prevents an inner attribute at the start of a source file being removed.
Note: The standard library
include!
macro applies byte order mark removal, CRLF normalization, and shebang removal to the file it reads. Theinclude_str!
andinclude_bytes!
macros do not.
Tokenization
The resulting sequence of characters is then converted into tokens as described in the remainder of this chapter.
Keywords
Rust divides keywords into three categories:
Strict keywords
These keywords can only be used in their correct contexts. They cannot be used as the names of:
- Items
- Variables and function parameters
- Fields and variants
- Type parameters
- Lifetime parameters or loop labels
- Macros or attributes
- Macro placeholders
- Crates
Lexer:
KW_AS :as
KW_BREAK :break
KW_CONST :const
KW_CONTINUE :continue
KW_CRATE :crate
KW_ELSE :else
KW_ENUM :enum
KW_EXTERN :extern
KW_FALSE :false
KW_FN :fn
KW_FOR :for
KW_IF :if
KW_IMPL :impl
KW_IN :in
KW_LET :let
KW_LOOP :loop
KW_MATCH :match
KW_MOD :mod
KW_MOVE :move
KW_MUT :mut
KW_PUB :pub
KW_REF :ref
KW_RETURN :return
KW_SELFVALUE :self
KW_SELFTYPE :Self
KW_STATIC :static
KW_STRUCT :struct
KW_SUPER :super
KW_TRAIT :trait
KW_TRUE :true
KW_TYPE :type
KW_UNSAFE :unsafe
KW_USE :use
KW_WHERE :where
KW_WHILE :while
The following keywords were added beginning in the 2018 edition.
Lexer 2018+
KW_ASYNC :async
KW_AWAIT :await
KW_DYN :dyn
Reserved keywords
These keywords aren’t used yet, but they are reserved for future use. They have the same restrictions as strict keywords. The reasoning behind this is to make current programs forward compatible with future versions of Rust by forbidding them to use these keywords.
Lexer
KW_ABSTRACT :abstract
KW_BECOME :become
KW_BOX :box
KW_DO :do
KW_FINAL :final
KW_MACRO :macro
KW_OVERRIDE :override
KW_PRIV :priv
KW_TYPEOF :typeof
KW_UNSIZED :unsized
KW_VIRTUAL :virtual
KW_YIELD :yield
The following keywords are reserved beginning in the 2018 edition.
Lexer 2018+
KW_TRY :try
Weak keywords
These keywords have special meaning only in certain contexts. For example, it
is possible to declare a variable or method with the name union
.
-
macro_rules
is used to create custom macros. -
union
is used to declare a union and is only a keyword when used in a union declaration. -
'static
is used for the static lifetime and cannot be used as a generic lifetime parameter or loop label// error[E0262]: invalid lifetime parameter name: `'static` fn invalid_lifetime_parameter<'static>(s: &'static str) -> &'static str { s }
-
In the 2015 edition,
dyn
is a keyword when used in a type position followed by a path that does not start with::
or<
, a lifetime, a question mark, afor
keyword or an opening parenthesis.Beginning in the 2018 edition,
dyn
has been promoted to a strict keyword.
Lexer
KW_MACRO_RULES :macro_rules
KW_UNION :union
KW_STATICLIFETIME :'static
Lexer 2015
KW_DYN :dyn
safe
is used for functions and statics, which has meaning in external blocks.
Identifiers
Lexer:
IDENTIFIER_OR_KEYWORD :
XID_Start XID_Continue*
|_
XID_Continue+RAW_IDENTIFIER :
r#
IDENTIFIER_OR_KEYWORD Exceptcrate
,self
,super
,Self
NON_KEYWORD_IDENTIFIER : IDENTIFIER_OR_KEYWORD Except a strict or reserved keyword
IDENTIFIER :
NON_KEYWORD_IDENTIFIER | RAW_IDENTIFIER
Identifiers follow the specification in Unicode Standard Annex #31 for Unicode version 15.0, with the additions described below. Some examples of identifiers:
foo
_identifier
r#true
Москва
東京
The profile used from UAX #31 is:
- Start :=
XID_Start
, plus the underscore character (U+005F) - Continue :=
XID_Continue
- Medial := empty
with the additional constraint that a single underscore character is not an identifier.
Note: Identifiers starting with an underscore are typically used to indicate an identifier that is intentionally unused, and will silence the unused warning in
rustc
.
Identifiers may not be a strict or reserved keyword without the r#
prefix described below in raw identifiers.
Zero width non-joiner (ZWNJ U+200C) and zero width joiner (ZWJ U+200D) characters are not allowed in identifiers.
Identifiers are restricted to the ASCII subset of XID_Start
and XID_Continue
in the following situations:
extern crate
declarations- External crate names referenced in a path
- Module names loaded from the filesystem without a
path
attribute no_mangle
attributed items- Item names in external blocks
Normalization
Identifiers are normalized using Normalization Form C (NFC) as defined in Unicode Standard Annex #15. Two identifiers are equal if their NFC forms are equal.
Procedural and declarative macros receive normalized identifiers in their input.
Raw identifiers
A raw identifier is like a normal identifier, but prefixed by r#
. (Note that
the r#
prefix is not included as part of the actual identifier.)
Unlike a normal identifier, a raw identifier may be any strict or reserved
keyword except the ones listed above for RAW_IDENTIFIER
.
Comments
Lexer
LINE_COMMENT :
//
(~[/
!
\n
] |//
) ~\n
*
|//
BLOCK_COMMENT :
/*
(~[*
!
] |**
| BlockCommentOrDoc) (BlockCommentOrDoc | ~*/
)**/
|/**/
|/***/
INNER_LINE_DOC :
//!
~[\n
IsolatedCR]*INNER_BLOCK_DOC :
/*!
( BlockCommentOrDoc | ~[*/
IsolatedCR] )**/
OUTER_LINE_DOC :
///
(~/
~[\n
IsolatedCR]*)?OUTER_BLOCK_DOC :
/**
(~*
| BlockCommentOrDoc ) (BlockCommentOrDoc | ~[*/
IsolatedCR])**/
BlockCommentOrDoc :
BLOCK_COMMENT
| OUTER_BLOCK_DOC
| INNER_BLOCK_DOCIsolatedCR :
\r
Non-doc comments
Comments follow the general C++ style of line (//
) and
block (/* ... */
) comment forms. Nested block comments are supported.
Non-doc comments are interpreted as a form of whitespace.
Doc comments
Line doc comments beginning with exactly three slashes (///
), and block
doc comments (/** ... */
), both outer doc comments, are interpreted as a
special syntax for doc
attributes. That is, they are equivalent to writing
#[doc="..."]
around the body of the comment, i.e., /// Foo
turns into
#[doc="Foo"]
and /** Bar */
turns into #[doc="Bar"]
.
Line comments beginning with //!
and block comments /*! ... */
are
doc comments that apply to the parent of the comment, rather than the item
that follows. That is, they are equivalent to writing #![doc="..."]
around
the body of the comment. //!
comments are usually used to document
modules that occupy a source file.
The character U+000D
(CR) is not allowed in doc comments.
Note: The sequence
U+000D
(CR) immediately followed byU+000A
(LF) would have been previously transformed into a singleU+000A
(LF).
Examples
#![allow(unused)] fn main() { //! A doc comment that applies to the implicit anonymous module of this crate pub mod outer_module { //! - Inner line doc //!! - Still an inner line doc (but with a bang at the beginning) /*! - Inner block doc */ /*!! - Still an inner block doc (but with a bang at the beginning) */ // - Only a comment /// - Outer line doc (exactly 3 slashes) //// - Only a comment /* - Only a comment */ /** - Outer block doc (exactly) 2 asterisks */ /*** - Only a comment */ pub mod inner_module {} pub mod nested_comments { /* In Rust /* we can /* nest comments */ */ */ // All three types of block comments can contain or be nested inside // any other type: /* /* */ /** */ /*! */ */ /*! /* */ /** */ /*! */ */ /** /* */ /** */ /*! */ */ pub mod dummy_item {} } pub mod degenerate_cases { // empty inner line doc //! // empty inner block doc /*!*/ // empty line comment // // empty outer line doc /// // empty block comment /**/ pub mod dummy_item {} // empty 2-asterisk block isn't a doc block, it is a block comment /***/ } /* The next one isn't allowed because outer doc comments require an item that will receive the doc */ /// Where is my item? mod boo {} } }
Whitespace
Whitespace is any non-empty string containing only characters that have the
Pattern_White_Space
Unicode property, namely:
U+0009
(horizontal tab,'\t'
)U+000A
(line feed,'\n'
)U+000B
(vertical tab)U+000C
(form feed)U+000D
(carriage return,'\r'
)U+0020
(space,' '
)U+0085
(next line)U+200E
(left-to-right mark)U+200F
(right-to-left mark)U+2028
(line separator)U+2029
(paragraph separator)
Rust is a “free-form” language, meaning that all forms of whitespace serve only to separate tokens in the grammar, and have no semantic significance.
A Rust program has identical meaning if each whitespace element is replaced with any other legal whitespace element, such as a single space character.
Tokens
Tokens are primitive productions in the grammar defined by regular (non-recursive) languages. Rust source input can be broken down into the following kinds of tokens:
Within this documentation’s grammar, “simple” tokens are given in string
table production form, and appear in monospace
font.
Literals
Literals are tokens used in literal expressions.
Examples
Characters and strings
Example | # sets1 | Characters | Escapes | |
---|---|---|---|---|
Character | 'H' | 0 | All Unicode | Quote & ASCII & Unicode |
String | "hello" | 0 | All Unicode | Quote & ASCII & Unicode |
Raw string | r#"hello"# | <256 | All Unicode | N/A |
Byte | b'H' | 0 | All ASCII | Quote & Byte |
Byte string | b"hello" | 0 | All ASCII | Quote & Byte |
Raw byte string | br#"hello"# | <256 | All ASCII | N/A |
C string | c"hello" | 0 | All Unicode | Quote & Byte & Unicode |
Raw C string | cr#"hello"# | <256 | All Unicode | N/A |
The number of #
s on each side of the same literal must be equivalent.
Note: Character and string literal tokens never include the sequence of
U+000D
(CR) immediately followed byU+000A
(LF): this pair would have been previously transformed into a singleU+000A
(LF).
ASCII escapes
Name | |
---|---|
\x41 | 7-bit character code (exactly 2 digits, up to 0x7F) |
\n | Newline |
\r | Carriage return |
\t | Tab |
\\ | Backslash |
\0 | Null |
Byte escapes
Name | |
---|---|
\x7F | 8-bit character code (exactly 2 digits) |
\n | Newline |
\r | Carriage return |
\t | Tab |
\\ | Backslash |
\0 | Null |
Unicode escapes
Name | |
---|---|
\u{7FFF} | 24-bit Unicode character code (up to 6 digits) |
Quote escapes
Name | |
---|---|
\' | Single quote |
\" | Double quote |
Numbers
Number literals2 | Example | Exponentiation |
---|---|---|
Decimal integer | 98_222 | N/A |
Hex integer | 0xff | N/A |
Octal integer | 0o77 | N/A |
Binary integer | 0b1111_0000 | N/A |
Floating-point | 123.0E+77 | Optional |
All number literals allow _
as a visual separator: 1_234.0E+18f64
Suffixes
A suffix is a sequence of characters following the primary part of a literal (without intervening whitespace), of the same form as a non-raw identifier or keyword.
Lexer
SUFFIX : IDENTIFIER_OR_KEYWORD
SUFFIX_NO_E : SUFFIX not beginning withe
orE
Any kind of literal (string, integer, etc) with any suffix is valid as a token.
A literal token with any suffix can be passed to a macro without producing an error.
The macro itself will decide how to interpret such a token and whether to produce an error or not.
In particular, the literal
fragment specifier for by-example macros matches literal tokens with arbitrary suffixes.
#![allow(unused)] fn main() { macro_rules! blackhole { ($tt:tt) => () } macro_rules! blackhole_lit { ($l:literal) => () } blackhole!("string"suffix); // OK blackhole_lit!(1suffix); // OK }
However, suffixes on literal tokens which are interpreted as literal expressions or patterns are restricted. Any suffixes are rejected on non-numeric literal tokens, and numeric literal tokens are accepted only with suffixes from the list below.
Integer | Floating-point |
---|---|
u8 , i8 , u16 , i16 , u32 , i32 , u64 , i64 , u128 , i128 , usize , isize | f32 , f64 |
Character and string literals
Character literals
Lexer
CHAR_LITERAL :
'
( ~['
\
\n \r \t] | QUOTE_ESCAPE | ASCII_ESCAPE | UNICODE_ESCAPE )'
SUFFIX?QUOTE_ESCAPE :
\'
|\"
ASCII_ESCAPE :
\x
OCT_DIGIT HEX_DIGIT
|\n
|\r
|\t
|\\
|\0
UNICODE_ESCAPE :
\u{
( HEX_DIGIT_
* )1..6}
A character literal is a single Unicode character enclosed within two
U+0027
(single-quote) characters, with the exception of U+0027
itself,
which must be escaped by a preceding U+005C
character (\
).
String literals
Lexer
STRING_LITERAL :
"
(
~["
\
IsolatedCR]
| QUOTE_ESCAPE
| ASCII_ESCAPE
| UNICODE_ESCAPE
| STRING_CONTINUE
)*"
SUFFIX?STRING_CONTINUE :
\
followed by \n
A string literal is a sequence of any Unicode characters enclosed within two
U+0022
(double-quote) characters, with the exception of U+0022
itself,
which must be escaped by a preceding U+005C
character (\
).
Line-breaks, represented by the character U+000A
(LF), are allowed in string literals.
When an unescaped U+005C
character (\
) occurs immediately before a line break, the line break does not appear in the string represented by the token.
See String continuation escapes for details.
The character U+000D
(CR) may not appear in a string literal other than as part of such a string continuation escape.
Character escapes
Some additional escapes are available in either character or non-raw string
literals. An escape starts with a U+005C
(\
) and continues with one of the
following forms:
- A 7-bit code point escape starts with
U+0078
(x
) and is followed by exactly two hex digits with value up to0x7F
. It denotes the ASCII character with value equal to the provided hex value. Higher values are not permitted because it is ambiguous whether they mean Unicode code points or byte values. - A 24-bit code point escape starts with
U+0075
(u
) and is followed by up to six hex digits surrounded by bracesU+007B
({
) andU+007D
(}
). It denotes the Unicode code point equal to the provided hex value. - A whitespace escape is one of the characters
U+006E
(n
),U+0072
(r
), orU+0074
(t
), denoting the Unicode valuesU+000A
(LF),U+000D
(CR) orU+0009
(HT) respectively. - The null escape is the character
U+0030
(0
) and denotes the Unicode valueU+0000
(NUL). - The backslash escape is the character
U+005C
(\
) which must be escaped in order to denote itself.
Raw string literals
Lexer
RAW_STRING_LITERAL :
r
RAW_STRING_CONTENT SUFFIX?RAW_STRING_CONTENT :
"
( ~ IsolatedCR )* (non-greedy)"
|#
RAW_STRING_CONTENT#
Raw string literals do not process any escapes. They start with the character
U+0072
(r
), followed by fewer than 256 of the character U+0023
(#
) and a
U+0022
(double-quote) character.
The raw string body can contain any sequence of Unicode characters other than U+000D
(CR).
It is terminated only by another U+0022
(double-quote) character, followed by the same number of U+0023
(#
) characters that preceded the opening U+0022
(double-quote) character.
All Unicode characters contained in the raw string body represent themselves,
the characters U+0022
(double-quote) (except when followed by at least as
many U+0023
(#
) characters as were used to start the raw string literal) or
U+005C
(\
) do not have any special meaning.
Examples for string literals:
#![allow(unused)] fn main() { "foo"; r"foo"; // foo "\"foo\""; r#""foo""#; // "foo" "foo #\"# bar"; r##"foo #"# bar"##; // foo #"# bar "\x52"; "R"; r"R"; // R "\\x52"; r"\x52"; // \x52 }
Byte and byte string literals
Byte literals
Lexer
BYTE_LITERAL :
b'
( ASCII_FOR_CHAR | BYTE_ESCAPE )'
SUFFIX?ASCII_FOR_CHAR :
any ASCII (i.e. 0x00 to 0x7F), except'
,\
, \n, \r or \tBYTE_ESCAPE :
\x
HEX_DIGIT HEX_DIGIT
|\n
|\r
|\t
|\\
|\0
|\'
|\"
A byte literal is a single ASCII character (in the U+0000
to U+007F
range) or a single escape preceded by the characters U+0062
(b
) and
U+0027
(single-quote), and followed by the character U+0027
. If the character
U+0027
is present within the literal, it must be escaped by a preceding
U+005C
(\
) character. It is equivalent to a u8
unsigned 8-bit integer
number literal.
Byte string literals
Lexer
BYTE_STRING_LITERAL :
b"
( ASCII_FOR_STRING | BYTE_ESCAPE | STRING_CONTINUE )*"
SUFFIX?ASCII_FOR_STRING :
any ASCII (i.e 0x00 to 0x7F), except"
,\
and IsolatedCR
A non-raw byte string literal is a sequence of ASCII characters and escapes,
preceded by the characters U+0062
(b
) and U+0022
(double-quote), and
followed by the character U+0022
. If the character U+0022
is present within
the literal, it must be escaped by a preceding U+005C
(\
) character.
Alternatively, a byte string literal can be a raw byte string literal, defined
below.
Line-breaks, represented by the character U+000A
(LF), are allowed in byte string literals.
When an unescaped U+005C
character (\
) occurs immediately before a line break, the line break does not appear in the string represented by the token.
See String continuation escapes for details.
The character U+000D
(CR) may not appear in a byte string literal other than as part of such a string continuation escape.
Some additional escapes are available in either byte or non-raw byte string
literals. An escape starts with a U+005C
(\
) and continues with one of the
following forms:
- A byte escape escape starts with
U+0078
(x
) and is followed by exactly two hex digits. It denotes the byte equal to the provided hex value. - A whitespace escape is one of the characters
U+006E
(n
),U+0072
(r
), orU+0074
(t
), denoting the bytes values0x0A
(ASCII LF),0x0D
(ASCII CR) or0x09
(ASCII HT) respectively. - The null escape is the character
U+0030
(0
) and denotes the byte value0x00
(ASCII NUL). - The backslash escape is the character
U+005C
(\
) which must be escaped in order to denote its ASCII encoding0x5C
.
Raw byte string literals
Lexer
RAW_BYTE_STRING_LITERAL :
br
RAW_BYTE_STRING_CONTENT SUFFIX?RAW_BYTE_STRING_CONTENT :
"
ASCII_FOR_RAW* (non-greedy)"
|#
RAW_BYTE_STRING_CONTENT#
ASCII_FOR_RAW :
any ASCII (i.e. 0x00 to 0x7F) except IsolatedCR
Raw byte string literals do not process any escapes. They start with the
character U+0062
(b
), followed by U+0072
(r
), followed by fewer than 256
of the character U+0023
(#
), and a U+0022
(double-quote) character.
The raw string body can contain any sequence of ASCII characters other than U+000D
(CR).
It is terminated only by another U+0022
(double-quote) character, followed by the same number of U+0023
(#
) characters that preceded the opening U+0022
(double-quote) character.
A raw byte string literal can not contain any non-ASCII byte.
All characters contained in the raw string body represent their ASCII encoding,
the characters U+0022
(double-quote) (except when followed by at least as
many U+0023
(#
) characters as were used to start the raw string literal) or
U+005C
(\
) do not have any special meaning.
Examples for byte string literals:
#![allow(unused)] fn main() { b"foo"; br"foo"; // foo b"\"foo\""; br#""foo""#; // "foo" b"foo #\"# bar"; br##"foo #"# bar"##; // foo #"# bar b"\x52"; b"R"; br"R"; // R b"\\x52"; br"\x52"; // \x52 }
C string and raw C string literals
C string literals
Lexer
C_STRING_LITERAL :
c"
(
~["
\
IsolatedCR NUL]
| BYTE_ESCAPE except\0
or\x00
| UNICODE_ESCAPE except\u{0}
,\u{00}
, …,\u{000000}
| STRING_CONTINUE
)*"
SUFFIX?
A C string literal is a sequence of Unicode characters and escapes,
preceded by the characters U+0063
(c
) and U+0022
(double-quote), and
followed by the character U+0022
. If the character U+0022
is present within
the literal, it must be escaped by a preceding U+005C
(\
) character.
Alternatively, a C string literal can be a raw C string literal, defined below.
C strings are implicitly terminated by byte 0x00
, so the C string literal
c""
is equivalent to manually constructing a &CStr
from the byte string
literal b"\x00"
. Other than the implicit terminator, byte 0x00
is not
permitted within a C string.
Line-breaks, represented by the character U+000A
(LF), are allowed in C string literals.
When an unescaped U+005C
character (\
) occurs immediately before a line break, the line break does not appear in the string represented by the token.
See String continuation escapes for details.
The character U+000D
(CR) may not appear in a C string literal other than as part of such a string continuation escape.
Some additional escapes are available in non-raw C string literals. An escape
starts with a U+005C
(\
) and continues with one of the following forms:
- A byte escape escape starts with
U+0078
(x
) and is followed by exactly two hex digits. It denotes the byte equal to the provided hex value. - A 24-bit code point escape starts with
U+0075
(u
) and is followed by up to six hex digits surrounded by bracesU+007B
({
) andU+007D
(}
). It denotes the Unicode code point equal to the provided hex value, encoded as UTF-8. - A whitespace escape is one of the characters
U+006E
(n
),U+0072
(r
), orU+0074
(t
), denoting the bytes values0x0A
(ASCII LF),0x0D
(ASCII CR) or0x09
(ASCII HT) respectively. - The backslash escape is the character
U+005C
(\
) which must be escaped in order to denote its ASCII encoding0x5C
.
A C string represents bytes with no defined encoding, but a C string literal
may contain Unicode characters above U+007F
. Such characters will be replaced
with the bytes of that character’s UTF-8 representation.
The following C string literals are equivalent:
#![allow(unused)] fn main() { c"æ"; // LATIN SMALL LETTER AE (U+00E6) c"\u{00E6}"; c"\xC3\xA6"; }
Edition differences: C string literals are accepted in the 2021 edition or later. In earlier additions the token
c""
is lexed asc ""
.
Raw C string literals
Lexer
RAW_C_STRING_LITERAL :
cr
RAW_C_STRING_CONTENT SUFFIX?RAW_C_STRING_CONTENT :
"
( ~ IsolatedCR NUL )* (non-greedy)"
|#
RAW_C_STRING_CONTENT#
Raw C string literals do not process any escapes. They start with the
character U+0063
(c
), followed by U+0072
(r
), followed by fewer than 256
of the character U+0023
(#
), and a U+0022
(double-quote) character.
The raw C string body can contain any sequence of Unicode characters other than U+0000
(NUL) and U+000D
(CR).
It is terminated only by another U+0022
(double-quote) character, followed by the same number of U+0023
(#
) characters that preceded the opening U+0022
(double-quote) character.
All characters contained in the raw C string body represent themselves in UTF-8
encoding. The characters U+0022
(double-quote) (except when followed by at
least as many U+0023
(#
) characters as were used to start the raw C string
literal) or U+005C
(\
) do not have any special meaning.
Edition differences: Raw C string literals are accepted in the 2021 edition or later. In earlier additions the token
cr""
is lexed ascr ""
, andcr#""#
is lexed ascr #""#
(which is non-grammatical).
Examples for C string and raw C string literals
#![allow(unused)] fn main() { c"foo"; cr"foo"; // foo c"\"foo\""; cr#""foo""#; // "foo" c"foo #\"# bar"; cr##"foo #"# bar"##; // foo #"# bar c"\x52"; c"R"; cr"R"; // R c"\\x52"; cr"\x52"; // \x52 }
Number literals
A number literal is either an integer literal or a floating-point literal. The grammar for recognizing the two kinds of literals is mixed.
Integer literals
Lexer
INTEGER_LITERAL :
( DEC_LITERAL | BIN_LITERAL | OCT_LITERAL | HEX_LITERAL ) SUFFIX_NO_E?DEC_LITERAL :
DEC_DIGIT (DEC_DIGIT|_
)*BIN_LITERAL :
0b
(BIN_DIGIT|_
)* BIN_DIGIT (BIN_DIGIT|_
)*OCT_LITERAL :
0o
(OCT_DIGIT|_
)* OCT_DIGIT (OCT_DIGIT|_
)*HEX_LITERAL :
0x
(HEX_DIGIT|_
)* HEX_DIGIT (HEX_DIGIT|_
)*BIN_DIGIT : [
0
-1
]OCT_DIGIT : [
0
-7
]DEC_DIGIT : [
0
-9
]HEX_DIGIT : [
0
-9
a
-f
A
-F
]
An integer literal has one of four forms:
- A decimal literal starts with a decimal digit and continues with any mixture of decimal digits and underscores.
- A hex literal starts with the character sequence
U+0030
U+0078
(0x
) and continues as any mixture (with at least one digit) of hex digits and underscores. - An octal literal starts with the character sequence
U+0030
U+006F
(0o
) and continues as any mixture (with at least one digit) of octal digits and underscores. - A binary literal starts with the character sequence
U+0030
U+0062
(0b
) and continues as any mixture (with at least one digit) of binary digits and underscores.
Like any literal, an integer literal may be followed (immediately, without any spaces) by a suffix as described above.
The suffix may not begin with e
or E
, as that would be interpreted as the exponent of a floating-point literal.
See Integer literal expressions for the effect of these suffixes.
Examples of integer literals which are accepted as literal expressions:
#![allow(unused)] fn main() { #![allow(overflowing_literals)] 123; 123i32; 123u32; 123_u32; 0xff; 0xff_u8; 0x01_f32; // integer 7986, not floating-point 1.0 0x01_e3; // integer 483, not floating-point 1000.0 0o70; 0o70_i16; 0b1111_1111_1001_0000; 0b1111_1111_1001_0000i64; 0b________1; 0usize; // These are too big for their type, but are accepted as literal expressions. 128_i8; 256_u8; // This is an integer literal, accepted as a floating-point literal expression. 5f32; }
Note that -1i8
, for example, is analyzed as two tokens: -
followed by 1i8
.
Examples of integer literals which are not accepted as literal expressions:
#![allow(unused)] fn main() { #[cfg(FALSE)] { 0invalidSuffix; 123AFB43; 0b010a; 0xAB_CD_EF_GH; 0b1111_f32; } }
Tuple index
Lexer
TUPLE_INDEX:
INTEGER_LITERAL
A tuple index is used to refer to the fields of tuples, tuple structs, and tuple variants.
Tuple indices are compared with the literal token directly. Tuple indices
start with 0
and each successive index increments the value by 1
as a
decimal value. Thus, only decimal values will match, and the value must not
have any extra 0
prefix characters.
#![allow(unused)] fn main() { let example = ("dog", "cat", "horse"); let dog = example.0; let cat = example.1; // The following examples are invalid. let cat = example.01; // ERROR no field named `01` let horse = example.0b10; // ERROR no field named `0b10` }
Note: Tuple indices may include certain suffixes, but this is not intended to be valid, and may be removed in a future version. See https://github.com/rust-lang/rust/issues/60210 for more information.
Floating-point literals
Lexer
FLOAT_LITERAL :
DEC_LITERAL.
(not immediately followed by.
,_
or an XID_Start character)
| DEC_LITERAL.
DEC_LITERAL SUFFIX_NO_E?
| DEC_LITERAL (.
DEC_LITERAL)? FLOAT_EXPONENT SUFFIX?FLOAT_EXPONENT :
(e
|E
) (+
|-
)? (DEC_DIGIT|_
)* DEC_DIGIT (DEC_DIGIT|_
)*
A floating-point literal has one of two forms:
- A decimal literal followed by a period character
U+002E
(.
). This is optionally followed by another decimal literal, with an optional exponent. - A single decimal literal followed by an exponent.
Like integer literals, a floating-point literal may be followed by a
suffix, so long as the pre-suffix part does not end with U+002E
(.
).
The suffix may not begin with e
or E
if the literal does not include an exponent.
See Floating-point literal expressions for the effect of these suffixes.
Examples of floating-point literals which are accepted as literal expressions:
#![allow(unused)] fn main() { 123.0f64; 0.1f64; 0.1f32; 12E+99_f64; let x: f64 = 2.; }
This last example is different because it is not possible to use the suffix
syntax with a floating point literal ending in a period. 2.f64
would attempt
to call a method named f64
on 2
.
Note that -1.0
, for example, is analyzed as two tokens: -
followed by 1.0
.
Examples of floating-point literals which are not accepted as literal expressions:
#![allow(unused)] fn main() { #[cfg(FALSE)] { 2.0f80; 2e5f80; 2e5e6; 2.0e5e6; 1.3e10u64; } }
Reserved forms similar to number literals
Lexer
RESERVED_NUMBER :
BIN_LITERAL [2
-9
]
| OCT_LITERAL [8
-9
]
| ( BIN_LITERAL | OCT_LITERAL | HEX_LITERAL ).
(not immediately followed by.
,_
or an XID_Start character)
| ( BIN_LITERAL | OCT_LITERAL ) (e
|E
)
|0b
_
* end of input or not BIN_DIGIT
|0o
_
* end of input or not OCT_DIGIT
|0x
_
* end of input or not HEX_DIGIT
| DEC_LITERAL ( . DEC_LITERAL)? (e
|E
) (+
|-
)? end of input or not DEC_DIGIT
The following lexical forms similar to number literals are reserved forms. Due to the possible ambiguity these raise, they are rejected by the tokenizer instead of being interpreted as separate tokens.
-
An unsuffixed binary or octal literal followed, without intervening whitespace, by a decimal digit out of the range for its radix.
-
An unsuffixed binary, octal, or hexadecimal literal followed, without intervening whitespace, by a period character (with the same restrictions on what follows the period as for floating-point literals).
-
An unsuffixed binary or octal literal followed, without intervening whitespace, by the character
e
orE
. -
Input which begins with one of the radix prefixes but is not a valid binary, octal, or hexadecimal literal (because it contains no digits).
-
Input which has the form of a floating-point literal with no digits in the exponent.
Examples of reserved forms:
#![allow(unused)] fn main() { 0b0102; // this is not `0b010` followed by `2` 0o1279; // this is not `0o127` followed by `9` 0x80.0; // this is not `0x80` followed by `.` and `0` 0b101e; // this is not a suffixed literal, or `0b101` followed by `e` 0b; // this is not an integer literal, or `0` followed by `b` 0b_; // this is not an integer literal, or `0` followed by `b_` 2e; // this is not a floating-point literal, or `2` followed by `e` 2.0e; // this is not a floating-point literal, or `2.0` followed by `e` 2em; // this is not a suffixed literal, or `2` followed by `em` 2.0em; // this is not a suffixed literal, or `2.0` followed by `em` }
Lifetimes and loop labels
Lexer
LIFETIME_TOKEN :
'
IDENTIFIER_OR_KEYWORD (not immediately followed by'
)
|'_
(not immediately followed by'
)LIFETIME_OR_LABEL :
'
NON_KEYWORD_IDENTIFIER (not immediately followed by'
)
|'_
(not immediately followed by'
)
Lifetime parameters and loop labels use LIFETIME_OR_LABEL tokens. Any LIFETIME_TOKEN will be accepted by the lexer, and for example, can be used in macros.
Punctuation
Punctuation symbol tokens are listed here for completeness. Their individual usages and meanings are defined in the linked pages.
Delimiters
Bracket punctuation is used in various parts of the grammar. An open bracket must always be paired with a close bracket. Brackets and the tokens within them are referred to as “token trees” in macros. The three types of brackets are:
Bracket | Type |
---|---|
{ } | Curly braces |
[ ] | Square brackets |
( ) | Parentheses |
Reserved prefixes
Lexer 2021+
RESERVED_TOKEN_DOUBLE_QUOTE : ( IDENTIFIER_OR_KEYWORD Exceptb
orc
orr
orbr
orcr
|_
)"
RESERVED_TOKEN_SINGLE_QUOTE : ( IDENTIFIER_OR_KEYWORD Exceptb
|_
)'
RESERVED_TOKEN_POUND : ( IDENTIFIER_OR_KEYWORD Exceptr
orbr
orcr
|_
)#
Some lexical forms known as reserved prefixes are reserved for future use.
Source input which would otherwise be lexically interpreted as a non-raw identifier (or a keyword or _
) which is immediately followed by a #
, '
, or "
character (without intervening whitespace) is identified as a reserved prefix.
Note that raw identifiers, raw string literals, and raw byte string literals may contain a #
character but are not interpreted as containing a reserved prefix.
Similarly the r
, b
, br
, c
, and cr
prefixes used in raw string literals, byte literals, byte string literals, raw byte string literals, C string literals, and raw C string literals are not interpreted as reserved prefixes.
Edition differences: Starting with the 2021 edition, reserved prefixes are reported as an error by the lexer (in particular, they cannot be passed to macros).
Before the 2021 edition, reserved prefixes are accepted by the lexer and interpreted as multiple tokens (for example, one token for the identifier or keyword, followed by a
#
token).Examples accepted in all editions:
#![allow(unused)] fn main() { macro_rules! lexes {($($_:tt)*) => {}} lexes!{a #foo} lexes!{continue 'foo} lexes!{match "..." {}} lexes!{r#let#foo} // three tokens: r#let # foo }
Examples accepted before the 2021 edition but rejected later:
#![allow(unused)] fn main() { macro_rules! lexes {($($_:tt)*) => {}} lexes!{a#foo} lexes!{continue'foo} lexes!{match"..." {}} }
Macros
The functionality and syntax of Rust can be extended with custom definitions
called macros. They are given names, and invoked through a consistent
syntax: some_extension!(...)
.
There are two ways to define new macros:
- Macros by Example define new syntax in a higher-level, declarative way.
- Procedural Macros define function-like macros, custom derives, and custom attributes using functions that operate on input tokens.
Macro Invocation
Syntax
MacroInvocation :
SimplePath!
DelimTokenTreeDelimTokenTree :
(
TokenTree*)
|[
TokenTree*]
|{
TokenTree*}
TokenTree :
Tokenexcept delimiters | DelimTokenTreeMacroInvocationSemi :
SimplePath!
(
TokenTree*)
;
| SimplePath!
[
TokenTree*]
;
| SimplePath!
{
TokenTree*}
A macro invocation expands a macro at compile time and replaces the invocation with the result of the macro. Macros may be invoked in the following situations:
- Expressions and statements
- Patterns
- Types
- Items including associated items
macro_rules
transcribers- External blocks
When used as an item or a statement, the MacroInvocationSemi form is used
where a semicolon is required at the end when not using curly braces.
Visibility qualifiers are never allowed before a macro invocation or
macro_rules
definition.
#![allow(unused)] fn main() { // Used as an expression. let x = vec![1,2,3]; // Used as a statement. println!("Hello!"); // Used in a pattern. macro_rules! pat { ($i:ident) => (Some($i)) } if let pat!(x) = Some(1) { assert_eq!(x, 1); } // Used in a type. macro_rules! Tuple { { $A:ty, $B:ty } => { ($A, $B) }; } type N2 = Tuple!(i32, i32); // Used as an item. use std::cell::RefCell; thread_local!(static FOO: RefCell<u32> = RefCell::new(1)); // Used as an associated item. macro_rules! const_maker { ($t:ty, $v:tt) => { const CONST: $t = $v; }; } trait T { const_maker!{i32, 7} } // Macro calls within macros. macro_rules! example { () => { println!("Macro call in a macro!") }; } // Outer macro `example` is expanded, then inner macro `println` is expanded. example!(); }
Macros By Example
Syntax
MacroRulesDefinition :
macro_rules
!
IDENTIFIER MacroRulesDefMacroRulesDef :
(
MacroRules)
;
|[
MacroRules]
;
|{
MacroRules}
MacroRules :
MacroRule (;
MacroRule )*;
?MacroRule :
MacroMatcher=>
MacroTranscriberMacroMatcher :
(
MacroMatch*)
|[
MacroMatch*]
|{
MacroMatch*}
MacroMatch :
Tokenexcept$
and delimiters
| MacroMatcher
|$
( IDENTIFIER_OR_KEYWORD exceptcrate
| RAW_IDENTIFIER |_
):
MacroFragSpec
|$
(
MacroMatch+)
MacroRepSep? MacroRepOpMacroFragSpec :
block
|expr
|ident
|item
|lifetime
|literal
|meta
|pat
|pat_param
|path
|stmt
|tt
|ty
|vis
MacroRepSep :
Tokenexcept delimiters and MacroRepOpMacroRepOp :
*
|+
|?
MacroTranscriber :
DelimTokenTree
macro_rules
allows users to define syntax extension in a declarative way. We
call such extensions “macros by example” or simply “macros”.
Each macro by example has a name, and one or more rules. Each rule has two parts: a matcher, describing the syntax that it matches, and a transcriber, describing the syntax that will replace a successfully matched invocation. Both the matcher and the transcriber must be surrounded by delimiters. Macros can expand to expressions, statements, items (including traits, impls, and foreign items), types, or patterns.
Transcribing
When a macro is invoked, the macro expander looks up macro invocations by name,
and tries each macro rule in turn. It transcribes the first successful match; if
this results in an error, then future matches are not tried. When matching, no
lookahead is performed; if the compiler cannot unambiguously determine how to
parse the macro invocation one token at a time, then it is an error. In the
following example, the compiler does not look ahead past the identifier to see
if the following token is a )
, even though that would allow it to parse the
invocation unambiguously:
#![allow(unused)] fn main() { macro_rules! ambiguity { ($($i:ident)* $j:ident) => { }; } ambiguity!(error); // Error: local ambiguity }
In both the matcher and the transcriber, the $
token is used to invoke special
behaviours from the macro engine (described below in Metavariables and
Repetitions). Tokens that aren’t part of such an invocation are matched and
transcribed literally, with one exception. The exception is that the outer
delimiters for the matcher will match any pair of delimiters. Thus, for
instance, the matcher (())
will match {()}
but not {{}}
. The character
$
cannot be matched or transcribed literally.
Forwarding a matched fragment
When forwarding a matched fragment to another macro-by-example, matchers in
the second macro will see an opaque AST of the fragment type. The second macro
can’t use literal tokens to match the fragments in the matcher, only a
fragment specifier of the same type. The ident
, lifetime
, and tt
fragment types are an exception, and can be matched by literal tokens. The
following illustrates this restriction:
#![allow(unused)] fn main() { macro_rules! foo { ($l:expr) => { bar!($l); } // ERROR: ^^ no rules expected this token in macro call } macro_rules! bar { (3) => {} } foo!(3); }
The following illustrates how tokens can be directly matched after matching a
tt
fragment:
#![allow(unused)] fn main() { // compiles OK macro_rules! foo { ($l:tt) => { bar!($l); } } macro_rules! bar { (3) => {} } foo!(3); }
Metavariables
In the matcher, $
name :
fragment-specifier matches a Rust syntax
fragment of the kind specified and binds it to the metavariable $
name. Valid
fragment specifiers are:
item
: an Itemblock
: a BlockExpressionstmt
: a Statement without the trailing semicolon (except for item statements that require semicolons)pat_param
: a PatternNoTopAltpat
: at least any PatternNoTopAlt, and possibly more depending on editionexpr
: an Expressionty
: a Typeident
: an IDENTIFIER_OR_KEYWORD or RAW_IDENTIFIERpath
: a TypePath style pathtt
: a TokenTree (a single token or tokens in matching delimiters()
,[]
, or{}
)meta
: an Attr, the contents of an attributelifetime
: a LIFETIME_TOKENvis
: a possibly empty Visibility qualifierliteral
: matches-
?LiteralExpression
In the transcriber, metavariables are referred to simply by $
name, since
the fragment kind is specified in the matcher. Metavariables are replaced with
the syntax element that matched them. The keyword metavariable $crate
can be
used to refer to the current crate; see Hygiene below. Metavariables can be
transcribed more than once or not at all.
For reasons of backwards compatibility, though _
is also an
expression, a standalone underscore is not matched by
the expr
fragment specifier. However, _
is matched by the expr
fragment
specifier when it appears as a subexpression.
For the same reason, a standalone const block is not matched but it is matched when appearing as a subexpression.
Edition differences: Starting with the 2021 edition,
pat
fragment-specifiers match top-level or-patterns (that is, they accept Pattern).Before the 2021 edition, they match exactly the same fragments as
pat_param
(that is, they accept PatternNoTopAlt).The relevant edition is the one in effect for the
macro_rules!
definition.
Repetitions
In both the matcher and transcriber, repetitions are indicated by placing the
tokens to be repeated inside $(
…)
, followed by a repetition operator,
optionally with a separator token between. The separator token can be any token
other than a delimiter or one of the repetition operators, but ;
and ,
are
the most common. For instance, $( $i:ident ),*
represents any number of
identifiers separated by commas. Nested repetitions are permitted.
The repetition operators are:
*
— indicates any number of repetitions.+
— indicates any number but at least one.?
— indicates an optional fragment with zero or one occurrence.
Since ?
represents at most one occurrence, it cannot be used with a
separator.
The repeated fragment both matches and transcribes to the specified number of
the fragment, separated by the separator token. Metavariables are matched to
every repetition of their corresponding fragment. For instance, the $( $i:ident ),*
example above matches $i
to all of the identifiers in the list.
During transcription, additional restrictions apply to repetitions so that the compiler knows how to expand them properly:
- A metavariable must appear in exactly the same number, kind, and nesting
order of repetitions in the transcriber as it did in the matcher. So for the
matcher
$( $i:ident ),*
, the transcribers=> { $i }
,=> { $( $( $i)* )* }
, and=> { $( $i )+ }
are all illegal, but=> { $( $i );* }
is correct and replaces a comma-separated list of identifiers with a semicolon-separated list. - Each repetition in the transcriber must contain at least one metavariable to
decide how many times to expand it. If multiple metavariables appear in the
same repetition, they must be bound to the same number of fragments. For
instance,
( $( $i:ident ),* ; $( $j:ident ),* ) => (( $( ($i,$j) ),* ))
must bind the same number of$i
fragments as$j
fragments. This means that invoking the macro with(a, b, c; d, e, f)
is legal and expands to((a,d), (b,e), (c,f))
, but(a, b, c; d, e)
is illegal because it does not have the same number. This requirement applies to every layer of nested repetitions.
Scoping, Exporting, and Importing
For historical reasons, the scoping of macros by example does not work entirely like items. Macros have two forms of scope: textual scope, and path-based scope. Textual scope is based on the order that things appear in source files, or even across multiple files, and is the default scoping. It is explained further below. Path-based scope works exactly the same way that item scoping does. The scoping, exporting, and importing of macros is controlled largely by attributes.
When a macro is invoked by an unqualified identifier (not part of a multi-part path), it is first looked up in textual scoping. If this does not yield any results, then it is looked up in path-based scoping. If the macro’s name is qualified with a path, then it is only looked up in path-based scoping.
use lazy_static::lazy_static; // Path-based import.
macro_rules! lazy_static { // Textual definition.
(lazy) => {};
}
lazy_static!{lazy} // Textual lookup finds our macro first.
self::lazy_static!{} // Path-based lookup ignores our macro, finds imported one.
Textual Scope
Textual scope is based largely on the order that things appear in source files,
and works similarly to the scope of local variables declared with let
except
it also applies at the module level. When macro_rules!
is used to define a
macro, the macro enters the scope after the definition (note that it can still
be used recursively, since names are looked up from the invocation site), up
until its surrounding scope, typically a module, is closed. This can enter child
modules and even span across multiple files:
//// src/lib.rs
mod has_macro {
// m!{} // Error: m is not in scope.
macro_rules! m {
() => {};
}
m!{} // OK: appears after declaration of m.
mod uses_macro;
}
// m!{} // Error: m is not in scope.
//// src/has_macro/uses_macro.rs
m!{} // OK: appears after declaration of m in src/lib.rs
It is not an error to define a macro multiple times; the most recent declaration will shadow the previous one unless it has gone out of scope.
#![allow(unused)] fn main() { macro_rules! m { (1) => {}; } m!(1); mod inner { m!(1); macro_rules! m { (2) => {}; } // m!(1); // Error: no rule matches '1' m!(2); macro_rules! m { (3) => {}; } m!(3); } m!(1); }
Macros can be declared and used locally inside functions as well, and work similarly:
#![allow(unused)] fn main() { fn foo() { // m!(); // Error: m is not in scope. macro_rules! m { () => {}; } m!(); } // m!(); // Error: m is not in scope. }
The macro_use
attribute
The macro_use
attribute has two purposes. First, it can be used to make a
module’s macro scope not end when the module is closed, by applying it to a
module:
#![allow(unused)] fn main() { #[macro_use] mod inner { macro_rules! m { () => {}; } } m!(); }
Second, it can be used to import macros from another crate, by attaching it to
an extern crate
declaration appearing in the crate’s root module. Macros
imported this way are imported into the macro_use
prelude, not textually,
which means that they can be shadowed by any other name. While macros imported
by #[macro_use]
can be used before the import statement, in case of a
conflict, the last macro imported wins. Optionally, a list of macros to import
can be specified using the MetaListIdents syntax; this is not supported
when #[macro_use]
is applied to a module.
#[macro_use(lazy_static)] // Or #[macro_use] to import all macros.
extern crate lazy_static;
lazy_static!{}
// self::lazy_static!{} // Error: lazy_static is not defined in `self`
Macros to be imported with #[macro_use]
must be exported with
#[macro_export]
, which is described below.
Path-Based Scope
By default, a macro has no path-based scope. However, if it has the
#[macro_export]
attribute, then it is declared in the crate root scope and can
be referred to normally as such:
#![allow(unused)] fn main() { self::m!(); m!(); // OK: Path-based lookup finds m in the current module. mod inner { super::m!(); crate::m!(); } mod mac { #[macro_export] macro_rules! m { () => {}; } } }
Macros labeled with #[macro_export]
are always pub
and can be referred to
by other crates, either by path or by #[macro_use]
as described above.
Hygiene
By default, all identifiers referred to in a macro are expanded as-is, and are
looked up at the macro’s invocation site. This can lead to issues if a macro
refers to an item or macro which isn’t in scope at the invocation site. To
alleviate this, the $crate
metavariable can be used at the start of a path to
force lookup to occur inside the crate defining the macro.
//// Definitions in the `helper_macro` crate.
#[macro_export]
macro_rules! helped {
// () => { helper!() } // This might lead to an error due to 'helper' not being in scope.
() => { $crate::helper!() }
}
#[macro_export]
macro_rules! helper {
() => { () }
}
//// Usage in another crate.
// Note that `helper_macro::helper` is not imported!
use helper_macro::helped;
fn unit() {
helped!();
}
Note that, because $crate
refers to the current crate, it must be used with a
fully qualified module path when referring to non-macro items:
#![allow(unused)] fn main() { pub mod inner { #[macro_export] macro_rules! call_foo { () => { $crate::inner::foo() }; } pub fn foo() {} } }
Additionally, even though $crate
allows a macro to refer to items within its
own crate when expanding, its use has no effect on visibility. An item or macro
referred to must still be visible from the invocation site. In the following
example, any attempt to invoke call_foo!()
from outside its crate will fail
because foo()
is not public.
#![allow(unused)] fn main() { #[macro_export] macro_rules! call_foo { () => { $crate::foo() }; } fn foo() {} }
Version & Edition differences: Prior to Rust 1.30,
$crate
andlocal_inner_macros
(below) were unsupported. They were added alongside path-based imports of macros (described above), to ensure that helper macros did not need to be manually imported by users of a macro-exporting crate. Crates written for earlier versions of Rust that use helper macros need to be modified to use$crate
orlocal_inner_macros
to work well with path-based imports.
When a macro is exported, the #[macro_export]
attribute can have the
local_inner_macros
keyword added to automatically prefix all contained macro
invocations with $crate::
. This is intended primarily as a tool to migrate
code written before $crate
was added to the language to work with Rust 2018’s
path-based imports of macros. Its use is discouraged in new code.
#![allow(unused)] fn main() { #[macro_export(local_inner_macros)] macro_rules! helped { () => { helper!() } // Automatically converted to $crate::helper!(). } #[macro_export] macro_rules! helper { () => { () } } }
Follow-set Ambiguity Restrictions
The parser used by the macro system is reasonably powerful, but it is limited in order to prevent ambiguity in current or future versions of the language. In particular, in addition to the rule about ambiguous expansions, a nonterminal matched by a metavariable must be followed by a token which has been decided can be safely used after that kind of match.
As an example, a macro matcher like $i:expr [ , ]
could in theory be accepted
in Rust today, since [,]
cannot be part of a legal expression and therefore
the parse would always be unambiguous. However, because [
can start trailing
expressions, [
is not a character which can safely be ruled out as coming
after an expression. If [,]
were accepted in a later version of Rust, this
matcher would become ambiguous or would misparse, breaking working code.
Matchers like $i:expr,
or $i:expr;
would be legal, however, because ,
and
;
are legal expression separators. The specific rules are:
expr
andstmt
may only be followed by one of:=>
,,
, or;
.pat_param
may only be followed by one of:=>
,,
,=
,|
,if
, orin
.pat
may only be followed by one of:=>
,,
,=
,if
, orin
.path
andty
may only be followed by one of:=>
,,
,=
,|
,;
,:
,>
,>>
,[
,{
,as
,where
, or a macro variable ofblock
fragment specifier.vis
may only be followed by one of:,
, an identifier other than a non-rawpriv
, any token that can begin a type, or a metavariable with aident
,ty
, orpath
fragment specifier.- All other fragment specifiers have no restrictions.
Edition differences: Before the 2021 edition,
pat
may also be followed by|
.
When repetitions are involved, then the rules apply to every possible number of expansions, taking separators into account. This means:
- If the repetition includes a separator, that separator must be able to follow the contents of the repetition.
- If the repetition can repeat multiple times (
*
or+
), then the contents must be able to follow themselves. - The contents of the repetition must be able to follow whatever comes before, and whatever comes after must be able to follow the contents of the repetition.
- If the repetition can match zero times (
*
or?
), then whatever comes after must be able to follow whatever comes before.
For more detail, see the formal specification.
Procedural Macros
Procedural macros allow creating syntax extensions as execution of a function. Procedural macros come in one of three flavors:
- Function-like macros -
custom!(...)
- Derive macros -
#[derive(CustomDerive)]
- Attribute macros -
#[CustomAttribute]
Procedural macros allow you to run code at compile time that operates over Rust syntax, both consuming and producing Rust syntax. You can sort of think of procedural macros as functions from an AST to another AST.
Procedural macros must be defined in the root of a crate with the crate type of
proc-macro
.
The macros may not be used from the crate where they are defined, and can only be used when imported in another crate.
Note: When using Cargo, Procedural macro crates are defined with the
proc-macro
key in your manifest:[lib] proc-macro = true
As functions, they must either return syntax, panic, or loop endlessly. Returned syntax either replaces or adds the syntax depending on the kind of procedural macro. Panics are caught by the compiler and are turned into a compiler error. Endless loops are not caught by the compiler which hangs the compiler.
Procedural macros run during compilation, and thus have the same resources that the compiler has. For example, standard input, error, and output are the same that the compiler has access to. Similarly, file access is the same. Because of this, procedural macros have the same security concerns that Cargo’s build scripts have.
Procedural macros have two ways of reporting errors. The first is to panic. The
second is to emit a compile_error
macro invocation.
The proc_macro
crate
Procedural macro crates almost always will link to the compiler-provided
proc_macro
crate. The proc_macro
crate provides types required for
writing procedural macros and facilities to make it easier.
This crate primarily contains a TokenStream
type. Procedural macros operate
over token streams instead of AST nodes, which is a far more stable interface
over time for both the compiler and for procedural macros to target. A
token stream is roughly equivalent to Vec<TokenTree>
where a TokenTree
can roughly be thought of as lexical token. For example foo
is an Ident
token, .
is a Punct
token, and 1.2
is a Literal
token. The TokenStream
type, unlike Vec<TokenTree>
, is cheap to clone.
All tokens have an associated Span
. A Span
is an opaque value that cannot
be modified but can be manufactured. Span
s represent an extent of source
code within a program and are primarily used for error reporting. While you
cannot modify a Span
itself, you can always change the Span
associated
with any token, such as through getting a Span
from another token.
Procedural macro hygiene
Procedural macros are unhygienic. This means they behave as if the output token stream was simply written inline to the code it’s next to. This means that it’s affected by external items and also affects external imports.
Macro authors need to be careful to ensure their macros work in as many contexts
as possible given this limitation. This often includes using absolute paths to
items in libraries (for example, ::std::option::Option
instead of Option
) or
by ensuring that generated functions have names that are unlikely to clash with
other functions (like __internal_foo
instead of foo
).
Function-like procedural macros
Function-like procedural macros are procedural macros that are invoked using
the macro invocation operator (!
).
These macros are defined by a public function with the proc_macro
attribute and a signature of (TokenStream) -> TokenStream
. The input
TokenStream
is what is inside the delimiters of the macro invocation and the
output TokenStream
replaces the entire macro invocation.
The proc_macro
attribute defines the macro in the macro namespace in the root of the crate.
For example, the following macro definition ignores its input and outputs a
function answer
into its scope.
#![crate_type = "proc-macro"]
extern crate proc_macro;
use proc_macro::TokenStream;
#[proc_macro]
pub fn make_answer(_item: TokenStream) -> TokenStream {
"fn answer() -> u32 { 42 }".parse().unwrap()
}
And then we use it in a binary crate to print “42” to standard output.
extern crate proc_macro_examples;
use proc_macro_examples::make_answer;
make_answer!();
fn main() {
println!("{}", answer());
}
Function-like procedural macros may be invoked in any macro invocation
position, which includes statements, expressions, patterns, type
expressions, item positions, including items in extern
blocks, inherent
and trait implementations, and trait definitions.
Derive macros
Derive macros define new inputs for the derive
attribute. These macros
can create new items given the token stream of a struct, enum, or union.
They can also define derive macro helper attributes.
Custom derive macros are defined by a public function with the
proc_macro_derive
attribute and a signature of (TokenStream) -> TokenStream
.
The proc_macro_derive
attribute defines the custom derive in the macro namespace in the root of the crate.
The input TokenStream
is the token stream of the item that has the derive
attribute on it. The output TokenStream
must be a set of items that are
then appended to the module or block that the item from the input
TokenStream
is in.
The following is an example of a derive macro. Instead of doing anything
useful with its input, it just appends a function answer
.
#![crate_type = "proc-macro"]
extern crate proc_macro;
use proc_macro::TokenStream;
#[proc_macro_derive(AnswerFn)]
pub fn derive_answer_fn(_item: TokenStream) -> TokenStream {
"fn answer() -> u32 { 42 }".parse().unwrap()
}
And then using said derive macro:
extern crate proc_macro_examples;
use proc_macro_examples::AnswerFn;
#[derive(AnswerFn)]
struct Struct;
fn main() {
assert_eq!(42, answer());
}
Derive macro helper attributes
Derive macros can add additional attributes into the scope of the item they are on. Said attributes are called derive macro helper attributes. These attributes are inert, and their only purpose is to be fed into the derive macro that defined them. That said, they can be seen by all macros.
The way to define helper attributes is to put an attributes
key in the
proc_macro_derive
macro with a comma separated list of identifiers that are
the names of the helper attributes.
For example, the following derive macro defines a helper attribute
helper
, but ultimately doesn’t do anything with it.
#![crate_type="proc-macro"]
extern crate proc_macro;
use proc_macro::TokenStream;
#[proc_macro_derive(HelperAttr, attributes(helper))]
pub fn derive_helper_attr(_item: TokenStream) -> TokenStream {
TokenStream::new()
}
And then usage on the derive macro on a struct:
#[derive(HelperAttr)]
struct Struct {
#[helper] field: ()
}
Attribute macros
Attribute macros define new outer attributes which can be
attached to items, including items in extern
blocks, inherent and trait
implementations, and trait definitions.
Attribute macros are defined by a public function with the
proc_macro_attribute
attribute that has a signature of (TokenStream, TokenStream) -> TokenStream
. The first TokenStream
is the delimited token
tree following the attribute’s name, not including the outer delimiters. If
the attribute is written as a bare attribute name, the attribute
TokenStream
is empty. The second TokenStream
is the rest of the item
including other attributes on the item. The returned TokenStream
replaces the item with an arbitrary number of items.
The proc_macro_attribute
attribute defines the attribute in the macro namespace in the root of the crate.
For example, this attribute macro takes the input stream and returns it as is, effectively being the no-op of attributes.
#![crate_type = "proc-macro"]
extern crate proc_macro;
use proc_macro::TokenStream;
#[proc_macro_attribute]
pub fn return_as_is(_attr: TokenStream, item: TokenStream) -> TokenStream {
item
}
This following example shows the stringified TokenStream
s that the attribute
macros see. The output will show in the output of the compiler. The output is
shown in the comments after the function prefixed with “out:”.
// my-macro/src/lib.rs
extern crate proc_macro;
use proc_macro::TokenStream;
#[proc_macro_attribute]
pub fn show_streams(attr: TokenStream, item: TokenStream) -> TokenStream {
println!("attr: \"{attr}\"");
println!("item: \"{item}\"");
item
}
// src/lib.rs
extern crate my_macro;
use my_macro::show_streams;
// Example: Basic function
#[show_streams]
fn invoke1() {}
// out: attr: ""
// out: item: "fn invoke1() {}"
// Example: Attribute with input
#[show_streams(bar)]
fn invoke2() {}
// out: attr: "bar"
// out: item: "fn invoke2() {}"
// Example: Multiple tokens in the input
#[show_streams(multiple => tokens)]
fn invoke3() {}
// out: attr: "multiple => tokens"
// out: item: "fn invoke3() {}"
// Example:
#[show_streams { delimiters }]
fn invoke4() {}
// out: attr: "delimiters"
// out: item: "fn invoke4() {}"
Declarative macro tokens and procedural macro tokens
Declarative macro_rules
macros and procedural macros use similar, but
different definitions for tokens (or rather TokenTree
s.)
Token trees in macro_rules
(corresponding to tt
matchers) are defined as
- Delimited groups (
(...)
,{...}
, etc) - All operators supported by the language, both single-character and
multi-character ones (
+
,+=
).- Note that this set doesn’t include the single quote
'
.
- Note that this set doesn’t include the single quote
- Literals (
"string"
,1
, etc)- Note that negation (e.g.
-1
) is never a part of such literal tokens, but a separate operator token.
- Note that negation (e.g.
- Identifiers, including keywords (
ident
,r#ident
,fn
) - Lifetimes (
'ident
) - Metavariable substitutions in
macro_rules
(e.g.$my_expr
inmacro_rules! mac { ($my_expr: expr) => { $my_expr } }
after themac
’s expansion, which will be considered a single token tree regardless of the passed expression)
Token trees in procedural macros are defined as
- Delimited groups (
(...)
,{...}
, etc) - All punctuation characters used in operators supported by the language (
+
, but not+=
), and also the single quote'
character (typically used in lifetimes, see below for lifetime splitting and joining behavior) - Literals (
"string"
,1
, etc)- Negation (e.g.
-1
) is supported as a part of integer and floating point literals.
- Negation (e.g.
- Identifiers, including keywords (
ident
,r#ident
,fn
)
Mismatches between these two definitions are accounted for when token streams
are passed to and from procedural macros.
Note that the conversions below may happen lazily, so they might not happen if
the tokens are not actually inspected.
When passed to a proc-macro
- All multi-character operators are broken into single characters.
- Lifetimes are broken into a
'
character and an identifier. - All metavariable substitutions are represented as their underlying token
streams.
- Such token streams may be wrapped into delimited groups (
Group
) with implicit delimiters (Delimiter::None
) when it’s necessary for preserving parsing priorities. tt
andident
substitutions are never wrapped into such groups and always represented as their underlying token trees.
- Such token streams may be wrapped into delimited groups (
When emitted from a proc macro
- Punctuation characters are glued into multi-character operators when applicable.
- Single quotes
'
joined with identifiers are glued into lifetimes. - Negative literals are converted into two tokens (the
-
and the literal) possibly wrapped into a delimited group (Group
) with implicit delimiters (Delimiter::None
) when it’s necessary for preserving parsing priorities.
Note that neither declarative nor procedural macros support doc comment tokens
(e.g. /// Doc
), so they are always converted to token streams representing
their equivalent #[doc = r"str"]
attributes when passed to macros.
Crates and source files
Syntax
Crate :
InnerAttribute*
Item*
Note: Although Rust, like any other language, can be implemented by an interpreter as well as a compiler, the only existing implementation is a compiler, and the language has always been designed to be compiled. For these reasons, this section assumes a compiler.
Rust’s semantics obey a phase distinction between compile-time and run-time.1 Semantic rules that have a static interpretation govern the success or failure of compilation, while semantic rules that have a dynamic interpretation govern the behavior of the program at run-time.
The compilation model centers on artifacts called crates. Each compilation processes a single crate in source form, and if successful, produces a single crate in binary form: either an executable or some sort of library.2
A crate is a unit of compilation and linking, as well as versioning, distribution, and runtime loading. A crate contains a tree of nested module scopes. The top level of this tree is a module that is anonymous (from the point of view of paths within the module) and any item within a crate has a canonical module path denoting its location within the crate’s module tree.
The Rust compiler is always invoked with a single source file as input, and
always produces a single output crate. The processing of that source file may
result in other source files being loaded as modules. Source files have the
extension .rs
.
A Rust source file describes a module, the name and location of which — in the module tree of the current crate — are defined from outside the source file: either by an explicit Module item in a referencing source file, or by the name of the crate itself. Every source file is a module, but not every module needs its own source file: module definitions can be nested within one file.
Each source file contains a sequence of zero or more Item definitions, and may optionally begin with any number of attributes that apply to the containing module, most of which influence the behavior of the compiler. The anonymous crate module can have additional attributes that apply to the crate as a whole.
Note: The file’s contents may be preceded by a shebang.
#![allow(unused)] fn main() { // Specify the crate name. #![crate_name = "projx"] // Specify the type of output artifact. #![crate_type = "lib"] // Turn on a warning. // This can be done in any module, not just the anonymous crate module. #![warn(non_camel_case_types)] }
Main Functions
A crate that contains a main
function can be compiled to an executable. If a
main
function is present, it must take no arguments, must not declare any
trait or lifetime bounds, must not have any where clauses, and its return
type must implement the Termination
trait.
fn main() {}
fn main() -> ! { std::process::exit(0); }
fn main() -> impl std::process::Termination { std::process::ExitCode::SUCCESS }
The main
function may be an import, e.g. from an external crate or from the current one.
#![allow(unused)] fn main() { mod foo { pub fn bar() { println!("Hello, world!"); } } use foo::bar as main; }
Note: Types with implementations of
Termination
in the standard library include:
()
!
Infallible
ExitCode
Result<T, E> where T: Termination, E: Debug
The no_main
attribute
The no_main
attribute may be applied at the crate level to disable
emitting the main
symbol for an executable binary. This is useful when some
other object being linked to defines main
.
The crate_name
attribute
The crate_name
attribute may be applied at the crate level to specify the
name of the crate with the MetaNameValueStr syntax.
#![allow(unused)] #![crate_name = "mycrate"] fn main() { }
The crate name must not be empty, and must only contain Unicode alphanumeric
or _
(U+005F) characters.
This distinction would also exist in an interpreter. Static checks like syntactic analysis, type checking, and lints should happen before the program is executed regardless of when it is executed.
A crate is somewhat analogous to an assembly in the ECMA-335 CLI model, a library in the SML/NJ Compilation Manager, a unit in the Owens and Flatt module system, or a configuration in Mesa.
Conditional compilation
Syntax
ConfigurationPredicate :
ConfigurationOption
| ConfigurationAll
| ConfigurationAny
| ConfigurationNotConfigurationOption :
IDENTIFIER (=
(STRING_LITERAL | RAW_STRING_LITERAL))?ConfigurationAll
all
(
ConfigurationPredicateList?)
ConfigurationAny
any
(
ConfigurationPredicateList?)
ConfigurationNot
not
(
ConfigurationPredicate)
ConfigurationPredicateList
ConfigurationPredicate (,
ConfigurationPredicate)*,
?
Conditionally compiled source code is source code that is compiled only under certain conditions. Source code can be made conditionally compiled using the cfg
and cfg_attr
attributes and the built-in cfg
macro. Whether to compile can depend on the target architecture of the compiled crate, arbitrary values passed to the compiler, and other things further described below.
Each form of conditional compilation takes a configuration predicate that evaluates to true or false. The predicate is one of the following:
- A configuration option. The predicate is true if the option is set, and false if it is unset.
all()
with a comma-separated list of configuration predicates. It is true if all of the given predicates are true, or if the list is empty.any()
with a comma-separated list of configuration predicates. It is true if at least one of the given predicates is true. If there are no predicates, it is false.not()
with a configuration predicate. It is true if its predicate is false and false if its predicate is true.
Configuration options are either names or key-value pairs, and are either set or unset.
Names are written as a single identifier, such as unix
.
Key-value pairs are written as an identifier, =
, and then a string, such as target_arch = "x86_64"
.
Note: Whitespace around the
=
is ignored, sofoo="bar"
andfoo = "bar"
are equivalent.
Keys do not need to be unique. For example, both feature = "std"
and feature = "serde"
can be set at the same time.
Set Configuration Options
Which configuration options are set is determined statically during the compilation of the crate. Some options are compiler-set based on data about the compilation. Other options are arbitrarily-set based on input passed to the compiler outside of the code. It is not possible to set a configuration option from within the source code of the crate being compiled.
Note: For
rustc
, arbitrary-set configuration options are set using the--cfg
flag.
Note: Configuration options with the key
feature
are a convention used by Cargo for specifying compile-time options and optional dependencies.
Warning: Arbitrarily-set configuration options can clash with compiler-set configuration options. For example, it is possible to do rustc --cfg "unix" program.rs
while compiling to a Windows target, and have both unix
and windows
configuration options set at the same time. Doing this would be unwise.
target_arch
Key-value option set once with the target’s CPU architecture. The value is similar to the first element of the platform’s target triple, but not identical.
Example values:
"x86"
"x86_64"
"mips"
"powerpc"
"powerpc64"
"arm"
"aarch64"
target_feature
Key-value option set for each platform feature available for the current compilation target.
Example values:
"avx"
"avx2"
"crt-static"
"rdrand"
"sse"
"sse2"
"sse4.1"
See the target_feature
attribute for more details on the available
features. An additional feature of crt-static
is available to the
target_feature
option to indicate that a static C runtime is available.
target_os
Key-value option set once with the target’s operating system. This value is similar to the second and third element of the platform’s target triple.
Example values:
"windows"
"macos"
"ios"
"linux"
"android"
"freebsd"
"dragonfly"
"openbsd"
"netbsd"
"none"
(typical for embedded targets)
target_family
Key-value option providing a more generic description of a target, such as the family of the
operating systems or architectures that the target generally falls into. Any number of
target_family
key-value pairs can be set.
Example values:
"unix"
"windows"
"wasm"
- Both
"unix"
and"wasm"
unix
and windows
unix
is set if target_family = "unix"
is set and windows
is set if
target_family = "windows"
is set.
target_env
Key-value option set with further disambiguating information about the target
platform with information about the ABI or libc
used. For historical reasons,
this value is only defined as not the empty-string when actually needed for
disambiguation. Thus, for example, on many GNU platforms, this value will be
empty. This value is similar to the fourth element of the platform’s target
triple. One difference is that embedded ABIs such as gnueabihf
will simply
define target_env
as "gnu"
.
Example values:
""
"gnu"
"msvc"
"musl"
"sgx"
target_abi
Key-value option set to further disambiguate the target_env
with information
about the target ABI. For historical reasons,
this value is only defined as not the empty-string when actually needed for
disambiguation. Thus, for example, on many GNU platforms, this value will be
empty.
Example values:
""
"llvm"
"eabihf"
"abi64"
"sim"
"macabi"
target_endian
Key-value option set once with either a value of “little” or “big” depending on the endianness of the target’s CPU.
target_pointer_width
Key-value option set once with the target’s pointer width in bits.
Example values:
"16"
"32"
"64"
target_vendor
Key-value option set once with the vendor of the target.
Example values:
"apple"
"fortanix"
"pc"
"unknown"
target_has_atomic
Key-value option set for each bit width that the target supports atomic loads, stores, and compare-and-swap operations.
When this cfg is present, all of the stable core::sync::atomic
APIs are available for
the relevant atomic width.
Possible values:
"8"
"16"
"32"
"64"
"128"
"ptr"
test
Enabled when compiling the test harness. Done with rustc
by using the
--test
flag. See Testing for more on testing support.
debug_assertions
Enabled by default when compiling without optimizations.
This can be used to enable extra debugging code in development but not in
production. For example, it controls the behavior of the standard library’s
debug_assert!
macro.
proc_macro
Set when the crate being compiled is being compiled with the proc_macro
crate type.
panic
Key-value option set depending on the panic strategy. Note that more values may be added in the future.
Example values:
"abort"
"unwind"
Forms of conditional compilation
The cfg
attribute
Syntax
CfgAttrAttribute :
cfg
(
ConfigurationPredicate)
The cfg
attribute conditionally includes the thing it is attached to based
on a configuration predicate.
It is written as cfg
, (
, a configuration predicate, and finally )
.
If the predicate is true, the thing is rewritten to not have the cfg
attribute
on it. If the predicate is false, the thing is removed from the source code.
When a crate-level cfg
has a false predicate, the behavior is slightly
different: any crate attributes preceding the cfg
are kept, and any crate
attributes following the cfg
are removed. This allows #![no_std]
and
#![no_core]
crates to avoid linking std
/core
even if a #![cfg(...)]
has
removed the entire crate.
Some examples on functions:
#![allow(unused)] fn main() { // The function is only included in the build when compiling for macOS #[cfg(target_os = "macos")] fn macos_only() { // ... } // This function is only included when either foo or bar is defined #[cfg(any(foo, bar))] fn needs_foo_or_bar() { // ... } // This function is only included when compiling for a unixish OS with a 32-bit // architecture #[cfg(all(unix, target_pointer_width = "32"))] fn on_32bit_unix() { // ... } // This function is only included when foo is not defined #[cfg(not(foo))] fn needs_not_foo() { // ... } // This function is only included when the panic strategy is set to unwind #[cfg(panic = "unwind")] fn when_unwinding() { // ... } }
The cfg
attribute is allowed anywhere attributes are allowed.
The cfg_attr
attribute
Syntax
CfgAttrAttribute :
cfg_attr
(
ConfigurationPredicate,
CfgAttrs?)
The cfg_attr
attribute conditionally includes attributes based on a
configuration predicate.
When the configuration predicate is true, this attribute expands out to the
attributes listed after the predicate. For example, the following module will
either be found at linux.rs
or windows.rs
based on the target.
#[cfg_attr(target_os = "linux", path = "linux.rs")]
#[cfg_attr(windows, path = "windows.rs")]
mod os;
Zero, one, or more attributes may be listed. Multiple attributes will each be expanded into separate attributes. For example:
#[cfg_attr(feature = "magic", sparkles, crackles)]
fn bewitched() {}
// When the `magic` feature flag is enabled, the above will expand to:
#[sparkles]
#[crackles]
fn bewitched() {}
Note: The
cfg_attr
can expand to anothercfg_attr
. For example,#[cfg_attr(target_os = "linux", cfg_attr(feature = "multithreaded", some_other_attribute))]
is valid. This example would be equivalent to#[cfg_attr(all(target_os = "linux", feature ="multithreaded"), some_other_attribute)]
.
The cfg_attr
attribute is allowed anywhere attributes are allowed.
The cfg
macro
The built-in cfg
macro takes in a single configuration predicate and evaluates
to the true
literal when the predicate is true and the false
literal when
it is false.
For example:
#![allow(unused)] fn main() { let machine_kind = if cfg!(unix) { "unix" } else if cfg!(windows) { "windows" } else { "unknown" }; println!("I'm running on a {} machine!", machine_kind); }
Items
Syntax:
Item:
OuterAttribute*
VisItem
| MacroItemVisItem:
Visibility?
(
Module
| ExternCrate
| UseDeclaration
| Function
| TypeAlias
| Struct
| Enumeration
| Union
| ConstantItem
| StaticItem
| Trait
| Implementation
| ExternBlock
)MacroItem:
MacroInvocationSemi
| MacroRulesDefinition
An item is a component of a crate. Items are organized within a crate by a nested set of modules. Every crate has a single “outermost” anonymous module; all further items within the crate have paths within the module tree of the crate.
Items are entirely determined at compile-time, generally remain fixed during execution, and may reside in read-only memory.
There are several kinds of items:
- modules
extern crate
declarationsuse
declarations- function definitions
- type definitions
- struct definitions
- enumeration definitions
- union definitions
- constant items
- static items
- trait definitions
- implementations
extern
blocks
Items may be declared in the root of the crate, a module, or a block expression.
A subset of items, called associated items, may be declared in traits and implementations.
A subset of items, called external items, may be declared in extern
blocks.
Items may be defined in any order, with the exception of macro_rules
which has its own scoping behavior.
Name resolution of item names allows items to be defined before or after where the item is referred to in the module or block.
See item scopes for information on the scoping rules of items.
Modules
Syntax:
Module :
unsafe
?mod
IDENTIFIER;
|unsafe
?mod
IDENTIFIER{
InnerAttribute*
Item*
}
A module is a container for zero or more items.
A module item is a module, surrounded in braces, named, and prefixed with the
keyword mod
. A module item introduces a new, named module into the tree of
modules making up a crate. Modules can nest arbitrarily.
An example of a module:
#![allow(unused)] fn main() { mod math { type Complex = (f64, f64); fn sin(f: f64) -> f64 { /* ... */ unimplemented!(); } fn cos(f: f64) -> f64 { /* ... */ unimplemented!(); } fn tan(f: f64) -> f64 { /* ... */ unimplemented!(); } } }
Modules are defined in the type namespace of the module or block where they are located. It is an error to define multiple items with the same name in the same namespace within a module. See the scopes chapter for more details on restrictions and shadowing behavior.
The unsafe
keyword is syntactically allowed to appear before the mod
keyword, but it is rejected at a semantic level. This allows macros to consume
the syntax and make use of the unsafe
keyword, before removing it from the
token stream.
Module Source Filenames
A module without a body is loaded from an external file. When the module does
not have a path
attribute, the path to the file mirrors the logical module
path. Ancestor module path components are directories, and the module’s
contents are in a file with the name of the module plus the .rs
extension.
For example, the following module structure can have this corresponding
filesystem structure:
Module Path | Filesystem Path | File Contents |
---|---|---|
crate | lib.rs | mod util; |
crate::util | util.rs | mod config; |
crate::util::config | util/config.rs |
Module filenames may also be the name of the module as a directory with the
contents in a file named mod.rs
within that directory. The above example can
alternately be expressed with crate::util
’s contents in a file named
util/mod.rs
. It is not allowed to have both util.rs
and util/mod.rs
.
Note: Prior to
rustc
1.30, usingmod.rs
files was the way to load a module with nested children. It is encouraged to use the new naming convention as it is more consistent, and avoids having many files namedmod.rs
within a project.
The path
attribute
The directories and files used for loading external file modules can be
influenced with the path
attribute.
For path
attributes on modules not inside inline module blocks, the file
path is relative to the directory the source file is located. For example, the
following code snippet would use the paths shown based on where it is located:
#[path = "foo.rs"]
mod c;
Source File | c ’s File Location | c ’s Module Path |
---|---|---|
src/a/b.rs | src/a/foo.rs | crate::a::b::c |
src/a/mod.rs | src/a/foo.rs | crate::a::c |
For path
attributes inside inline module blocks, the relative location of
the file path depends on the kind of source file the path
attribute is
located in. “mod-rs” source files are root modules (such as lib.rs
or
main.rs
) and modules with files named mod.rs
. “non-mod-rs” source files
are all other module files. Paths for path
attributes inside inline module
blocks in a mod-rs file are relative to the directory of the mod-rs file
including the inline module components as directories. For non-mod-rs files,
it is the same except the path starts with a directory with the name of the
non-mod-rs module. For example, the following code snippet would use the paths
shown based on where it is located:
mod inline {
#[path = "other.rs"]
mod inner;
}
Source File | inner ’s File Location | inner ’s Module Path |
---|---|---|
src/a/b.rs | src/a/b/inline/other.rs | crate::a::b::inline::inner |
src/a/mod.rs | src/a/inline/other.rs | crate::a::inline::inner |
An example of combining the above rules of path
attributes on inline modules
and nested modules within (applies to both mod-rs and non-mod-rs files):
#[path = "thread_files"]
mod thread {
// Load the `local_data` module from `thread_files/tls.rs` relative to
// this source file's directory.
#[path = "tls.rs"]
mod local_data;
}
Attributes on Modules
Modules, like all items, accept outer attributes. They also accept inner
attributes: either after {
for a module with a body, or at the beginning of the
source file, after the optional BOM and shebang.
The built-in attributes that have meaning on a module are cfg
,
deprecated
, doc
, the lint check attributes, path
, and
no_implicit_prelude
. Modules also accept macro attributes.
Extern crate declarations
Syntax:
ExternCrate :
extern
crate
CrateRef AsClause?;
CrateRef :
IDENTIFIER |self
AsClause :
as
( IDENTIFIER |_
)
An extern crate
declaration specifies a dependency on an external crate.
The external crate is then bound into the declaring scope as the given identifier in the type namespace.
Additionally, if the extern crate
appears in the crate root, then the crate name is also added to the extern prelude, making it automatically in scope in all modules.
The as
clause can be used to bind the imported crate to a different name.
The external crate is resolved to a specific soname
at compile time, and a
runtime linkage requirement to that soname
is passed to the linker for
loading at runtime. The soname
is resolved at compile time by scanning the
compiler’s library path and matching the optional crate_name
provided against
the crate_name
attributes that were declared on the external crate when it was
compiled. If no crate_name
is provided, a default name
attribute is assumed,
equal to the identifier given in the extern crate
declaration.
The self
crate may be imported which creates a binding to the current crate.
In this case the as
clause must be used to specify the name to bind it to.
Three examples of extern crate
declarations:
extern crate pcre;
extern crate std; // equivalent to: extern crate std as std;
extern crate std as ruststd; // linking to 'std' under another name
When naming Rust crates, hyphens are disallowed. However, Cargo packages may
make use of them. In such case, when Cargo.toml
doesn’t specify a crate name,
Cargo will transparently replace -
with _
(Refer to RFC 940 for more
details).
Here is an example:
// Importing the Cargo package hello-world
extern crate hello_world; // hyphen replaced with an underscore
Underscore Imports
An external crate dependency can be declared without binding its name in scope
by using an underscore with the form extern crate foo as _
. This may be
useful for crates that only need to be linked, but are never referenced, and
will avoid being reported as unused.
The macro_use
attribute works as usual and imports the macro names
into the macro_use
prelude.
The no_link
attribute
The no_link
attribute may be specified on an extern crate
item to
prevent linking the crate into the output. This is commonly used to load a
crate to access only its macros.
Use declarations
Syntax:
UseDeclaration :
use
UseTree;
UseTree :
(SimplePath?::
)?*
| (SimplePath?::
)?{
(UseTree (,
UseTree )*,
?)?}
| SimplePath (as
( IDENTIFIER |_
) )?
A use declaration creates one or more local name bindings synonymous with
some other path. Usually a use
declaration is used to shorten the path
required to refer to a module item. These declarations may appear in modules
and blocks, usually at the top.
A use
declaration is also sometimes called an import, or, if it is public, a re-export.
Use declarations support a number of convenient shortcuts:
- Simultaneously binding a list of paths with a common prefix, using the
brace syntax
use a::b::{c, d, e::f, g::h::i};
- Simultaneously binding a list of paths with a common prefix and their common
parent module, using the
self
keyword, such asuse a::b::{self, c, d::e};
- Rebinding the target name as a new local name, using the syntax
use p::q::r as x;
. This can also be used with the last two features:use a::b::{self as ab, c as abc}
. - Binding all paths matching a given prefix, using the asterisk wildcard syntax
use a::b::*;
. - Nesting groups of the previous features multiple times, such as
use a::b::{self as ab, c, d::{*, e::f}};
An example of use
declarations:
use std::collections::hash_map::{self, HashMap}; fn foo<T>(_: T){} fn bar(map1: HashMap<String, usize>, map2: hash_map::HashMap<String, usize>){} fn main() { // use declarations can also exist inside of functions use std::option::Option::{Some, None}; // Equivalent to 'foo(vec![std::option::Option::Some(1.0f64), // std::option::Option::None]);' foo(vec![Some(1.0f64), None]); // Both `hash_map` and `HashMap` are in scope. let map1 = HashMap::new(); let map2 = hash_map::HashMap::new(); bar(map1, map2); }
use
Visibility
Like items, use
declarations are private to the containing module, by
default. Also like items, a use
declaration can be public, if qualified by
the pub
keyword. Such a use
declaration serves to re-export a name. A
public use
declaration can therefore redirect some public name to a
different target definition: even a definition with a private canonical path,
inside a different module. If a sequence of such redirections form a cycle or
cannot be resolved unambiguously, they represent a compile-time error.
An example of re-exporting:
mod quux { pub use self::foo::{bar, baz}; pub mod foo { pub fn bar() {} pub fn baz() {} } } fn main() { quux::bar(); quux::baz(); }
In this example, the module quux
re-exports two public names defined in
foo
.
use
Paths
The paths that are allowed in a use
item follow the SimplePath grammar and are similar to the paths that may be used in an expression.
They may create bindings for:
They cannot import associated items, generic parameters, local variables, paths with Self
, or tool attributes. More restrictions are described below.
use
will create bindings for all namespaces from the imported entities, with the exception that a self
import will only import from the type namespace (as described below).
For example, the following illustrates creating bindings for the same name in two namespaces:
#![allow(unused)] fn main() { mod stuff { pub struct Foo(pub i32); } // Imports the `Foo` type and the `Foo` constructor. use stuff::Foo; fn example() { let ctor = Foo; // Uses `Foo` from the value namespace. let x: Foo = ctor(123); // Uses `Foo` From the type namespace. } }
Edition differences: In the 2015 edition,
use
paths are relative to the crate root. For example:mod foo { pub mod example { pub mod iter {} } pub mod baz { pub fn foobaz() {} } } mod bar { // Resolves `foo` from the crate root. use foo::example::iter; // The `::` prefix explicitly resolves `foo` // from the crate root. use ::foo::baz::foobaz; } fn main() {}
The 2015 edition does not allow use declarations to reference the extern prelude. Thus,
extern crate
declarations are still required in 2015 to reference an external crate in ause
declaration. Beginning with the 2018 edition,use
declarations can specify an external crate dependency the same wayextern crate
can.
as
renames
The as
keyword can be used to change the name of an imported entity.
For example:
#![allow(unused)] fn main() { // Creates a non-public alias `bar` for the function `foo`. use inner::foo as bar; mod inner { pub fn foo() {} } }
Brace syntax
Braces can be used in the last segment of the path to import multiple entities from the previous segment, or, if there are no previous segments, from the current scope. Braces can be nested, creating a tree of paths, where each grouping of segments is logically combined with its parent to create a full path.
#![allow(unused)] fn main() { // Creates bindings to: // - `std::collections::BTreeSet` // - `std::collections::hash_map` // - `std::collections::hash_map::HashMap` use std::collections::{BTreeSet, hash_map::{self, HashMap}}; }
An empty brace does not import anything, though the leading path is validated that it is accessible.
Edition differences: In the 2015 edition, paths are relative to the crate root, so an import such as
use {foo, bar};
will import the namesfoo
andbar
from the crate root, whereas starting in 2018, those names are relative to the current scope.
self
imports
The keyword self
may be used within brace syntax to create a binding of the parent entity under its own name.
mod stuff { pub fn foo() {} pub fn bar() {} } mod example { // Creates a binding for `stuff` and `foo`. use crate::stuff::{self, foo}; pub fn baz() { foo(); stuff::bar(); } } fn main() {}
self
only creates a binding from the type namespace of the parent entity.
For example, in the following, only the foo
mod is imported:
mod bar { pub mod foo {} pub fn foo() {} } // This only imports the module `foo`. The function `foo` lives in // the value namespace and is not imported. use bar::foo::{self}; fn main() { foo(); //~ ERROR `foo` is a module }
Note:
self
may also be used as the first segment of a path. The usage ofself
as the first segment and inside ause
brace is logically the same; it means the current module of the parent segment, or the current module if there is no parent segment. Seeself
in the paths chapter for more information on the meaning of a leadingself
.
Glob imports
The *
character may be used as the last segment of a use
path to import all importable entities from the entity of the preceding segment.
For example:
#![allow(unused)] fn main() { // Creates a non-public alias to `bar`. use foo::*; mod foo { fn i_am_private() {} enum Example { V1, V2, } pub fn bar() { // Creates local aliases to `V1` and `V2` // of the `Example` enum. use Example::*; let x = V1; } } }
Items and named imports are allowed to shadow names from glob imports in the same namespace. That is, if there is a name already defined by another item in the same namespace, the glob import will be shadowed. For example:
#![allow(unused)] fn main() { // This creates a binding to the `clashing::Foo` tuple struct // constructor, but does not import its type because that would // conflict with the `Foo` struct defined here. // // Note that the order of definition here is unimportant. use clashing::*; struct Foo { field: f32, } fn do_stuff() { // Uses the constructor from `clashing::Foo`. let f1 = Foo(123); // The struct expression uses the type from // the `Foo` struct defined above. let f2 = Foo { field: 1.0 }; // `Bar` is also in scope due to the glob import. let z = Bar {}; } mod clashing { pub struct Foo(pub i32); pub struct Bar {} } }
*
cannot be used as the first or intermediate segments.
*
cannot be used to import a module’s contents into itself (such as use self::*;
).
Edition differences: In the 2015 edition, paths are relative to the crate root, so an import such as
use *;
is valid, and it means to import everything from the crate root. This cannot be used in the crate root itself.
Underscore Imports
Items can be imported without binding to a name by using an underscore with
the form use path as _
. This is particularly useful to import a trait so
that its methods may be used without importing the trait’s symbol, for example
if the trait’s symbol may conflict with another symbol. Another example is to
link an external crate without importing its name.
Asterisk glob imports will import items imported with _
in their unnameable
form.
mod foo { pub trait Zoo { fn zoo(&self) {} } impl<T> Zoo for T {} } use self::foo::Zoo as _; struct Zoo; // Underscore import avoids name conflict with this item. fn main() { let z = Zoo; z.zoo(); }
The unique, unnameable symbols are created after macro expansion so that
macros may safely emit multiple references to _
imports. For example, the
following should not produce an error:
#![allow(unused)] fn main() { macro_rules! m { ($item: item) => { $item $item } } m!(use std as _;); // This expands to: // use std as _; // use std as _; }
Restrictions
The following are restrictions for valid use
declarations:
use crate;
must useas
to define the name to which to bind the crate root.use {self};
is an error; there must be a leading segment when usingself
.- As with any item definition,
use
imports cannot create duplicate bindings of the same name in the same namespace in a module or block. use
paths with$crate
are not allowed in amacro_rules
expansion.use
paths cannot refer to enum variants through a type alias. For example:#![allow(unused)] fn main() { enum MyEnum { MyVariant } type TypeAlias = MyEnum; use MyEnum::MyVariant; //~ OK use TypeAlias::MyVariant; //~ ERROR }
Ambiguities
Note: This section is incomplete.
Some situations are an error when there is an ambiguity as to which name a use
declaration refers. This happens when there are two name candidates that do not resolve to the same entity.
Glob imports are allowed to import conflicting names in the same namespace as long as the name is not used. For example:
mod foo { pub struct Qux; } mod bar { pub struct Qux; } use foo::*; use bar::*; //~ OK, no name conflict. fn main() { // This would be an error, due to the ambiguity. //let x = Qux; }
Multiple glob imports are allowed to import the same name, and that name is allowed to be used, if the imports are of the same item (following re-exports). The visibility of the name is the maximum visibility of the imports. For example:
mod foo { pub struct Qux; } mod bar { pub use super::foo::Qux; } // These both import the same `Qux`. The visibility of `Qux` // is `pub` because that is the maximum visibility between // these two `use` declarations. pub use bar::*; use foo::*; fn main() { let _: Qux = Qux; }
Functions
Syntax
Function :
FunctionQualifiersfn
IDENTIFIER GenericParams?
(
FunctionParameters?)
FunctionReturnType? WhereClause?
( BlockExpression |;
)FunctionQualifiers :
const
?async
1? ItemSafety? (extern
Abi?)?ItemSafety :
safe
2 |unsafe
Abi :
STRING_LITERAL | RAW_STRING_LITERALFunctionParameters :
SelfParam,
?
| (SelfParam,
)? FunctionParam (,
FunctionParam)*,
?SelfParam :
OuterAttribute* ( ShorthandSelf | TypedSelf )ShorthandSelf :
(&
|&
Lifetime)?mut
?self
TypedSelf :
mut
?self
:
TypeFunctionParam :
OuterAttribute* ( FunctionParamPattern |...
| Type 3 )FunctionParamPattern :
PatternNoTopAlt:
( Type |...
)FunctionReturnType :
->
Type1The
async
qualifier is not allowed in the 2015 edition.2The
safe
function qualifier is only allowed semantically withinextern
blocks.3Function parameters with only a type are only allowed in an associated function of a trait item in the 2015 edition.
A function consists of a block (that’s the body of the function),
along with a name, a set of parameters, and an output type.
Other than a name, all these are optional.
Functions are declared with the keyword fn
which defines the given name in the value namespace of the module or block where it is located.
Functions may declare a set of input variables as parameters, through which the caller passes arguments into the function, and the output type of the value the function will return to its caller on completion.
If the output type is not explicitly stated, it is the unit type.
When referred to, a function yields a first-class value of the corresponding zero-sized function item type, which when called evaluates to a direct call to the function.
For example, this is a simple function:
#![allow(unused)] fn main() { fn answer_to_life_the_universe_and_everything() -> i32 { return 42; } }
The safe
function is semantically only allowed when used in an extern
block.
Function parameters
Function parameters are irrefutable patterns, so any pattern that is valid in
an else-less let
binding is also valid as a parameter:
#![allow(unused)] fn main() { fn first((value, _): (i32, i32)) -> i32 { value } }
If the first parameter is a SelfParam, this indicates that the function is a method. Functions with a self parameter may only appear as an associated function in a trait or implementation.
A parameter with the ...
token indicates a variadic function, and may only
be used as the last parameter of an external block function. The variadic
parameter may have an optional identifier, such as args: ...
.
Function body
The body block of a function is conceptually wrapped in another block that first binds the
argument patterns and then return
s the value of the function’s body. This
means that the tail expression of the block, if evaluated, ends up being
returned to the caller. As usual, an explicit return expression within
the body of the function will short-cut that implicit return, if reached.
For example, the function above behaves as if it was written as:
// argument_0 is the actual first argument passed from the caller
let (value, _) = argument_0;
return {
value
};
Functions without a body block are terminated with a semicolon. This form may only appear in a trait or external block.
Generic functions
A generic function allows one or more parameterized types to appear in its signature. Each type parameter must be explicitly declared in an angle-bracket-enclosed and comma-separated list, following the function name.
#![allow(unused)] fn main() { // foo is generic over A and B fn foo<A, B>(x: A, y: B) { } }
Inside the function signature and body, the name of the type parameter can be
used as a type name. Trait bounds can be specified for type
parameters to allow methods with that trait to be called on values of that
type. This is specified using the where
syntax:
#![allow(unused)] fn main() { use std::fmt::Debug; fn foo<T>(x: T) where T: Debug { } }
When a generic function is referenced, its type is instantiated based on the
context of the reference. For example, calling the foo
function here:
#![allow(unused)] fn main() { use std::fmt::Debug; fn foo<T>(x: &[T]) where T: Debug { // details elided } foo(&[1, 2]); }
will instantiate type parameter T
with i32
.
The type parameters can also be explicitly supplied in a trailing path
component after the function name. This might be necessary if there is not
sufficient context to determine the type parameters. For example,
mem::size_of::<u32>() == 4
.
Extern function qualifier
The extern
function qualifier allows providing function definitions that can
be called with a particular ABI:
extern "ABI" fn foo() { /* ... */ }
These are often used in combination with external block items which provide function declarations that can be used to call functions without providing their definition:
unsafe extern "ABI" {
unsafe fn foo(); /* no body */
safe fn bar(); /* no body */
}
unsafe { foo() };
bar();
When "extern" Abi?*
is omitted from FunctionQualifiers
in function items,
the ABI "Rust"
is assigned. For example:
#![allow(unused)] fn main() { fn foo() {} }
is equivalent to:
#![allow(unused)] fn main() { extern "Rust" fn foo() {} }
Functions can be called by foreign code, and using an ABI that differs from Rust allows, for example, to provide functions that can be called from other programming languages like C:
#![allow(unused)] fn main() { // Declares a function with the "C" ABI extern "C" fn new_i32() -> i32 { 0 } // Declares a function with the "stdcall" ABI #[cfg(target_arch = "x86_64")] extern "stdcall" fn new_i32_stdcall() -> i32 { 0 } }
Just as with external block, when the extern
keyword is used and the "ABI"
is omitted, the ABI used defaults to "C"
. That is, this:
#![allow(unused)] fn main() { extern fn new_i32() -> i32 { 0 } let fptr: extern fn() -> i32 = new_i32; }
is equivalent to:
#![allow(unused)] fn main() { extern "C" fn new_i32() -> i32 { 0 } let fptr: extern "C" fn() -> i32 = new_i32; }
Functions with an ABI that differs from "Rust"
do not support unwinding in the
exact same way that Rust does. Therefore, unwinding past the end of functions
with such ABIs causes the process to abort.
Note: The LLVM backend of the
rustc
implementation aborts the process by executing an illegal instruction.
Const functions
Functions qualified with the const
keyword are const functions, as are
tuple struct and tuple variant constructors. Const functions can be
called from within const contexts.
Const functions may use the extern
function qualifier, but only with the "Rust"
and "C"
ABIs.
Const functions are not allowed to be async.
Async functions
Functions may be qualified as async, and this can also be combined with the
unsafe
qualifier:
#![allow(unused)] fn main() { async fn regular_example() { } async unsafe fn unsafe_example() { } }
Async functions do no work when called: instead, they capture their arguments into a future. When polled, that future will execute the function’s body.
An async function is roughly equivalent to a function
that returns impl Future
and with an async move
block as
its body:
#![allow(unused)] fn main() { // Source async fn example(x: &str) -> usize { x.len() } }
is roughly equivalent to:
#![allow(unused)] fn main() { use std::future::Future; // Desugared fn example<'a>(x: &'a str) -> impl Future<Output = usize> + 'a { async move { x.len() } } }
The actual desugaring is more complex:
- The return type in the desugaring is assumed to capture all lifetime
parameters from the
async fn
declaration. This can be seen in the desugared example above, which explicitly outlives, and hence captures,'a
. - The
async move
block in the body captures all function parameters, including those that are unused or bound to a_
pattern. This ensures that function parameters are dropped in the same order as they would be if the function were not async, except that the drop occurs when the returned future has been fully awaited.
For more information on the effect of async, see async
blocks.
Edition differences: Async functions are only available beginning with Rust 2018.
Combining async
and unsafe
It is legal to declare a function that is both async and unsafe. The
resulting function is unsafe to call and (like any async function)
returns a future. This future is just an ordinary future and thus an
unsafe
context is not required to “await” it:
#![allow(unused)] fn main() { // Returns a future that, when awaited, dereferences `x`. // // Soundness condition: `x` must be safe to dereference until // the resulting future is complete. async unsafe fn unsafe_example(x: *const i32) -> i32 { *x } async fn safe_example() { // An `unsafe` block is required to invoke the function initially: let p = 22; let future = unsafe { unsafe_example(&p) }; // But no `unsafe` block required here. This will // read the value of `p`: let q = future.await; } }
Note that this behavior is a consequence of the desugaring to a
function that returns an impl Future
– in this case, the function
we desugar to is an unsafe
function, but the return value remains
the same.
Unsafe is used on an async function in precisely the same way that it
is used on other functions: it indicates that the function imposes
some additional obligations on its caller to ensure soundness. As in any
other unsafe function, these conditions may extend beyond the initial
call itself – in the snippet above, for example, the unsafe_example
function took a pointer x
as argument, and then (when awaited)
dereferenced that pointer. This implies that x
would have to be
valid until the future is finished executing, and it is the caller’s
responsibility to ensure that.
Attributes on functions
Outer attributes are allowed on functions. Inner
attributes are allowed directly after the {
inside its body block.
This example shows an inner attribute on a function. The function is documented with just the word “Example”.
#![allow(unused)] fn main() { fn documented() { #![doc = "Example"] } }
Note: Except for lints, it is idiomatic to only use outer attributes on function items.
The attributes that have meaning on a function are cfg
, cfg_attr
, deprecated
,
doc
, export_name
, link_section
, no_mangle
, the lint check
attributes, must_use
, the procedural macro attributes, the testing
attributes, and the optimization hint attributes. Functions also accept
attributes macros.
Attributes on function parameters
Outer attributes are allowed on function parameters and the
permitted built-in attributes are restricted to cfg
, cfg_attr
, allow
,
warn
, deny
, and forbid
.
#![allow(unused)] fn main() { fn len( #[cfg(windows)] slice: &[u16], #[cfg(not(windows))] slice: &[u8], ) -> usize { slice.len() } }
Inert helper attributes used by procedural macro attributes applied to items are also
allowed but be careful to not include these inert attributes in your final TokenStream
.
For example, the following code defines an inert some_inert_attribute
attribute that
is not formally defined anywhere and the some_proc_macro_attribute
procedural macro is
responsible for detecting its presence and removing it from the output token stream.
#[some_proc_macro_attribute]
fn foo_oof(#[some_inert_attribute] arg: u8) {
}
Type aliases
Syntax
TypeAlias :
type
IDENTIFIER GenericParams? (:
TypeParamBounds )? WhereClause? (=
Type WhereClause?)?;
A type alias defines a new name for an existing type in the type namespace of the module or block where it is located.
Type aliases are declared with the keyword type
.
Every value has a single, specific type, but may implement several different traits, and may be compatible with several different type constraints.
For example, the following defines the type Point
as a synonym for the type
(u8, u8)
, the type of pairs of unsigned 8 bit integers:
#![allow(unused)] fn main() { type Point = (u8, u8); let p: Point = (41, 68); }
A type alias to a tuple-struct or unit-struct cannot be used to qualify that type’s constructor:
#![allow(unused)] fn main() { struct MyStruct(u32); use MyStruct as UseAlias; type TypeAlias = MyStruct; let _ = UseAlias(5); // OK let _ = TypeAlias(5); // Doesn't work }
A type alias, when not used as an associated type, must include a Type and may not include TypeParamBounds.
A type alias, when used as an associated type in a trait, must not include a Type specification but may include TypeParamBounds.
A type alias, when used as an associated type in a trait impl, must include a Type specification and may not include TypeParamBounds.
Where clauses before the equals sign on a type alias in a trait impl (like
type TypeAlias<T> where T: Foo = Bar<T>
) are deprecated. Where clauses after
the equals sign (like type TypeAlias<T> = Bar<T> where T: Foo
) are preferred.
Structs
Syntax
Struct :
StructStruct
| TupleStructStructStruct :
struct
IDENTIFIER GenericParams? WhereClause? ({
StructFields?}
|;
)TupleStruct :
struct
IDENTIFIER GenericParams?(
TupleFields?)
WhereClause?;
StructFields :
StructField (,
StructField)*,
?StructField :
OuterAttribute*
Visibility?
IDENTIFIER:
TypeTupleFields :
TupleField (,
TupleField)*,
?TupleField :
OuterAttribute*
Visibility?
Type
A struct is a nominal struct type defined with the keyword struct
.
A struct declaration defines the given name in the type namespace of the module or block where it is located.
An example of a struct
item and its use:
#![allow(unused)] fn main() { struct Point {x: i32, y: i32} let p = Point {x: 10, y: 11}; let px: i32 = p.x; }
A tuple struct is a nominal tuple type, and is also defined with the keyword struct
.
In addition to defining a type, it also defines a constructor of the same name in the value namespace.
The constructor is a function which can be called to create a new instance of the struct.
For example:
#![allow(unused)] fn main() { struct Point(i32, i32); let p = Point(10, 11); let px: i32 = match p { Point(x, _) => x }; }
A unit-like struct is a struct without any fields, defined by leaving off the list of fields entirely. Such a struct implicitly defines a constant of its type with the same name. For example:
#![allow(unused)] fn main() { struct Cookie; let c = [Cookie, Cookie {}, Cookie, Cookie {}]; }
is equivalent to
#![allow(unused)] fn main() { struct Cookie {} const Cookie: Cookie = Cookie {}; let c = [Cookie, Cookie {}, Cookie, Cookie {}]; }
The precise memory layout of a struct is not specified. One can specify a
particular layout using the repr
attribute.
Enumerations
Syntax
Enumeration :
enum
IDENTIFIER GenericParams? WhereClause?{
EnumItems?}
EnumItems :
EnumItem (,
EnumItem )*,
?EnumItem :
OuterAttribute* Visibility?
IDENTIFIER ( EnumItemTuple | EnumItemStruct )? EnumItemDiscriminant?EnumItemTuple :
(
TupleFields?)
EnumItemStruct :
{
StructFields?}
EnumItemDiscriminant :
=
Expression
An enumeration, also referred to as an enum, is a simultaneous definition of a nominal enumerated type as well as a set of constructors, that can be used to create or pattern-match values of the corresponding enumerated type.
Enumerations are declared with the keyword enum
.
The enum
declaration defines the enumeration type in the type namespace of the module or block where it is located.
An example of an enum
item and its use:
#![allow(unused)] fn main() { enum Animal { Dog, Cat, } let mut a: Animal = Animal::Dog; a = Animal::Cat; }
Enum constructors can have either named or unnamed fields:
#![allow(unused)] fn main() { enum Animal { Dog(String, f64), Cat { name: String, weight: f64 }, } let mut a: Animal = Animal::Dog("Cocoa".to_string(), 37.2); a = Animal::Cat { name: "Spotty".to_string(), weight: 2.7 }; }
In this example, Cat
is a struct-like enum variant, whereas Dog
is simply
called an enum variant.
An enum where no constructors contain fields are called a field-less enum. For example, this is a fieldless enum:
#![allow(unused)] fn main() { enum Fieldless { Tuple(), Struct{}, Unit, } }
If a field-less enum only contains unit variants, the enum is called an unit-only enum. For example:
#![allow(unused)] fn main() { enum Enum { Foo = 3, Bar = 2, Baz = 1, } }
Variant constructors are similar to struct definitions, and can be referenced by a path from the enumeration name, including in use declarations. Each variant defines its type in the type namespace, though that type cannot be used as a type specifier. Tuple-like and unit-like variants also define a constructor in the value namespace.
A struct-like variant can be instantiated with a struct expression. A tuple-like variant can be instantiated with a call expression or a struct expression. A unit-like variant can be instantiated with a path expression or a struct expression. For example:
#![allow(unused)] fn main() { enum Examples { UnitLike, TupleLike(i32), StructLike { value: i32 }, } use Examples::*; // Creates aliases to all variants. let x = UnitLike; // Path expression of the const item. let x = UnitLike {}; // Struct expression. let y = TupleLike(123); // Call expression. let y = TupleLike { 0: 123 }; // Struct expression using integer field names. let z = StructLike { value: 123 }; // Struct expression. }
Discriminants
Each enum instance has a discriminant: an integer logically associated to it that is used to determine which variant it holds.
Under the default representation, the discriminant is interpreted as
an isize
value. However, the compiler is allowed to use a smaller type (or
another means of distinguishing variants) in its actual memory layout.
Assigning discriminant values
Explicit discriminants
In two circumstances, the discriminant of a variant may be explicitly set by
following the variant name with =
and a constant expression:
-
if the enumeration is “unit-only”.
-
if a primitive representation is used. For example:
#![allow(unused)] fn main() { #[repr(u8)] enum Enum { Unit = 3, Tuple(u16), Struct { a: u8, b: u16, } = 1, } }
Implicit discriminants
If a discriminant for a variant is not specified, then it is set to one higher than the discriminant of the previous variant in the declaration. If the discriminant of the first variant in the declaration is unspecified, then it is set to zero.
#![allow(unused)] fn main() { enum Foo { Bar, // 0 Baz = 123, // 123 Quux, // 124 } let baz_discriminant = Foo::Baz as u32; assert_eq!(baz_discriminant, 123); }
Restrictions
It is an error when two variants share the same discriminant.
#![allow(unused)] fn main() { enum SharedDiscriminantError { SharedA = 1, SharedB = 1 } enum SharedDiscriminantError2 { Zero, // 0 One, // 1 OneToo = 1 // 1 (collision with previous!) } }
It is also an error to have an unspecified discriminant where the previous discriminant is the maximum value for the size of the discriminant.
#![allow(unused)] fn main() { #[repr(u8)] enum OverflowingDiscriminantError { Max = 255, MaxPlusOne // Would be 256, but that overflows the enum. } #[repr(u8)] enum OverflowingDiscriminantError2 { MaxMinusOne = 254, // 254 Max, // 255 MaxPlusOne // Would be 256, but that overflows the enum. } }
Accessing discriminant
Via mem::discriminant
std::mem::discriminant
returns an opaque reference to the discriminant of
an enum value which can be compared. This cannot be used to get the value
of the discriminant.
Casting
If an enumeration is unit-only (with no tuple and struct variants), then its discriminant can be directly accessed with a numeric cast; e.g.:
#![allow(unused)] fn main() { enum Enum { Foo, Bar, Baz, } assert_eq!(0, Enum::Foo as isize); assert_eq!(1, Enum::Bar as isize); assert_eq!(2, Enum::Baz as isize); }
Field-less enums can be casted if they do not have explicit discriminants, or where only unit variants are explicit.
#![allow(unused)] fn main() { enum Fieldless { Tuple(), Struct{}, Unit, } assert_eq!(0, Fieldless::Tuple() as isize); assert_eq!(1, Fieldless::Struct{} as isize); assert_eq!(2, Fieldless::Unit as isize); #[repr(u8)] enum FieldlessWithDiscrimants { First = 10, Tuple(), Second = 20, Struct{}, Unit, } assert_eq!(10, FieldlessWithDiscrimants::First as u8); assert_eq!(11, FieldlessWithDiscrimants::Tuple() as u8); assert_eq!(20, FieldlessWithDiscrimants::Second as u8); assert_eq!(21, FieldlessWithDiscrimants::Struct{} as u8); assert_eq!(22, FieldlessWithDiscrimants::Unit as u8); }
Pointer casting
If the enumeration specifies a primitive representation, then the discriminant may be reliably accessed via unsafe pointer casting:
#![allow(unused)] fn main() { #[repr(u8)] enum Enum { Unit, Tuple(bool), Struct{a: bool}, } impl Enum { fn discriminant(&self) -> u8 { unsafe { *(self as *const Self as *const u8) } } } let unit_like = Enum::Unit; let tuple_like = Enum::Tuple(true); let struct_like = Enum::Struct{a: false}; assert_eq!(0, unit_like.discriminant()); assert_eq!(1, tuple_like.discriminant()); assert_eq!(2, struct_like.discriminant()); }
Zero-variant enums
Enums with zero variants are known as zero-variant enums. As they have no valid values, they cannot be instantiated.
#![allow(unused)] fn main() { enum ZeroVariants {} }
Zero-variant enums are equivalent to the never type, but they cannot be coerced into other types.
#![allow(unused)] fn main() { enum ZeroVariants {} let x: ZeroVariants = panic!(); let y: u32 = x; // mismatched type error }
Variant visibility
Enum variants syntactically allow a Visibility annotation, but this is rejected when the enum is validated. This allows items to be parsed with a unified syntax across different contexts where they are used.
#![allow(unused)] fn main() { macro_rules! mac_variant { ($vis:vis $name:ident) => { enum $name { $vis Unit, $vis Tuple(u8, u16), $vis Struct { f: u8 }, } } } // Empty `vis` is allowed. mac_variant! { E } // This is allowed, since it is removed before being validated. #[cfg(FALSE)] enum E { pub U, pub(crate) T(u8), pub(super) T { f: String } } }
Unions
Syntax
Union :
union
IDENTIFIER GenericParams? WhereClause?{
StructFields?}
A union declaration uses the same syntax as a struct declaration, except with
union
in place of struct
.
A union declaration defines the given name in the type namespace of the module or block where it is located.
#![allow(unused)] fn main() { #[repr(C)] union MyUnion { f1: u32, f2: f32, } }
The key property of unions is that all fields of a union share common storage. As a result, writes to one field of a union can overwrite its other fields, and size of a union is determined by the size of its largest field.
Union field types are restricted to the following subset of types:
Copy
types- References (
&T
and&mut T
for arbitraryT
) ManuallyDrop<T>
(for arbitraryT
)- Tuples and arrays containing only allowed union field types
This restriction ensures, in particular, that union fields never need to be
dropped. Like for structs and enums, it is possible to impl Drop
for a union
to manually define what happens when it gets dropped.
Unions without any fields are not accepted by the compiler, but can be accepted by macros.
Initialization of a union
A value of a union type can be created using the same syntax that is used for struct types, except that it must specify exactly one field:
#![allow(unused)] fn main() { union MyUnion { f1: u32, f2: f32 } let u = MyUnion { f1: 1 }; }
The expression above creates a value of type MyUnion
and initializes the
storage using field f1
. The union can be accessed using the same syntax as
struct fields:
#![allow(unused)] fn main() { union MyUnion { f1: u32, f2: f32 } let u = MyUnion { f1: 1 }; let f = unsafe { u.f1 }; }
Reading and writing union fields
Unions have no notion of an “active field”. Instead, every union access just
interprets the storage as the type of the field used for the access. Reading a
union field reads the bits of the union at the field’s type. Fields might have a
non-zero offset (except when the C representation is used); in that case the
bits starting at the offset of the fields are read. It is the programmer’s
responsibility to make sure that the data is valid at the field’s type. Failing
to do so results in undefined behavior. For example, reading the value 3
from a field of the boolean type is undefined behavior. Effectively,
writing to and then reading from a union with the C representation is
analogous to a transmute
from the type used for writing to the type used for
reading.
Consequently, all reads of union fields have to be placed in unsafe
blocks:
#![allow(unused)] fn main() { union MyUnion { f1: u32, f2: f32 } let u = MyUnion { f1: 1 }; unsafe { let f = u.f1; } }
Commonly, code using unions will provide safe wrappers around unsafe union field accesses.
In contrast, writes to union fields are safe, since they just overwrite arbitrary data, but cannot cause undefined behavior. (Note that union field types can never have drop glue, so a union field write will never implicitly drop anything.)
Pattern matching on unions
Another way to access union fields is to use pattern matching. Pattern matching
on union fields uses the same syntax as struct patterns, except that the pattern
must specify exactly one field. Since pattern matching is like reading the union
with a particular field, it has to be placed in unsafe
blocks as well.
#![allow(unused)] fn main() { union MyUnion { f1: u32, f2: f32 } fn f(u: MyUnion) { unsafe { match u { MyUnion { f1: 10 } => { println!("ten"); } MyUnion { f2 } => { println!("{}", f2); } } } } }
Pattern matching may match a union as a field of a larger structure. In particular, when using a Rust union to implement a C tagged union via FFI, this allows matching on the tag and the corresponding field simultaneously:
#![allow(unused)] fn main() { #[repr(u32)] enum Tag { I, F } #[repr(C)] union U { i: i32, f: f32, } #[repr(C)] struct Value { tag: Tag, u: U, } fn is_zero(v: Value) -> bool { unsafe { match v { Value { tag: Tag::I, u: U { i: 0 } } => true, Value { tag: Tag::F, u: U { f: num } } if num == 0.0 => true, _ => false, } } } }
References to union fields
Since union fields share common storage, gaining write access to one field of a union can give write access to all its remaining fields. Borrow checking rules have to be adjusted to account for this fact. As a result, if one field of a union is borrowed, all its remaining fields are borrowed as well for the same lifetime.
#![allow(unused)] fn main() { union MyUnion { f1: u32, f2: f32 } // ERROR: cannot borrow `u` (via `u.f2`) as mutable more than once at a time fn test() { let mut u = MyUnion { f1: 1 }; unsafe { let b1 = &mut u.f1; // ---- first mutable borrow occurs here (via `u.f1`) let b2 = &mut u.f2; // ^^^^ second mutable borrow occurs here (via `u.f2`) *b1 = 5; } // - first borrow ends here assert_eq!(unsafe { u.f1 }, 5); } }
As you could see, in many aspects (except for layouts, safety, and ownership) unions behave exactly like structs, largely as a consequence of inheriting their syntactic shape from structs. This is also true for many unmentioned aspects of Rust language (such as privacy, name resolution, type inference, generics, trait implementations, inherent implementations, coherence, pattern checking, etc etc etc).
Constant items
Syntax
ConstantItem :
const
( IDENTIFIER |_
):
Type (=
Expression )?;
A constant item is an optionally named constant value which is not associated
with a specific memory location in the program. Constants are essentially inlined
wherever they are used, meaning that they are copied directly into the relevant
context when used. This includes usage of constants from external crates, and
non-Copy
types. References to the same constant are not necessarily
guaranteed to refer to the same memory address.
The constant declaration defines the constant value in the value namespace of the module or block where it is located.
Constants must be explicitly typed. The type must have a 'static
lifetime: any
references in the initializer must have 'static
lifetimes.
Constants may refer to the address of other constants, in which case the
address will have elided lifetimes where applicable, otherwise – in most cases
– defaulting to the static
lifetime. (See static lifetime
elision.) The compiler is, however, still at liberty to translate the constant
many times, so the address referred to may not be stable.
#![allow(unused)] fn main() { const BIT1: u32 = 1 << 0; const BIT2: u32 = 1 << 1; const BITS: [u32; 2] = [BIT1, BIT2]; const STRING: &'static str = "bitstring"; struct BitsNStrings<'a> { mybits: [u32; 2], mystring: &'a str, } const BITS_N_STRINGS: BitsNStrings<'static> = BitsNStrings { mybits: BITS, mystring: STRING, }; }
The constant expression may only be omitted in a trait definition.
Constants with Destructors
Constants can contain destructors. Destructors are run when the value goes out of scope.
#![allow(unused)] fn main() { struct TypeWithDestructor(i32); impl Drop for TypeWithDestructor { fn drop(&mut self) { println!("Dropped. Held {}.", self.0); } } const ZERO_WITH_DESTRUCTOR: TypeWithDestructor = TypeWithDestructor(0); fn create_and_drop_zero_with_destructor() { let x = ZERO_WITH_DESTRUCTOR; // x gets dropped at end of function, calling drop. // prints "Dropped. Held 0.". } }
Unnamed constant
Unlike an associated constant, a free constant may be unnamed by using an underscore instead of the name. For example:
#![allow(unused)] fn main() { const _: () = { struct _SameNameTwice; }; // OK although it is the same name as above: const _: () = { struct _SameNameTwice; }; }
As with underscore imports, macros may safely emit the same unnamed constant in the same scope more than once. For example, the following should not produce an error:
#![allow(unused)] fn main() { macro_rules! m { ($item: item) => { $item $item } } m!(const _: () = ();); // This expands to: // const _: () = (); // const _: () = (); }
Evaluation
Free constants are always evaluated at compile-time to surface panics. This happens even within an unused function:
#![allow(unused)] fn main() { // Compile-time panic const PANIC: () = std::unimplemented!(); fn unused_generic_function<T>() { // A failing compile-time assertion const _: () = assert!(usize::BITS == 0); } }
Static items
Syntax
StaticItem :
ItemSafety?1static
mut
? IDENTIFIER:
Type (=
Expression )?;
1The
safe
andunsafe
function qualifiers are only allowed semantically withinextern
blocks.
A static item is similar to a constant, except that it represents a precise
memory location in the program. All references to the static refer to the same
memory location. Static items have the static
lifetime, which outlives all
other lifetimes in a Rust program. Static items do not call drop
at the
end of the program.
The static declaration defines a static value in the value namespace of the module or block where it is located.
The static initializer is a constant expression evaluated at compile time. Static initializers may refer to other statics.
Non-mut
static items that contain a type that is not interior mutable may
be placed in read-only memory.
All access to a static is safe, but there are a number of restrictions on statics:
- The type must have the
Sync
trait bound to allow thread-safe access. - Constants cannot refer to statics.
The initializer expression must be omitted in an external block, and must be provided for free static items.
The safe
and unsafe
qualifiers are semantically only allowed when used in an external block.
Statics & generics
A static item defined in a generic scope (for example in a blanket or default implementation) will result in exactly one static item being defined, as if the static definition was pulled out of the current scope into the module. There will not be one item per monomorphization.
This code:
use std::sync::atomic::{AtomicUsize, Ordering}; trait Tr { fn default_impl() { static COUNTER: AtomicUsize = AtomicUsize::new(0); println!("default_impl: counter was {}", COUNTER.fetch_add(1, Ordering::Relaxed)); } fn blanket_impl(); } struct Ty1 {} struct Ty2 {} impl<T> Tr for T { fn blanket_impl() { static COUNTER: AtomicUsize = AtomicUsize::new(0); println!("blanket_impl: counter was {}", COUNTER.fetch_add(1, Ordering::Relaxed)); } } fn main() { <Ty1 as Tr>::default_impl(); <Ty2 as Tr>::default_impl(); <Ty1 as Tr>::blanket_impl(); <Ty2 as Tr>::blanket_impl(); }
prints
default_impl: counter was 0
default_impl: counter was 1
blanket_impl: counter was 0
blanket_impl: counter was 1
Mutable statics
If a static item is declared with the mut
keyword, then it is allowed to be
modified by the program. One of Rust’s goals is to make concurrency bugs hard
to run into, and this is obviously a very large source of race conditions or
other bugs. For this reason, an unsafe
block is required when either reading
or writing a mutable static variable. Care should be taken to ensure that
modifications to a mutable static are safe with respect to other threads
running in the same process.
Mutable statics are still very useful, however. They can be used with C
libraries and can also be bound from C libraries in an extern
block.
#![allow(unused)] fn main() { fn atomic_add(_: *mut u32, _: u32) -> u32 { 2 } static mut LEVELS: u32 = 0; // This violates the idea of no shared state, and this doesn't internally // protect against races, so this function is `unsafe` unsafe fn bump_levels_unsafe() -> u32 { unsafe { let ret = LEVELS; LEVELS += 1; return ret; } } // As an alternative to `bump_levels_unsafe`, this function is safe, assuming // that we have an atomic_add function which returns the old value. This // function is safe only if no other code accesses the static in a non-atomic // fashion. If such accesses are possible (such as in `bump_levels_unsafe`), // then this would need to be `unsafe` to indicate to the caller that they // must still guard against concurrent access. fn bump_levels_safe() -> u32 { unsafe { return atomic_add(&raw mut LEVELS, 1); } } }
Mutable statics have the same restrictions as normal statics, except that the
type does not have to implement the Sync
trait.
Using Statics or Consts
It can be confusing whether or not you should use a constant item or a static item. Constants should, in general, be preferred over statics unless one of the following are true:
- Large amounts of data are being stored
- The single-address property of statics is required.
- Interior mutability is required.
Traits
Syntax
Trait :
unsafe
?trait
IDENTIFIER GenericParams? (:
TypeParamBounds? )? WhereClause?{
InnerAttribute*
AssociatedItem*
}
A trait describes an abstract interface that types can implement. This interface consists of associated items, which come in three varieties:
The trait declaration defines a trait in the type namespace of the module or block where it is located. Associated items are defined as members of the trait within their respective namespaces. Associated types are defined in the type namespace. Associated constants and associated functions are defined in the value namespace.
All traits define an implicit type parameter Self
that refers to “the type
that is implementing this interface”. Traits may also contain additional type
parameters. These type parameters, including Self
, may be constrained by
other traits and so forth as usual.
Traits are implemented for specific types through separate implementations.
Trait functions may omit the function body by replacing it with a semicolon. This indicates that the implementation must define the function. If the trait function defines a body, this definition acts as a default for any implementation which does not override it. Similarly, associated constants may omit the equals sign and expression to indicate implementations must define the constant value. Associated types must never define the type, the type may only be specified in an implementation.
#![allow(unused)] fn main() { // Examples of associated trait items with and without definitions. trait Example { const CONST_NO_DEFAULT: i32; const CONST_WITH_DEFAULT: i32 = 99; type TypeNoDefault; fn method_without_default(&self); fn method_with_default(&self) {} } }
Trait functions are not allowed to be const
.
Trait bounds
Generic items may use traits as bounds on their type parameters.
Generic Traits
Type parameters can be specified for a trait to make it generic. These appear after the trait name, using the same syntax used in generic functions.
#![allow(unused)] fn main() { trait Seq<T> { fn len(&self) -> u32; fn elt_at(&self, n: u32) -> T; fn iter<F>(&self, f: F) where F: Fn(T); } }
Object Safety
Object safe traits can be the base trait of a trait object. A trait is object safe if it has the following qualities (defined in RFC 255):
- All supertraits must also be object safe.
Sized
must not be a supertrait. In other words, it must not requireSelf: Sized
.- It must not have any associated constants.
- It must not have any associated types with generics.
- All associated functions must either be dispatchable from a trait object or be explicitly non-dispatchable:
- Dispatchable functions must:
- Not have any type parameters (although lifetime parameters are allowed).
- Be a method that does not use
Self
except in the type of the receiver. - Have a receiver with one of the following types:
- Not have an opaque return type; that is,
- Not be an
async fn
(which has a hiddenFuture
type). - Not have a return position
impl Trait
type (fn example(&self) -> impl Trait
).
- Not be an
- Not have a
where Self: Sized
bound (receiver type ofSelf
(i.e.self
) implies this).
- Explicitly non-dispatchable functions require:
- Have a
where Self: Sized
bound (receiver type ofSelf
(i.e.self
) implies this).
- Have a
- Dispatchable functions must:
#![allow(unused)] fn main() { use std::rc::Rc; use std::sync::Arc; use std::pin::Pin; // Examples of object safe methods. trait TraitMethods { fn by_ref(self: &Self) {} fn by_ref_mut(self: &mut Self) {} fn by_box(self: Box<Self>) {} fn by_rc(self: Rc<Self>) {} fn by_arc(self: Arc<Self>) {} fn by_pin(self: Pin<&Self>) {} fn with_lifetime<'a>(self: &'a Self) {} fn nested_pin(self: Pin<Arc<Self>>) {} } struct S; impl TraitMethods for S {} let t: Box<dyn TraitMethods> = Box::new(S); }
#![allow(unused)] fn main() { // This trait is object-safe, but these methods cannot be dispatched on a trait object. trait NonDispatchable { // Non-methods cannot be dispatched. fn foo() where Self: Sized {} // Self type isn't known until runtime. fn returns(&self) -> Self where Self: Sized; // `other` may be a different concrete type of the receiver. fn param(&self, other: Self) where Self: Sized {} // Generics are not compatible with vtables. fn typed<T>(&self, x: T) where Self: Sized {} } struct S; impl NonDispatchable for S { fn returns(&self) -> Self where Self: Sized { S } } let obj: Box<dyn NonDispatchable> = Box::new(S); obj.returns(); // ERROR: cannot call with Self return obj.param(S); // ERROR: cannot call with Self parameter obj.typed(1); // ERROR: cannot call with generic type }
#![allow(unused)] fn main() { use std::rc::Rc; // Examples of non-object safe traits. trait NotObjectSafe { const CONST: i32 = 1; // ERROR: cannot have associated const fn foo() {} // ERROR: associated function without Sized fn returns(&self) -> Self; // ERROR: Self in return type fn typed<T>(&self, x: T) {} // ERROR: has generic type parameters fn nested(self: Rc<Box<Self>>) {} // ERROR: nested receiver not yet supported } struct S; impl NotObjectSafe for S { fn returns(&self) -> Self { S } } let obj: Box<dyn NotObjectSafe> = Box::new(S); // ERROR }
#![allow(unused)] fn main() { // Self: Sized traits are not object-safe. trait TraitWithSize where Self: Sized {} struct S; impl TraitWithSize for S {} let obj: Box<dyn TraitWithSize> = Box::new(S); // ERROR }
#![allow(unused)] fn main() { // Not object safe if `Self` is a type argument. trait Super<A> {} trait WithSelf: Super<Self> where Self: Sized {} struct S; impl<A> Super<A> for S {} impl WithSelf for S {} let obj: Box<dyn WithSelf> = Box::new(S); // ERROR: cannot use `Self` type parameter }
Supertraits
Supertraits are traits that are required to be implemented for a type to implement a specific trait. Furthermore, anywhere a generic or trait object is bounded by a trait, it has access to the associated items of its supertraits.
Supertraits are declared by trait bounds on the Self
type of a trait and
transitively the supertraits of the traits declared in those trait bounds. It is
an error for a trait to be its own supertrait.
The trait with a supertrait is called a subtrait of its supertrait.
The following is an example of declaring Shape
to be a supertrait of Circle
.
#![allow(unused)] fn main() { trait Shape { fn area(&self) -> f64; } trait Circle : Shape { fn radius(&self) -> f64; } }
And the following is the same example, except using where clauses.
#![allow(unused)] fn main() { trait Shape { fn area(&self) -> f64; } trait Circle where Self: Shape { fn radius(&self) -> f64; } }
This next example gives radius
a default implementation using the area
function from Shape
.
#![allow(unused)] fn main() { trait Shape { fn area(&self) -> f64; } trait Circle where Self: Shape { fn radius(&self) -> f64 { // A = pi * r^2 // so algebraically, // r = sqrt(A / pi) (self.area() /std::f64::consts::PI).sqrt() } } }
This next example calls a supertrait method on a generic parameter.
#![allow(unused)] fn main() { trait Shape { fn area(&self) -> f64; } trait Circle : Shape { fn radius(&self) -> f64; } fn print_area_and_radius<C: Circle>(c: C) { // Here we call the area method from the supertrait `Shape` of `Circle`. println!("Area: {}", c.area()); println!("Radius: {}", c.radius()); } }
Similarly, here is an example of calling supertrait methods on trait objects.
#![allow(unused)] fn main() { trait Shape { fn area(&self) -> f64; } trait Circle : Shape { fn radius(&self) -> f64; } struct UnitCircle; impl Shape for UnitCircle { fn area(&self) -> f64 { std::f64::consts::PI } } impl Circle for UnitCircle { fn radius(&self) -> f64 { 1.0 } } let circle = UnitCircle; let circle = Box::new(circle) as Box<dyn Circle>; let nonsense = circle.radius() * circle.area(); }
Unsafe traits
Traits items that begin with the unsafe
keyword indicate that implementing the
trait may be unsafe. It is safe to use a correctly implemented unsafe trait.
The trait implementation must also begin with the unsafe
keyword.
Sync
and Send
are examples of unsafe traits.
Parameter patterns
Function or method declarations without a body only allow IDENTIFIER or
_
wild card patterns. mut
IDENTIFIER is currently
allowed, but it is deprecated and will become a hard error in the future.
In the 2015 edition, the pattern for a trait function or method parameter is optional:
#![allow(unused)] fn main() { // 2015 Edition trait T { fn f(i32); // Parameter identifiers are not required. } }
The kinds of patterns for parameters is limited to one of the following:
- IDENTIFIER
mut
IDENTIFIER_
&
IDENTIFIER&&
IDENTIFIER
Beginning in the 2018 edition, function or method parameter patterns are no longer optional. Also, all irrefutable patterns are allowed as long as there is a body. Without a body, the limitations listed above are still in effect.
#![allow(unused)] fn main() { trait T { fn f1((a, b): (i32, i32)) {} fn f2(_: (i32, i32)); // Cannot use tuple pattern without a body. } }
Item visibility
Trait items syntactically allow a Visibility annotation, but this is
rejected when the trait is validated. This allows items to be parsed with a
unified syntax across different contexts where they are used. As an example,
an empty vis
macro fragment specifier can be used for trait items, where the
macro rule may be used in other situations where visibility is allowed.
macro_rules! create_method { ($vis:vis $name:ident) => { $vis fn $name(&self) {} }; } trait T1 { // Empty `vis` is allowed. create_method! { method_of_t1 } } struct S; impl S { // Visibility is allowed here. create_method! { pub method_of_s } } impl T1 for S {} fn main() { let s = S; s.method_of_t1(); s.method_of_s(); }
Implementations
Syntax
Implementation :
InherentImpl | TraitImplInherentImpl :
impl
GenericParams? Type WhereClause?{
InnerAttribute*
AssociatedItem*
}
TraitImpl :
unsafe
?impl
GenericParams?!
? TypePathfor
Type
WhereClause?
{
InnerAttribute*
AssociatedItem*
}
An implementation is an item that associates items with an implementing type.
Implementations are defined with the keyword impl
and contain functions
that belong to an instance of the type that is being implemented or to the
type statically.
There are two types of implementations:
- inherent implementations
- trait implementations
Inherent Implementations
An inherent implementation is defined as the sequence of the impl
keyword,
generic type declarations, a path to a nominal type, a where clause, and a
bracketed set of associable items.
The nominal type is called the implementing type and the associable items are the associated items to the implementing type.
Inherent implementations associate the contained items to the implementing type. Inherent implementations can contain associated functions (including methods) and associated constants. They cannot contain associated type aliases.
The path to an associated item is any path to the implementing type, followed by the associated item’s identifier as the final path component.
A type can also have multiple inherent implementations. An implementing type must be defined within the same crate as the original type definition.
pub mod color { pub struct Color(pub u8, pub u8, pub u8); impl Color { pub const WHITE: Color = Color(255, 255, 255); } } mod values { use super::color::Color; impl Color { pub fn red() -> Color { Color(255, 0, 0) } } } pub use self::color::Color; fn main() { // Actual path to the implementing type and impl in the same module. color::Color::WHITE; // Impl blocks in different modules are still accessed through a path to the type. color::Color::red(); // Re-exported paths to the implementing type also work. Color::red(); // Does not work, because use in `values` is not pub. // values::Color::red(); }
Trait Implementations
A trait implementation is defined like an inherent implementation except that
the optional generic type declarations are followed by a trait, followed
by the keyword for
, followed by a path to a nominal type.
The trait is known as the implemented trait. The implementing type implements the implemented trait.
A trait implementation must define all non-default associated items declared by the implemented trait, may redefine default associated items defined by the implemented trait, and cannot define any other items.
The path to the associated items is <
followed by a path to the implementing
type followed by as
followed by a path to the trait followed by >
as a path
component followed by the associated item’s path component.
Unsafe traits require the trait implementation to begin with the unsafe
keyword.
#![allow(unused)] fn main() { #[derive(Copy, Clone)] struct Point {x: f64, y: f64}; type Surface = i32; struct BoundingBox {x: f64, y: f64, width: f64, height: f64}; trait Shape { fn draw(&self, s: Surface); fn bounding_box(&self) -> BoundingBox; } fn do_draw_circle(s: Surface, c: Circle) { } struct Circle { radius: f64, center: Point, } impl Copy for Circle {} impl Clone for Circle { fn clone(&self) -> Circle { *self } } impl Shape for Circle { fn draw(&self, s: Surface) { do_draw_circle(s, *self); } fn bounding_box(&self) -> BoundingBox { let r = self.radius; BoundingBox { x: self.center.x - r, y: self.center.y - r, width: 2.0 * r, height: 2.0 * r, } } } }
Trait Implementation Coherence
A trait implementation is considered incoherent if either the orphan rules check fails or there are overlapping implementation instances.
Two trait implementations overlap when there is a non-empty intersection of the traits the implementation is for, the implementations can be instantiated with the same type.
Orphan rules
Given impl<P1..=Pn> Trait<T1..=Tn> for T0
, an impl
is valid only if at
least one of the following is true:
Trait
is a local trait- All of
- At least one of the types
T0..=Tn
must be a local type. LetTi
be the first such type. - No uncovered type parameters
P1..=Pn
may appear inT0..Ti
(excludingTi
)
- At least one of the types
Only the appearance of uncovered type parameters is restricted.
Note that for the purposes of coherence, fundamental types are
special. The T
in Box<T>
is not considered covered, and Box<LocalType>
is considered local.
Generic Implementations
An implementation can take generic parameters, which can be used in the rest
of the implementation. Implementation parameters are written directly after the
impl
keyword.
#![allow(unused)] fn main() { trait Seq<T> { fn dummy(&self, _: T) { } } impl<T> Seq<T> for Vec<T> { /* ... */ } impl Seq<bool> for u32 { /* Treat the integer as a sequence of bits */ } }
Generic parameters constrain an implementation if the parameter appears at least once in one of:
- The implemented trait, if it has one
- The implementing type
- As an associated type in the bounds of a type that contains another parameter that constrains the implementation
Type and const parameters must always constrain the implementation. Lifetimes must constrain the implementation if the lifetime is used in an associated type.
Examples of constraining situations:
#![allow(unused)] fn main() { trait Trait{} trait GenericTrait<T> {} trait HasAssocType { type Ty; } struct Struct; struct GenericStruct<T>(T); struct ConstGenericStruct<const N: usize>([(); N]); // T constrains by being an argument to GenericTrait. impl<T> GenericTrait<T> for i32 { /* ... */ } // T constrains by being an argument to GenericStruct impl<T> Trait for GenericStruct<T> { /* ... */ } // Likewise, N constrains by being an argument to ConstGenericStruct impl<const N: usize> Trait for ConstGenericStruct<N> { /* ... */ } // T constrains by being in an associated type in a bound for type `U` which is // itself a generic parameter constraining the trait. impl<T, U> GenericTrait<U> for u32 where U: HasAssocType<Ty = T> { /* ... */ } // Like previous, except the type is `(U, isize)`. `U` appears inside the type // that includes `T`, and is not the type itself. impl<T, U> GenericStruct<U> where (U, isize): HasAssocType<Ty = T> { /* ... */ } }
Examples of non-constraining situations:
#![allow(unused)] fn main() { // The rest of these are errors, since they have type or const parameters that // do not constrain. // T does not constrain since it does not appear at all. impl<T> Struct { /* ... */ } // N does not constrain for the same reason. impl<const N: usize> Struct { /* ... */ } // Usage of T inside the implementation does not constrain the impl. impl<T> Struct { fn uses_t(t: &T) { /* ... */ } } // T is used as an associated type in the bounds for U, but U does not constrain. impl<T, U> Struct where U: HasAssocType<Ty = T> { /* ... */ } // T is used in the bounds, but not as an associated type, so it does not constrain. impl<T, U> GenericTrait<U> for u32 where U: GenericTrait<T> {} }
Example of an allowed unconstraining lifetime parameter:
#![allow(unused)] fn main() { struct Struct; impl<'a> Struct {} }
Example of a disallowed unconstraining lifetime parameter:
#![allow(unused)] fn main() { struct Struct; trait HasAssocType { type Ty; } impl<'a> HasAssocType for Struct { type Ty = &'a Struct; } }
Attributes on Implementations
Implementations may contain outer attributes before the impl
keyword and
inner attributes inside the brackets that contain the associated items. Inner
attributes must come before any associated items. The attributes that have
meaning here are cfg
, deprecated
, doc
, and the lint check
attributes.
External blocks
Syntax
ExternBlock :
unsafe
?extern
Abi?{
InnerAttribute*
ExternalItem*
}
ExternalItem :
OuterAttribute* (
MacroInvocationSemi
| ( Visibility? ( StaticItem | Function ) )
)
External blocks provide declarations of items that are not defined in the current crate and are the basis of Rust’s foreign function interface. These are akin to unchecked imports.
Two kinds of item declarations are allowed in external blocks: functions and
statics. Calling functions or accessing statics that are declared in external
blocks is only allowed in an unsafe
context.
The external block defines its functions and statics in the value namespace of the module or block where it is located.
Functions
Functions within external blocks are declared in the same way as other Rust
functions, with the exception that they must not have a body and are instead
terminated by a semicolon. Patterns are not allowed in parameters, only
IDENTIFIER or _
may be used. The safe
and unsafe
function qualifiers are
allowed, but other function qualifiers (e.g. const
, async
, extern
) are
not.
Functions within external blocks may be called by Rust code, just like functions defined in Rust. The Rust compiler automatically translates between the Rust ABI and the foreign ABI.
A function declared in an extern block is implicitly unsafe
unless the safe
function qualifier is present.
When coerced to a function pointer, a function declared in an extern block has
type extern "abi" for<'l1, ..., 'lm> fn(A1, ..., An) -> R
, where 'l1
,
… 'lm
are its lifetime parameters, A1
, …, An
are the declared types of
its parameters, R
is the declared return type.
Statics
Statics within external blocks are declared in the same way as statics outside of external blocks,
except that they do not have an expression initializing their value.
Unless a static item declared in an extern block is qualified as safe
, it is unsafe
to access that item, whether or
not it’s mutable, because there is nothing guaranteeing that the bit pattern at the static’s
memory is valid for the type it is declared with, since some arbitrary (e.g. C) code is in charge
of initializing the static.
Extern statics can be either immutable or mutable just like statics outside of external blocks.
An immutable static must be initialized before any Rust code is executed. It is not enough for
the static to be initialized before Rust code reads from it.
Once Rust code runs, mutating an immutable static (from inside or outside Rust) is UB,
except if the mutation happens to bytes inside of an UnsafeCell
.
ABI
By default external blocks assume that the library they are calling uses the
standard C ABI on the specific platform. Other ABIs may be specified using an
abi
string, as shown here:
#![allow(unused)] fn main() { // Interface to the Windows API unsafe extern "stdcall" { } }
There are three ABI strings which are cross-platform, and which all compilers are guaranteed to support:
unsafe extern "Rust"
– The default ABI when you write a normalfn foo()
in any Rust code.unsafe extern "C"
– This is the same asextern fn foo()
; whatever the default your C compiler supports.unsafe extern "system"
– Usually the same asextern "C"
, except on Win32, in which case it’s"stdcall"
, or what you should use to link to the Windows API itself
There are also some platform-specific ABI strings:
unsafe extern "cdecl"
– The default for x86_32 C code.unsafe extern "stdcall"
– The default for the Win32 API on x86_32.unsafe extern "win64"
– The default for C code on x86_64 Windows.unsafe extern "sysv64"
– The default for C code on non-Windows x86_64.unsafe extern "aapcs"
– The default for ARM.unsafe extern "fastcall"
– Thefastcall
ABI – corresponds to MSVC’s__fastcall
and GCC and clang’s__attribute__((fastcall))
unsafe extern "vectorcall"
– Thevectorcall
ABI – corresponds to MSVC’s__vectorcall
and clang’s__attribute__((vectorcall))
unsafe extern "thiscall"
– The default for C++ member functions on MSVC – corresponds to MSVC’s__thiscall
and GCC and clang’s__attribute__((thiscall))
unsafe extern "efiapi"
– The ABI used for UEFI functions.
Variadic functions
Functions within external blocks may be variadic by specifying ...
as the
last argument. The variadic parameter may optionally be specified with an
identifier.
#![allow(unused)] fn main() { unsafe extern "C" { safe fn foo(...); unsafe fn bar(x: i32, ...); unsafe fn with_name(format: *const u8, args: ...); } }
Attributes on extern blocks
The following attributes control the behavior of external blocks.
The link
attribute
The link
attribute specifies the name of a native library that the
compiler should link with for the items within an extern
block. It uses the
MetaListNameValueStr syntax to specify its inputs. The name
key is the
name of the native library to link. The kind
key is an optional value which
specifies the kind of library with the following possible values:
dylib
— Indicates a dynamic library. This is the default ifkind
is not specified.static
— Indicates a static library.framework
— Indicates a macOS framework. This is only valid for macOS targets.raw-dylib
— Indicates a dynamic library where the compiler will generate an import library to link against (seedylib
versusraw-dylib
below for details). This is only valid for Windows targets.
The name
key must be included if kind
is specified.
The optional modifiers
argument is a way to specify linking modifiers for the
library to link.
Modifiers are specified as a comma-delimited string with each modifier prefixed
with either a +
or -
to indicate that the modifier is enabled or disabled,
respectively.
Specifying multiple modifiers
arguments in a single link
attribute,
or multiple identical modifiers in the same modifiers
argument is not currently supported.
Example: #[link(name = "mylib", kind = "static", modifiers = "+whole-archive")]
.
The wasm_import_module
key may be used to specify the WebAssembly module
name for the items within an extern
block when importing symbols from the
host environment. The default module name is env
if wasm_import_module
is
not specified.
#[link(name = "crypto")]
unsafe extern {
// …
}
#[link(name = "CoreFoundation", kind = "framework")]
unsafe extern {
// …
}
#[link(wasm_import_module = "foo")]
unsafe extern {
// …
}
It is valid to add the link
attribute on an empty extern block. You can use
this to satisfy the linking requirements of extern blocks elsewhere in your
code (including upstream crates) instead of adding the attribute to each extern
block.
Linking modifiers: bundle
This modifier is only compatible with the static
linking kind.
Using any other kind will result in a compiler error.
When building a rlib or staticlib +bundle
means that the native static library
will be packed into the rlib or staticlib archive, and then retrieved from there
during linking of the final binary.
When building a rlib -bundle
means that the native static library is registered as a dependency
of that rlib “by name”, and object files from it are included only during linking of the final
binary, the file search by that name is also performed during final linking.
When building a staticlib -bundle
means that the native static library is simply not included
into the archive and some higher level build system will need to add it later during linking of
the final binary.
This modifier has no effect when building other targets like executables or dynamic libraries.
The default for this modifier is +bundle
.
More implementation details about this modifier can be found in
bundle
documentation for rustc.
Linking modifiers: whole-archive
This modifier is only compatible with the static
linking kind.
Using any other kind will result in a compiler error.
+whole-archive
means that the static library is linked as a whole archive
without throwing any object files away.
The default for this modifier is -whole-archive
.
More implementation details about this modifier can be found in
whole-archive
documentation for rustc.
Linking modifiers: verbatim
This modifier is compatible with all linking kinds.
+verbatim
means that rustc itself won’t add any target-specified library prefixes or suffixes
(like lib
or .a
) to the library name, and will try its best to ask for the same thing from the
linker.
-verbatim
means that rustc will either add a target-specific prefix and suffix to the library
name before passing it to linker, or won’t prevent linker from implicitly adding it.
The default for this modifier is -verbatim
.
More implementation details about this modifier can be found in
verbatim
documentation for rustc.
dylib
versus raw-dylib
On Windows, linking against a dynamic library requires that an import library is provided to the linker: this is a special static library that declares all of the symbols exported by the dynamic library in such a way that the linker knows that they have to be dynamically loaded at runtime.
Specifying kind = "dylib"
instructs the Rust compiler to link an import
library based on the name
key. The linker will then use its normal library
resolution logic to find that import library. Alternatively, specifying
kind = "raw-dylib"
instructs the compiler to generate an import library
during compilation and provide that to the linker instead.
raw-dylib
is only supported on Windows. Using it when targeting other
platforms will result in a compiler error.
The import_name_type
key
On x86 Windows, names of functions are “decorated” (i.e., have a specific prefix
and/or suffix added) to indicate their calling convention. For example, a
stdcall
calling convention function with the name fn1
that has no arguments
would be decorated as _fn1@0
. However, the PE Format does also permit names
to have no prefix or be undecorated. Additionally, the MSVC and GNU toolchains
use different decorations for the same calling conventions which means, by
default, some Win32 functions cannot be called using the raw-dylib
link kind
via the GNU toolchain.
To allow for these differences, when using the raw-dylib
link kind you may
also specify the import_name_type
key with one of the following values to
change how functions are named in the generated import library:
decorated
: The function name will be fully-decorated using the MSVC toolchain format.noprefix
: The function name will be decorated using the MSVC toolchain format, but skipping the leading?
,@
, or optionally_
.undecorated
: The function name will not be decorated.
If the import_name_type
key is not specified, then the function name will be
fully-decorated using the target toolchain’s format.
Variables are never decorated and so the import_name_type
key has no effect on
how they are named in the generated import library.
The import_name_type
key is only supported on x86 Windows. Using it when
targeting other platforms will result in a compiler error.
The link_name
attribute
The link_name
attribute may be specified on declarations inside an extern
block to indicate the symbol to import for the given function or static. It
uses the MetaNameValueStr syntax to specify the name of the symbol.
#![allow(unused)] fn main() { unsafe extern { #[link_name = "actual_symbol_name"] safe fn name_in_rust(); } }
Using this attribute with the link_ordinal
attribute will result in a
compiler error.
The link_ordinal
attribute
The link_ordinal
attribute can be applied on declarations inside an extern
block to indicate the numeric ordinal to use when generating the import library
to link against. An ordinal is a unique number per symbol exported by a dynamic
library on Windows and can be used when the library is being loaded to find
that symbol rather than having to look it up by name.
Warning: link_ordinal
should only be used in cases where the ordinal of the
symbol is known to be stable: if the ordinal of a symbol is not explicitly set
when its containing binary is built then one will be automatically assigned to
it, and that assigned ordinal may change between builds of the binary.
#[link(name = "exporter", kind = "raw-dylib")]
unsafe extern "stdcall" {
#[link_ordinal(15)]
safe fn imported_function_stdcall(i: i32);
}
This attribute is only used with the raw-dylib
linking kind.
Using any other kind will result in a compiler error.
Using this attribute with the link_name
attribute will result in a
compiler error.
Attributes on function parameters
Attributes on extern function parameters follow the same rules and restrictions as regular function parameters.
Generic parameters
Syntax
GenericParams :
<
>
|<
(GenericParam,
)* GenericParam,
?>
GenericParam :
OuterAttribute* ( LifetimeParam | TypeParam | ConstParam )LifetimeParam :
LIFETIME_OR_LABEL (:
LifetimeBounds )?TypeParam :
IDENTIFIER (:
TypeParamBounds? )? (=
Type )?ConstParam:
const
IDENTIFIER:
Type (=
Block | IDENTIFIER | -?LITERAL )?
Functions, type aliases, structs, enumerations, unions, traits, and
implementations may be parameterized by types, constants, and lifetimes. These
parameters are listed in angle brackets (<...>
),
usually immediately after the name of the item and before its definition. For
implementations, which don’t have a name, they come directly after impl
.
The order of generic parameters is restricted to lifetime parameters and then type and const parameters intermixed.
The same parameter name may not be declared more than once in a GenericParams list.
Some examples of items with type, const, and lifetime parameters:
#![allow(unused)] fn main() { fn foo<'a, T>() {} trait A<U> {} struct Ref<'a, T> where T: 'a { r: &'a T } struct InnerArray<T, const N: usize>([T; N]); struct EitherOrderWorks<const N: bool, U>(U); }
Generic parameters are in scope within the item definition where they are declared. They are not in scope for items declared within the body of a function as described in item declarations. See generic parameter scopes for more details.
References, raw pointers, arrays, slices, tuples, and function pointers have lifetime or type parameters as well, but are not referred to with path syntax.
'_
is not a valid lifetime parameter.
Const generics
Const generic parameters allow items to be generic over constant values. The const identifier introduces a name in the value namespace for the constant parameter, and all instances of the item must be instantiated with a value of the given type.
The only allowed types of const parameters are u8
, u16
, u32
, u64
, u128
, usize
,
i8
, i16
, i32
, i64
, i128
, isize
, char
and bool
.
Const parameters can be used anywhere a const item can be used, with the exception that when used in a type or array repeat expression, it must be standalone (as described below). That is, they are allowed in the following places:
- As an applied const to any type which forms a part of the signature of the item in question.
- As part of a const expression used to define an associated const, or as a parameter to an associated type.
- As a value in any runtime expression in the body of any functions in the item.
- As a parameter to any type used in the body of any functions in the item.
- As a part of the type of any fields in the item.
#![allow(unused)] fn main() { // Examples where const generic parameters can be used. // Used in the signature of the item itself. fn foo<const N: usize>(arr: [i32; N]) { // Used as a type within a function body. let x: [i32; N]; // Used as an expression. println!("{}", N * 2); } // Used as a field of a struct. struct Foo<const N: usize>([i32; N]); impl<const N: usize> Foo<N> { // Used as an associated constant. const CONST: usize = N * 4; } trait Trait { type Output; } impl<const N: usize> Trait for Foo<N> { // Used as an associated type. type Output = [i32; N]; } }
#![allow(unused)] fn main() { // Examples where const generic parameters cannot be used. fn foo<const N: usize>() { // Cannot use in item definitions within a function body. const BAD_CONST: [usize; N] = [1; N]; static BAD_STATIC: [usize; N] = [1; N]; fn inner(bad_arg: [usize; N]) { let bad_value = N * 2; } type BadAlias = [usize; N]; struct BadStruct([usize; N]); } }
As a further restriction, const parameters may only appear as a standalone
argument inside of a type or array repeat expression. In those contexts,
they may only be used as a single segment path expression, possibly inside a
block (such as N
or {N}
). That is, they cannot be combined with other
expressions.
#![allow(unused)] fn main() { // Examples where const parameters may not be used. // Not allowed to combine in other expressions in types, such as the // arithmetic expression in the return type here. fn bad_function<const N: usize>() -> [u8; {N + 1}] { // Similarly not allowed for array repeat expressions. [1; {N + 1}] } }
A const argument in a path specifies the const value to use for that item.
The argument must be a const expression of the type ascribed to the const
parameter. The const expression must be a block expression
(surrounded with braces) unless it is a single path segment (an IDENTIFIER)
or a literal (with a possibly leading -
token).
Note: This syntactic restriction is necessary to avoid requiring infinite lookahead when parsing an expression inside of a type.
#![allow(unused)] fn main() { fn double<const N: i32>() { println!("doubled: {}", N * 2); } const SOME_CONST: i32 = 12; fn example() { // Example usage of a const argument. double::<9>(); double::<-123>(); double::<{7 + 8}>(); double::<SOME_CONST>(); double::<{ SOME_CONST + 5 }>(); } }
When there is ambiguity if a generic argument could be resolved as either a type or const argument, it is always resolved as a type. Placing the argument in a block expression can force it to be interpreted as a const argument.
#![allow(unused)] fn main() { type N = u32; struct Foo<const N: usize>; // The following is an error, because `N` is interpreted as the type alias `N`. fn foo<const N: usize>() -> Foo<N> { todo!() } // ERROR // Can be fixed by wrapping in braces to force it to be interpreted as the `N` // const parameter: fn bar<const N: usize>() -> Foo<{ N }> { todo!() } // ok }
Unlike type and lifetime parameters, const parameters can be declared without being used inside of a parameterized item, with the exception of implementations as described in generic implementations:
#![allow(unused)] fn main() { // ok struct Foo<const N: usize>; enum Bar<const M: usize> { A, B } // ERROR: unused parameter struct Baz<T>; struct Biz<'a>; struct Unconstrained; impl<const N: usize> Unconstrained {} }
When resolving a trait bound obligation, the exhaustiveness of all
implementations of const parameters is not considered when determining if the
bound is satisfied. For example, in the following, even though all possible
const values for the bool
type are implemented, it is still an error that
the trait bound is not satisfied:
#![allow(unused)] fn main() { struct Foo<const B: bool>; trait Bar {} impl Bar for Foo<true> {} impl Bar for Foo<false> {} fn needs_bar(_: impl Bar) {} fn generic<const B: bool>() { let v = Foo::<B>; needs_bar(v); // ERROR: trait bound `Foo<B>: Bar` is not satisfied } }
Where clauses
Syntax
WhereClause :
where
( WhereClauseItem,
)* WhereClauseItem ?WhereClauseItem :
LifetimeWhereClauseItem
| TypeBoundWhereClauseItemLifetimeWhereClauseItem :
Lifetime:
LifetimeBoundsTypeBoundWhereClauseItem :
ForLifetimes? Type:
TypeParamBounds?
Where clauses provide another way to specify bounds on type and lifetime parameters as well as a way to specify bounds on types that aren’t type parameters.
The for
keyword can be used to introduce higher-ranked lifetimes. It only
allows LifetimeParam parameters.
#![allow(unused)] fn main() { struct A<T> where T: Iterator, // Could use A<T: Iterator> instead T::Item: Copy, // Bound on an associated type String: PartialEq<T>, // Bound on `String`, using the type parameter i32: Default, // Allowed, but not useful { f: T, } }
Attributes
Generic lifetime and type parameters allow attributes on them. There are no built-in attributes that do anything in this position, although custom derive attributes may give meaning to it.
This example shows using a custom derive attribute to modify the meaning of a generic parameter.
// Assume that the derive for MyFlexibleClone declared `my_flexible_clone` as
// an attribute it understands.
#[derive(MyFlexibleClone)]
struct Foo<#[my_flexible_clone(unbounded)] H> {
a: *const H
}
Associated Items
Syntax
AssociatedItem :
OuterAttribute* (
MacroInvocationSemi
| ( Visibility? ( TypeAlias | ConstantItem | Function ) )
)
Associated Items are the items declared in traits or defined in implementations. They are called this because they are defined on an associate type — the type in the implementation. They are a subset of the kinds of items you can declare in a module. Specifically, there are associated functions (including methods), associated types, and associated constants.
Associated items are useful when the associated item logically is related to the
associating item. For example, the is_some
method on Option
is intrinsically
related to Options, so should be associated.
Every associated item kind comes in two varieties: definitions that contain the actual implementation and declarations that declare signatures for definitions.
It is the declarations that make up the contract of traits and what is available on generic types.
Associated functions and methods
Associated functions are functions associated with a type.
An associated function declaration declares a signature for an associated
function definition. It is written as a function item, except the
function body is replaced with a ;
.
The identifier is the name of the function. The generics, parameter list, return type, and where clause of the associated function must be the same as the associated function declarations’s.
An associated function definition defines a function associated with another type. It is written the same as a function item.
An example of a common associated function is a new
function that returns
a value of the type the associated function is associated with.
struct Struct { field: i32 } impl Struct { fn new() -> Struct { Struct { field: 0i32 } } } fn main () { let _struct = Struct::new(); }
When the associated function is declared on a trait, the function can also be
called with a path that is a path to the trait appended by the name of the
trait. When this happens, it is substituted for <_ as Trait>::function_name
.
#![allow(unused)] fn main() { trait Num { fn from_i32(n: i32) -> Self; } impl Num for f64 { fn from_i32(n: i32) -> f64 { n as f64 } } // These 4 are all equivalent in this case. let _: f64 = Num::from_i32(42); let _: f64 = <_ as Num>::from_i32(42); let _: f64 = <f64 as Num>::from_i32(42); let _: f64 = f64::from_i32(42); }
Methods
Associated functions whose first parameter is named self
are called methods
and may be invoked using the method call operator, for example, x.foo()
, as
well as the usual function call notation.
If the type of the self
parameter is specified, it is limited to types resolving
to one generated by the following grammar (where 'lt
denotes some arbitrary
lifetime):
P = &'lt S | &'lt mut S | Box<S> | Rc<S> | Arc<S> | Pin<P>
S = Self | P
The Self
terminal in this grammar denotes a type resolving to the implementing type.
This can also include the contextual type alias Self
, other type aliases,
or associated type projections resolving to the implementing type.
#![allow(unused)] fn main() { use std::rc::Rc; use std::sync::Arc; use std::pin::Pin; // Examples of methods implemented on struct `Example`. struct Example; type Alias = Example; trait Trait { type Output; } impl Trait for Example { type Output = Example; } impl Example { fn by_value(self: Self) {} fn by_ref(self: &Self) {} fn by_ref_mut(self: &mut Self) {} fn by_box(self: Box<Self>) {} fn by_rc(self: Rc<Self>) {} fn by_arc(self: Arc<Self>) {} fn by_pin(self: Pin<&Self>) {} fn explicit_type(self: Arc<Example>) {} fn with_lifetime<'a>(self: &'a Self) {} fn nested<'a>(self: &mut &'a Arc<Rc<Box<Alias>>>) {} fn via_projection(self: <Example as Trait>::Output) {} } }
Shorthand syntax can be used without specifying a type, which have the following equivalents:
Shorthand | Equivalent |
---|---|
self | self: Self |
&'lifetime self | self: &'lifetime Self |
&'lifetime mut self | self: &'lifetime mut Self |
Note: Lifetimes can be, and usually are, elided with this shorthand.
If the self
parameter is prefixed with mut
, it becomes a mutable variable,
similar to regular parameters using a mut
identifier pattern. For example:
#![allow(unused)] fn main() { trait Changer: Sized { fn change(mut self) {} fn modify(mut self: Box<Self>) {} } }
As an example of methods on a trait, consider the following:
#![allow(unused)] fn main() { type Surface = i32; type BoundingBox = i32; trait Shape { fn draw(&self, surface: Surface); fn bounding_box(&self) -> BoundingBox; } }
This defines a trait with two methods. All values that have implementations
of this trait while the trait is in scope can have their draw
and
bounding_box
methods called.
#![allow(unused)] fn main() { type Surface = i32; type BoundingBox = i32; trait Shape { fn draw(&self, surface: Surface); fn bounding_box(&self) -> BoundingBox; } struct Circle { // ... } impl Shape for Circle { // ... fn draw(&self, _: Surface) {} fn bounding_box(&self) -> BoundingBox { 0i32 } } impl Circle { fn new() -> Circle { Circle{} } } let circle_shape = Circle::new(); let bounding_box = circle_shape.bounding_box(); }
Edition differences: In the 2015 edition, it is possible to declare trait methods with anonymous parameters (e.g.
fn foo(u8)
). This is deprecated and an error as of the 2018 edition. All parameters must have an argument name.
Attributes on method parameters
Attributes on method parameters follow the same rules and restrictions as regular function parameters.
Associated Types
Associated types are type aliases associated with another type. Associated types cannot be defined in inherent implementations nor can they be given a default implementation in traits.
An associated type declaration declares a signature for associated type
definitions. It is written in one of the following forms, where Assoc
is the
name of the associated type, Params
is a comma-separated list of type,
lifetime or const parameters, Bounds
is a plus-separated list of trait bounds
that the associated type must meet, and WhereBounds
is a comma-separated list
of bounds that the parameters must meet:
type Assoc;
type Assoc: Bounds;
type Assoc<Params>;
type Assoc<Params>: Bounds;
type Assoc<Params> where WhereBounds;
type Assoc<Params>: Bounds where WhereBounds;
The identifier is the name of the declared type alias. The optional trait bounds
must be fulfilled by the implementations of the type alias.
There is an implicit Sized
bound on associated types that can be relaxed using the special ?Sized
bound.
An associated type definition defines a type alias for the implementation
of a trait on a type. They are written similarly to an associated type declaration,
but cannot contain Bounds
, but instead must contain a Type
:
type Assoc = Type;
type Assoc<Params> = Type; // the type `Type` here may reference `Params`
type Assoc<Params> = Type where WhereBounds;
type Assoc<Params> where WhereBounds = Type; // deprecated, prefer the form above
If a type Item
has an associated type Assoc
from a trait Trait
, then
<Item as Trait>::Assoc
is a type that is an alias of the type specified in the
associated type definition. Furthermore, if Item
is a type parameter, then
Item::Assoc
can be used in type parameters.
Associated types may include generic parameters and where clauses; these are
often referred to as generic associated types, or GATs. If the type Thing
has an associated type Item
from a trait Trait
with the generics <'a>
, the
type can be named like <Thing as Trait>::Item<'x>
, where 'x
is some lifetime
in scope. In this case, 'x
will be used wherever 'a
appears in the associated
type definitions on impls.
trait AssociatedType { // Associated type declaration type Assoc; } struct Struct; struct OtherStruct; impl AssociatedType for Struct { // Associated type definition type Assoc = OtherStruct; } impl OtherStruct { fn new() -> OtherStruct { OtherStruct } } fn main() { // Usage of the associated type to refer to OtherStruct as <Struct as AssociatedType>::Assoc let _other_struct: OtherStruct = <Struct as AssociatedType>::Assoc::new(); }
An example of associated types with generics and where clauses:
struct ArrayLender<'a, T>(&'a mut [T; 16]); trait Lend { // Generic associated type declaration type Lender<'a> where Self: 'a; fn lend<'a>(&'a mut self) -> Self::Lender<'a>; } impl<T> Lend for [T; 16] { // Generic associated type definition type Lender<'a> = ArrayLender<'a, T> where Self: 'a; fn lend<'a>(&'a mut self) -> Self::Lender<'a> { ArrayLender(self) } } fn borrow<'a, T: Lend>(array: &'a mut T) -> <T as Lend>::Lender<'a> { array.lend() } fn main() { let mut array = [0usize; 16]; let lender = borrow(&mut array); }
Associated Types Container Example
Consider the following example of a Container
trait. Notice that the type is
available for use in the method signatures:
#![allow(unused)] fn main() { trait Container { type E; fn empty() -> Self; fn insert(&mut self, elem: Self::E); } }
In order for a type to implement this trait, it must not only provide
implementations for every method, but it must specify the type E
. Here’s an
implementation of Container
for the standard library type Vec
:
#![allow(unused)] fn main() { trait Container { type E; fn empty() -> Self; fn insert(&mut self, elem: Self::E); } impl<T> Container for Vec<T> { type E = T; fn empty() -> Vec<T> { Vec::new() } fn insert(&mut self, x: T) { self.push(x); } } }
Relationship between Bounds
and WhereBounds
In this example:
#![allow(unused)] fn main() { use std::fmt::Debug; trait Example { type Output<T>: Ord where T: Debug; } }
Given a reference to the associated type like <X as Example>::Output<Y>
, the associated type itself must be Ord
, and the type Y
must be Debug
.
Required where clauses on generic associated types
Generic associated type declarations on traits currently may require a list of where clauses, dependent on functions in the trait and how the GAT is used. These rules may be loosened in the future; updates can be found on the generic associated types initiative repository.
In a few words, these where clauses are required in order to maximize the allowed definitions of the associated type in impls. To do this, any clauses that can be proven to hold on functions (using the parameters of the function or trait) where a GAT appears as an input or output must also be written on the GAT itself.
#![allow(unused)] fn main() { trait LendingIterator { type Item<'x> where Self: 'x; fn next<'a>(&'a mut self) -> Self::Item<'a>; } }
In the above, on the next
function, we can prove that Self: 'a
, because of
the implied bounds from &'a mut self
; therefore, we must write the equivalent
bound on the GAT itself: where Self: 'x
.
When there are multiple functions in a trait that use the GAT, then the intersection of the bounds from the different functions are used, rather than the union.
#![allow(unused)] fn main() { trait Check<T> { type Checker<'x>; fn create_checker<'a>(item: &'a T) -> Self::Checker<'a>; fn do_check(checker: Self::Checker<'_>); } }
In this example, no bounds are required on the type Checker<'a>;
. While we
know that T: 'a
on create_checker
, we do not know that on do_check
. However,
if do_check
was commented out, then the where T: 'x
bound would be required
on Checker
.
The bounds on associated types also propagate required where clauses.
#![allow(unused)] fn main() { trait Iterable { type Item<'a> where Self: 'a; type Iterator<'a>: Iterator<Item = Self::Item<'a>> where Self: 'a; fn iter<'a>(&'a self) -> Self::Iterator<'a>; } }
Here, where Self: 'a
is required on Item
because of iter
. However, Item
is used in the bounds of Iterator
, the where Self: 'a
clause is also required
there.
Finally, any explicit uses of 'static
on GATs in the trait do not count towards
the required bounds.
#![allow(unused)] fn main() { trait StaticReturn { type Y<'a>; fn foo(&self) -> Self::Y<'static>; } }
Associated Constants
Associated constants are constants associated with a type.
An associated constant declaration declares a signature for associated
constant definitions. It is written as const
, then an identifier,
then :
, then a type, finished by a ;
.
The identifier is the name of the constant used in the path. The type is the type that the definition has to implement.
An associated constant definition defines a constant associated with a type. It is written the same as a constant item.
Associated constant definitions undergo constant evaluation only when referenced. Further, definitions that include generic parameters are evaluated after monomorphization.
struct Struct; struct GenericStruct<const ID: i32>; impl Struct { // Definition not immediately evaluated const PANIC: () = panic!("compile-time panic"); } impl<const ID: i32> GenericStruct<ID> { // Definition not immediately evaluated const NON_ZERO: () = if ID == 0 { panic!("contradiction") }; } fn main() { // Referencing Struct::PANIC causes compilation error let _ = Struct::PANIC; // Fine, ID is not 0 let _ = GenericStruct::<1>::NON_ZERO; // Compilation error from evaluating NON_ZERO with ID=0 let _ = GenericStruct::<0>::NON_ZERO; }
Associated Constants Examples
A basic example:
trait ConstantId { const ID: i32; } struct Struct; impl ConstantId for Struct { const ID: i32 = 1; } fn main() { assert_eq!(1, Struct::ID); }
Using default values:
trait ConstantIdDefault { const ID: i32 = 1; } struct Struct; struct OtherStruct; impl ConstantIdDefault for Struct {} impl ConstantIdDefault for OtherStruct { const ID: i32 = 5; } fn main() { assert_eq!(1, Struct::ID); assert_eq!(5, OtherStruct::ID); }
Attributes
Syntax
InnerAttribute :
#
!
[
Attr]
OuterAttribute :
#
[
Attr]
Attr :
SimplePath AttrInput?
|unsafe
(
SimplePath AttrInput?)
AttrInput :
DelimTokenTree
|=
Expression
An attribute is a general, free-form metadatum that is interpreted according to name, convention, language, and compiler version. Attributes are modeled on Attributes in ECMA-335, with the syntax coming from ECMA-334 (C#).
Inner attributes, written with a bang (!
) after the hash (#
), apply to the
item that the attribute is declared within. Outer attributes, written without
the bang after the hash, apply to the thing that follows the attribute.
The attribute consists of a path to the attribute, followed by an optional
delimited token tree whose interpretation is defined by the attribute.
Attributes other than macro attributes also allow the input to be an equals
sign (=
) followed by an expression. See the meta item
syntax below for more details.
An attribute may be unsafe to apply. To avoid undefined behavior when using
these attributes, certain obligations that cannot be checked by the compiler
must be met. To assert these have been, the attribute is wrapped in
unsafe(..)
, e.g. #[unsafe(no_mangle)]
.
The following attributes are unsafe:
Attributes can be classified into the following kinds:
Attributes may be applied to many things in the language:
- All item declarations accept outer attributes while external blocks, functions, implementations, and modules accept inner attributes.
- Most statements accept outer attributes (see Expression Attributes for limitations on expression statements).
- Block expressions accept outer and inner attributes, but only when they are the outer expression of an expression statement or the final expression of another block expression.
- Enum variants and struct and union fields accept outer attributes.
- Match expression arms accept outer attributes.
- Generic lifetime or type parameter accept outer attributes.
- Expressions accept outer attributes in limited situations, see Expression Attributes for details.
- Function, closure and function pointer
parameters accept outer attributes. This includes attributes on variadic parameters
denoted with
...
in function pointers and external blocks.
Some examples of attributes:
#![allow(unused)] fn main() { // General metadata applied to the enclosing module or crate. #![crate_type = "lib"] // A function marked as a unit test #[test] fn test_foo() { /* ... */ } // A conditionally-compiled module #[cfg(target_os = "linux")] mod bar { /* ... */ } // A lint attribute used to suppress a warning/error #[allow(non_camel_case_types)] type int8_t = i8; // Inner attribute applies to the entire function. fn some_unused_variables() { #![allow(unused_variables)] let x = (); let y = (); let z = (); } }
Meta Item Attribute Syntax
A “meta item” is the syntax used for the Attr rule by most built-in attributes. It has the following grammar:
Syntax
MetaItem :
SimplePath
| SimplePath=
Expression
| SimplePath(
MetaSeq?)
MetaSeq :
MetaItemInner (,
MetaItemInner )*,
?MetaItemInner :
MetaItem
| Expression
Expressions in meta items must macro-expand to literal expressions, which must not include integer or float type suffixes. Expressions which are not literal expressions will be syntactically accepted (and can be passed to proc-macros), but will be rejected after parsing.
Note that if the attribute appears within another macro, it will be expanded
after that outer macro. For example, the following code will expand the
Serialize
proc-macro first, which must preserve the include_str!
call in
order for it to be expanded:
#[derive(Serialize)]
struct Foo {
#[doc = include_str!("x.md")]
x: u32
}
Additionally, macros in attributes will be expanded only after all other attributes applied to the item:
#[macro_attr1] // expanded first
#[doc = mac!()] // `mac!` is expanded fourth.
#[macro_attr2] // expanded second
#[derive(MacroDerive1, MacroDerive2)] // expanded third
fn foo() {}
Various built-in attributes use different subsets of the meta item syntax to specify their inputs. The following grammar rules show some commonly used forms:
Syntax
MetaWord:
IDENTIFIERMetaNameValueStr:
IDENTIFIER=
(STRING_LITERAL | RAW_STRING_LITERAL)MetaListPaths:
IDENTIFIER(
( SimplePath (,
SimplePath)*,
? )?)
MetaListIdents:
IDENTIFIER(
( IDENTIFIER (,
IDENTIFIER)*,
? )?)
MetaListNameValueStr:
IDENTIFIER(
( MetaNameValueStr (,
MetaNameValueStr)*,
? )?)
Some examples of meta items are:
Style | Example |
---|---|
MetaWord | no_std |
MetaNameValueStr | doc = "example" |
MetaListPaths | allow(unused, clippy::inline_always) |
MetaListIdents | macro_use(foo, bar) |
MetaListNameValueStr | link(name = "CoreFoundation", kind = "framework") |
Active and inert attributes
An attribute is either active or inert. During attribute processing, active attributes remove themselves from the thing they are on while inert attributes stay on.
The cfg
and cfg_attr
attributes are active. The test
attribute is
inert when compiling for tests and active otherwise. Attribute macros are
active. All other attributes are inert.
Tool attributes
The compiler may allow attributes for external tools where each tool resides in its own module in the tool prelude. The first segment of the attribute path is the name of the tool, with one or more additional segments whose interpretation is up to the tool.
When a tool is not in use, the tool’s attributes are accepted without a warning. When the tool is in use, the tool is responsible for processing and interpretation of its attributes.
Tool attributes are not available if the no_implicit_prelude
attribute is
used.
#![allow(unused)] fn main() { // Tells the rustfmt tool to not format the following element. #[rustfmt::skip] struct S { } // Controls the "cyclomatic complexity" threshold for the clippy tool. #[clippy::cyclomatic_complexity = "100"] pub fn f() {} }
Note:
rustc
currently recognizes the tools “clippy”, “rustfmt”, “diagnostic”, “miri” and “rust_analyzer”.
Built-in attributes index
The following is an index of all built-in attributes.
- Conditional compilation
- Testing
test
— Marks a function as a test.ignore
— Disables a test function.should_panic
— Indicates a test should generate a panic.
- Derive
derive
— Automatic trait implementations.automatically_derived
— Marker for implementations created byderive
.
- Macros
macro_export
— Exports amacro_rules
macro for cross-crate usage.macro_use
— Expands macro visibility, or imports macros from other crates.proc_macro
— Defines a function-like macro.proc_macro_derive
— Defines a derive macro.proc_macro_attribute
— Defines an attribute macro.
- Diagnostics
allow
,expect
,warn
,deny
,forbid
— Alters the default lint level.deprecated
— Generates deprecation notices.must_use
— Generates a lint for unused values.diagnostic::on_unimplemented
— Hints the compiler to emit a certain error message if a trait is not implemented.
- ABI, linking, symbols, and FFI
link
— Specifies a native library to link with anextern
block.link_name
— Specifies the name of the symbol for functions or statics in anextern
block.link_ordinal
— Specifies the ordinal of the symbol for functions or statics in anextern
block.no_link
— Prevents linking an extern crate.repr
— Controls type layout.crate_type
— Specifies the type of crate (library, executable, etc.).no_main
— Disables emitting themain
symbol.export_name
— Specifies the exported symbol name for a function or static.link_section
— Specifies the section of an object file to use for a function or static.no_mangle
— Disables symbol name encoding.used
— Forces the compiler to keep a static item in the output object file.crate_name
— Specifies the crate name.
- Code generation
inline
— Hint to inline code.cold
— Hint that a function is unlikely to be called.no_builtins
— Disables use of certain built-in functions.target_feature
— Configure platform-specific code generation.track_caller
- Pass the parent call location tostd::panic::Location::caller()
.instruction_set
- Specify the instruction set used to generate a functions code
- Documentation
doc
— Specifies documentation. See The Rustdoc Book for more information. Doc comments are transformed intodoc
attributes.
- Preludes
no_std
— Removes std from the prelude.no_implicit_prelude
— Disables prelude lookups within a module.
- Modules
path
— Specifies the filename for a module.
- Limits
recursion_limit
— Sets the maximum recursion limit for certain compile-time operations.type_length_limit
— Sets the maximum size of a polymorphic type.
- Runtime
panic_handler
— Sets the function to handle panics.global_allocator
— Sets the global memory allocator.windows_subsystem
— Specifies the windows subsystem to link with.
- Features
feature
— Used to enable unstable or experimental compiler features. See The Unstable Book for features implemented inrustc
.
- Type System
non_exhaustive
— Indicate that a type will have more fields/variants added in future.
- Debugger
debugger_visualizer
— Embeds a file that specifies debugger output for a type.collapse_debuginfo
— Controls how macro invocations are encoded in debuginfo.
Testing attributes
The following attributes are used for specifying functions for performing
tests. Compiling a crate in “test” mode enables building the test functions
along with a test harness for executing the tests. Enabling the test mode also
enables the test
conditional compilation option.
The test
attribute
The test
attribute marks a function to be executed as a test. These
functions are only compiled when in test mode. Test functions must be free,
monomorphic functions that take no arguments, and the return type must implement the Termination
trait, for example:
()
Result<T, E> where T: Termination, E: Debug
!
Note: The test mode is enabled by passing the
--test
argument torustc
or usingcargo test
.
The test harness calls the returned value’s report
method, and classifies the test as passed or failed depending on whether the resulting ExitCode
represents successful termination.
In particular:
- Tests that return
()
pass as long as they terminate and do not panic. - Tests that return a
Result<(), E>
pass as long as they returnOk(())
. - Tests that return
ExitCode::SUCCESS
pass, and tests that returnExitCode::FAILURE
fail. - Tests that do not terminate neither pass nor fail.
#![allow(unused)] fn main() { use std::io; fn setup_the_thing() -> io::Result<i32> { Ok(1) } fn do_the_thing(s: &i32) -> io::Result<()> { Ok(()) } #[test] fn test_the_thing() -> io::Result<()> { let state = setup_the_thing()?; // expected to succeed do_the_thing(&state)?; // expected to succeed Ok(()) } }
The ignore
attribute
A function annotated with the test
attribute can also be annotated with the
ignore
attribute. The ignore
attribute tells the test harness to not
execute that function as a test. It will still be compiled when in test mode.
The ignore
attribute may optionally be written with the MetaNameValueStr
syntax to specify a reason why the test is ignored.
#![allow(unused)] fn main() { #[test] #[ignore = "not yet implemented"] fn mytest() { // … } }
Note: The
rustc
test harness supports the--include-ignored
flag to force ignored tests to be run.
The should_panic
attribute
A function annotated with the test
attribute that returns ()
can also be
annotated with the should_panic
attribute. The should_panic
attribute
makes the test only pass if it actually panics.
The should_panic
attribute may optionally take an input string that must
appear within the panic message. If the string is not found in the message,
then the test will fail. The string may be passed using the
MetaNameValueStr syntax or the MetaListNameValueStr syntax with an
expected
field.
#![allow(unused)] fn main() { #[test] #[should_panic(expected = "values don't match")] fn mytest() { assert_eq!(1, 2, "values don't match"); } }
Derive
The derive
attribute allows new items to be automatically generated for
data structures. It uses the MetaListPaths syntax to specify a list of
traits to implement or paths to derive macros to process.
For example, the following will create an impl
item for the
PartialEq
and Clone
traits for Foo
, and the type parameter T
will be
given the PartialEq
or Clone
constraints for the appropriate impl
:
#![allow(unused)] fn main() { #[derive(PartialEq, Clone)] struct Foo<T> { a: i32, b: T, } }
The generated impl
for PartialEq
is equivalent to
#![allow(unused)] fn main() { struct Foo<T> { a: i32, b: T } impl<T: PartialEq> PartialEq for Foo<T> { fn eq(&self, other: &Foo<T>) -> bool { self.a == other.a && self.b == other.b } } }
You can implement derive
for your own traits through procedural macros.
The automatically_derived
attribute
The automatically_derived
attribute is automatically added to
implementations created by the derive
attribute for built-in traits. It
has no direct effect, but it may be used by tools and diagnostic lints to
detect these automatically generated implementations.
Diagnostic attributes
The following attributes are used for controlling or generating diagnostic messages during compilation.
Lint check attributes
A lint check names a potentially undesirable coding pattern, such as
unreachable code or omitted documentation. The lint attributes allow
,
expect
, warn
, deny
, and forbid
use the MetaListPaths syntax
to specify a list of lint names to change the lint level for the entity
to which the attribute applies.
For any lint check C
:
#[allow(C)]
overrides the check forC
so that violations will go unreported.#[expect(C)]
indicates that lintC
is expected to be emitted. The attribute will suppress the emission ofC
or issue a warning, if the expectation is unfulfilled.#[warn(C)]
warns about violations ofC
but continues compilation.#[deny(C)]
signals an error after encountering a violation ofC
,#[forbid(C)]
is the same asdeny(C)
, but also forbids changing the lint level afterwards,
Note: The lint checks supported by
rustc
can be found viarustc -W help
, along with their default settings and are documented in the rustc book.
#![allow(unused)] fn main() { pub mod m1 { // Missing documentation is ignored here #[allow(missing_docs)] pub fn undocumented_one() -> i32 { 1 } // Missing documentation signals a warning here #[warn(missing_docs)] pub fn undocumented_too() -> i32 { 2 } // Missing documentation signals an error here #[deny(missing_docs)] pub fn undocumented_end() -> i32 { 3 } } }
Lint attributes can override the level specified from a previous attribute, as long as the level does not attempt to change a forbidden lint. Previous attributes are those from a higher level in the syntax tree, or from a previous attribute on the same entity as listed in left-to-right source order.
This example shows how one can use allow
and warn
to toggle a particular
check on and off:
#![allow(unused)] fn main() { #[warn(missing_docs)] pub mod m2 { #[allow(missing_docs)] pub mod nested { // Missing documentation is ignored here pub fn undocumented_one() -> i32 { 1 } // Missing documentation signals a warning here, // despite the allow above. #[warn(missing_docs)] pub fn undocumented_two() -> i32 { 2 } } // Missing documentation signals a warning here pub fn undocumented_too() -> i32 { 3 } } }
This example shows how one can use forbid
to disallow uses of allow
or
expect
for that lint check:
#![allow(unused)] fn main() { #[forbid(missing_docs)] pub mod m3 { // Attempting to toggle warning signals an error here #[allow(missing_docs)] /// Returns 2. pub fn undocumented_too() -> i32 { 2 } } }
Note:
rustc
allows setting lint levels on the command-line, and also supports setting caps on the lints that are reported.
Lint Reasons
All lint attributes support an additional reason
parameter, to give context why
a certain attribute was added. This reason will be displayed as part of the lint
message if the lint is emitted at the defined level.
#![allow(unused)] fn main() { // `keyword_idents` is allowed by default. Here we deny it to // avoid migration of identifiers when we update the edition. #![deny( keyword_idents, reason = "we want to avoid these idents to be future compatible" )] // This name was allowed in Rust's 2015 edition. We still aim to avoid // this to be future compatible and not confuse end users. fn dyn() {} }
Here is another example, where the lint is allowed with a reason:
#![allow(unused)] fn main() { use std::path::PathBuf; pub fn get_path() -> PathBuf { // The `reason` parameter on `allow` attributes acts as documentation for the reader. #[allow(unused_mut, reason = "this is only modified on some platforms")] let mut file_name = PathBuf::from("git"); #[cfg(target_os = "windows")] file_name.set_extension("exe"); file_name } }
The #[expect]
attribute
The #[expect(C)]
attribute creates a lint expectation for lint C
. The
expectation will be fulfilled, if a #[warn(C)]
attribute at the same location
would result in a lint emission. If the expectation is unfulfilled, because
lint C
would not be emitted, the unfulfilled_lint_expectations
lint will
be emitted at the attribute.
fn main() { // This `#[expect]` attribute creates a lint expectation, that the `unused_variables` // lint would be emitted by the following statement. This expectation is // unfulfilled, since the `question` variable is used by the `println!` macro. // Therefore, the `unfulfilled_lint_expectations` lint will be emitted at the // attribute. #[expect(unused_variables)] let question = "who lives in a pineapple under the sea?"; println!("{question}"); // This `#[expect]` attribute creates a lint expectation that will be fulfilled, since // the `answer` variable is never used. The `unused_variables` lint, that would usually // be emitted, is suppressed. No warning will be issued for the statement or attribute. #[expect(unused_variables)] let answer = "SpongeBob SquarePants!"; }
The lint expectation is only fulfilled by lint emissions which have been suppressed by
the expect
attribute. If the lint level is modified in the scope with other level
attributes like allow
or warn
, the lint emission will be handled accordingly and the
expectation will remain unfulfilled.
#![allow(unused)] fn main() { #[expect(unused_variables)] fn select_song() { // This will emit the `unused_variables` lint at the warn level // as defined by the `warn` attribute. This will not fulfill the // expectation above the function. #[warn(unused_variables)] let song_name = "Crab Rave"; // The `allow` attribute suppresses the lint emission. This will not // fulfill the expectation as it has been suppressed by the `allow` // attribute and not the `expect` attribute above the function. #[allow(unused_variables)] let song_creator = "Noisestorm"; // This `expect` attribute will suppress the `unused_variables` lint emission // at the variable. The `expect` attribute above the function will still not // be fulfilled, since this lint emission has been suppressed by the local // expect attribute. #[expect(unused_variables)] let song_version = "Monstercat Release"; } }
If the expect
attribute contains several lints, each one is expected separately. For a
lint group it’s enough if one lint inside the group has been emitted:
#![allow(unused)] fn main() { // This expectation will be fulfilled by the unused value inside the function // since the emitted `unused_variables` lint is inside the `unused` lint group. #[expect(unused)] pub fn thoughts() { let unused = "I'm running out of examples"; } pub fn another_example() { // This attribute creates two lint expectations. The `unused_mut` lint will be // suppressed and with that fulfill the first expectation. The `unused_variables` // wouldn't be emitted, since the variable is used. That expectation will therefore // be unsatisfied, and a warning will be emitted. #[expect(unused_mut, unused_variables)] let mut link = "https://www.rust-lang.org/"; println!("Welcome to our community: {link}"); } }
Note: The behavior of
#[expect(unfulfilled_lint_expectations)]
is currently defined to always generate theunfulfilled_lint_expectations
lint.
Lint groups
Lints may be organized into named groups so that the level of related lints can be adjusted together. Using a named group is equivalent to listing out the lints within that group.
#![allow(unused)] fn main() { // This allows all lints in the "unused" group. #[allow(unused)] // This overrides the "unused_must_use" lint from the "unused" // group to deny. #[deny(unused_must_use)] fn example() { // This does not generate a warning because the "unused_variables" // lint is in the "unused" group. let x = 1; // This generates an error because the result is unused and // "unused_must_use" is marked as "deny". std::fs::remove_file("some_file"); // ERROR: unused `Result` that must be used } }
There is a special group named “warnings” which includes all lints at the “warn” level. The “warnings” group ignores attribute order and applies to all lints that would otherwise warn within the entity.
#![allow(unused)] fn main() { unsafe fn an_unsafe_fn() {} // The order of these two attributes does not matter. #[deny(warnings)] // The unsafe_code lint is normally "allow" by default. #[warn(unsafe_code)] fn example_err() { // This is an error because the `unsafe_code` warning has // been lifted to "deny". unsafe { an_unsafe_fn() } // ERROR: usage of `unsafe` block } }
Tool lint attributes
Tool lints allows using scoped lints, to allow
, warn
, deny
or forbid
lints of certain tools.
Tool lints only get checked when the associated tool is active. If a lint
attribute, such as allow
, references a nonexistent tool lint, the compiler
will not warn about the nonexistent lint until you use the tool.
Otherwise, they work just like regular lint attributes:
// set the entire `pedantic` clippy lint group to warn #![warn(clippy::pedantic)] // silence warnings from the `filter_map` clippy lint #![allow(clippy::filter_map)] fn main() { // ... } // silence the `cmp_nan` clippy lint just for this function #[allow(clippy::cmp_nan)] fn foo() { // ... }
Note:
rustc
currently recognizes the tool lints for “clippy” and “rustdoc”.
The deprecated
attribute
The deprecated
attribute marks an item as deprecated. rustc
will issue
warnings on usage of #[deprecated]
items. rustdoc
will show item
deprecation, including the since
version and note
, if available.
The deprecated
attribute has several forms:
deprecated
— Issues a generic message.deprecated = "message"
— Includes the given string in the deprecation message.- MetaListNameValueStr syntax with two optional fields:
since
— Specifies a version number when the item was deprecated.rustc
does not currently interpret the string, but external tools like Clippy may check the validity of the value.note
— Specifies a string that should be included in the deprecation message. This is typically used to provide an explanation about the deprecation and preferred alternatives.
The deprecated
attribute may be applied to any item, trait item, enum
variant, struct field, external block item, or macro definition. It
cannot be applied to trait implementation items. When applied to an item
containing other items, such as a module or implementation, all child
items inherit the deprecation attribute.
Here is an example:
#![allow(unused)] fn main() { #[deprecated(since = "5.2.0", note = "foo was rarely used. Users should instead use bar")] pub fn foo() {} pub fn bar() {} }
The RFC contains motivations and more details.
The must_use
attribute
The must_use
attribute is used to issue a diagnostic warning when a value
is not “used”. It can be applied to user-defined composite types
(struct
s, enum
s, and union
s), functions,
and traits.
The must_use
attribute may include a message by using the
MetaNameValueStr syntax such as #[must_use = "example message"]
. The
message will be given alongside the warning.
When used on user-defined composite types, if the expression of an
expression statement has that type, then the unused_must_use
lint is
violated.
#![allow(unused)] fn main() { #[must_use] struct MustUse { // some fields } impl MustUse { fn new() -> MustUse { MustUse {} } } // Violates the `unused_must_use` lint. MustUse::new(); }
When used on a function, if the expression of an expression statement is a
call expression to that function, then the unused_must_use
lint is
violated.
#![allow(unused)] fn main() { #[must_use] fn five() -> i32 { 5i32 } // Violates the unused_must_use lint. five(); }
When used on a trait declaration, a call expression of an expression
statement to a function that returns an impl trait or a dyn trait of that trait violates
the unused_must_use
lint.
#![allow(unused)] fn main() { #[must_use] trait Critical {} impl Critical for i32 {} fn get_critical() -> impl Critical { 4i32 } // Violates the `unused_must_use` lint. get_critical(); }
When used on a function in a trait declaration, then the behavior also applies when the call expression is a function from an implementation of the trait.
#![allow(unused)] fn main() { trait Trait { #[must_use] fn use_me(&self) -> i32; } impl Trait for i32 { fn use_me(&self) -> i32 { 0i32 } } // Violates the `unused_must_use` lint. 5i32.use_me(); }
When used on a function in a trait implementation, the attribute does nothing.
Note: Trivial no-op expressions containing the value will not violate the lint. Examples include wrapping the value in a type that does not implement
Drop
and then not using that type and being the final expression of a block expression that is not used.#![allow(unused)] fn main() { #[must_use] fn five() -> i32 { 5i32 } // None of these violate the unused_must_use lint. (five(),); Some(five()); { five() }; if true { five() } else { 0i32 }; match true { _ => five() }; }
Note: It is idiomatic to use a let statement with a pattern of
_
when a must-used value is purposely discarded.#![allow(unused)] fn main() { #[must_use] fn five() -> i32 { 5i32 } // Does not violate the unused_must_use lint. let _ = five(); }
The diagnostic
tool attribute namespace
The #[diagnostic]
attribute namespace is a home for attributes to influence compile-time error messages.
The hints provided by these attributes are not guaranteed to be used.
Unknown attributes in this namespace are accepted, though they may emit warnings for unused attributes.
Additionally, invalid inputs to known attributes will typically be a warning (see the attribute definitions for details).
This is meant to allow adding or discarding attributes and changing inputs in the future to allow changes without the need to keep the non-meaningful attributes or options working.
The diagnostic::on_unimplemented
attribute
The #[diagnostic::on_unimplemented]
attribute is a hint to the compiler to supplement the error message that would normally be generated in scenarios where a trait is required but not implemented on a type.
The attribute should be placed on a trait declaration, though it is not an error to be located in other positions.
The attribute uses the MetaListNameValueStr syntax to specify its inputs, though any malformed input to the attribute is not considered as an error to provide both forwards and backwards compatibility.
The following keys have the given meaning:
message
— The text for the top level error message.label
— The text for the label shown inline in the broken code in the error message.note
— Provides additional notes.
The note
option can appear several times, which results in several note messages being emitted.
If any of the other options appears several times the first occurrence of the relevant option specifies the actually used value.
Any other occurrence generates an lint warning.
For any other non-existing option a lint-warning is generated.
All three options accept a string as an argument, interpreted using the same formatting as a std::fmt
string.
Format parameters with the given named parameter will be replaced with the following text:
{Self}
— The name of the type implementing the trait.{
GenericParameterName}
— The name of the generic argument’s type for the given generic parameter.
Any other format parameter will generate a warning, but will otherwise be included in the string as-is.
Invalid format strings may generate a warning, but are otherwise allowed, but may not display as intended. Format specifiers may generate a warning, but are otherwise ignored.
In this example:
#[diagnostic::on_unimplemented( message = "My Message for `ImportantTrait<{A}>` implemented for `{Self}`", label = "My Label", note = "Note 1", note = "Note 2" )] trait ImportantTrait<A> {} fn use_my_trait(_: impl ImportantTrait<i32>) {} fn main() { use_my_trait(String::new()); }
the compiler may generate an error message which looks like this:
error[E0277]: My Message for `ImportantTrait<i32>` implemented for `String`
--> src/main.rs:14:18
|
14 | use_my_trait(String::new());
| ------------ ^^^^^^^^^^^^^ My Label
| |
| required by a bound introduced by this call
|
= help: the trait `ImportantTrait<i32>` is not implemented for `String`
= note: Note 1
= note: Note 2
Code generation attributes
The following attributes are used for controlling code generation.
Optimization hints
The cold
and inline
attributes give suggestions to generate code in a
way that may be faster than what it would do without the hint. The attributes
are only hints, and may be ignored.
Both attributes can be used on functions. When applied to a function in a trait, they apply only to that function when used as a default function for a trait implementation and not to all trait implementations. The attributes have no effect on a trait function without a body.
The inline
attribute
The inline
attribute suggests that a copy of the attributed function
should be placed in the caller, rather than generating code to call the
function where it is defined.
Note: The
rustc
compiler automatically inlines functions based on internal heuristics. Incorrectly inlining functions can make the program slower, so this attribute should be used with care.
There are three ways to use the inline attribute:
#[inline]
suggests performing an inline expansion.#[inline(always)]
suggests that an inline expansion should always be performed.#[inline(never)]
suggests that an inline expansion should never be performed.
Note:
#[inline]
in every form is a hint, with no requirements on the language to place a copy of the attributed function in the caller.
The cold
attribute
The cold
attribute suggests that the attributed function is unlikely to
be called.
The no_builtins
attribute
The no_builtins
attribute may be applied at the crate level to disable
optimizing certain code patterns to invocations of library functions that are
assumed to exist.
The target_feature
attribute
The target_feature
attribute may be applied to a function to
enable code generation of that function for specific platform architecture
features. It uses the MetaListNameValueStr syntax with a single key of
enable
whose value is a string of comma-separated feature names to enable.
#![allow(unused)] fn main() { #[cfg(target_feature = "avx2")] #[target_feature(enable = "avx2")] unsafe fn foo_avx2() {} }
Each target architecture has a set of features that may be enabled. It is an error to specify a feature for a target architecture that the crate is not being compiled for.
It is undefined behavior to call a function that is compiled with a feature that is not supported on the current platform the code is running on, except if the platform explicitly documents this to be safe.
Functions marked with target_feature
are not inlined into a context that
does not support the given features. The #[inline(always)]
attribute may not
be used with a target_feature
attribute.
Available features
The following is a list of the available feature names.
x86
or x86_64
Executing code with unsupported features is undefined behavior on this platform.
Hence this platform requires that #[target_feature]
is only applied to unsafe
functions.
Feature | Implicitly Enables | Description |
---|---|---|
adx | ADX — Multi-Precision Add-Carry Instruction Extensions | |
aes | sse2 | AES — Advanced Encryption Standard |
avx | sse4.2 | AVX — Advanced Vector Extensions |
avx2 | avx | AVX2 — Advanced Vector Extensions 2 |
bmi1 | BMI1 — Bit Manipulation Instruction Sets | |
bmi2 | BMI2 — Bit Manipulation Instruction Sets 2 | |
cmpxchg16b | cmpxchg16b — Compares and exchange 16 bytes (128 bits) of data atomically | |
f16c | avx | F16C — 16-bit floating point conversion instructions |
fma | avx | FMA3 — Three-operand fused multiply-add |
fxsr | fxsave and fxrstor — Save and restore x87 FPU, MMX Technology, and SSE State | |
lzcnt | lzcnt — Leading zeros count | |
movbe | movbe — Move data after swapping bytes | |
pclmulqdq | sse2 | pclmulqdq — Packed carry-less multiplication quadword |
popcnt | popcnt — Count of bits set to 1 | |
rdrand | rdrand — Read random number | |
rdseed | rdseed — Read random seed | |
sha | sse2 | SHA — Secure Hash Algorithm |
sse | SSE — Streaming SIMD Extensions | |
sse2 | sse | SSE2 — Streaming SIMD Extensions 2 |
sse3 | sse2 | SSE3 — Streaming SIMD Extensions 3 |
sse4.1 | ssse3 | SSE4.1 — Streaming SIMD Extensions 4.1 |
sse4.2 | sse4.1 | SSE4.2 — Streaming SIMD Extensions 4.2 |
ssse3 | sse3 | SSSE3 — Supplemental Streaming SIMD Extensions 3 |
xsave | xsave — Save processor extended states | |
xsavec | xsavec — Save processor extended states with compaction | |
xsaveopt | xsaveopt — Save processor extended states optimized | |
xsaves | xsaves — Save processor extended states supervisor |
aarch64
This platform requires that #[target_feature]
is only applied to unsafe
functions.
Further documentation on these features can be found in the ARM Architecture Reference Manual, or elsewhere on developer.arm.com.
Note: The following pairs of features should both be marked as enabled or disabled together if used:
paca
andpacg
, which LLVM currently implements as one feature.
Feature | Implicitly Enables | Feature Name |
---|---|---|
aes | neon | FEAT_AES & FEAT_PMULL — Advanced SIMD AES & PMULL instructions |
bf16 | FEAT_BF16 — BFloat16 instructions | |
bti | FEAT_BTI — Branch Target Identification | |
crc | FEAT_CRC — CRC32 checksum instructions | |
dit | FEAT_DIT — Data Independent Timing instructions | |
dotprod | FEAT_DotProd — Advanced SIMD Int8 dot product instructions | |
dpb | FEAT_DPB — Data cache clean to point of persistence | |
dpb2 | FEAT_DPB2 — Data cache clean to point of deep persistence | |
f32mm | sve | FEAT_F32MM — SVE single-precision FP matrix multiply instruction |
f64mm | sve | FEAT_F64MM — SVE double-precision FP matrix multiply instruction |
fcma | neon | FEAT_FCMA — Floating point complex number support |
fhm | fp16 | FEAT_FHM — Half-precision FP FMLAL instructions |
flagm | FEAT_FlagM — Conditional flag manipulation | |
fp16 | neon | FEAT_FP16 — Half-precision FP data processing |
frintts | FEAT_FRINTTS — Floating-point to int helper instructions | |
i8mm | FEAT_I8MM — Int8 Matrix Multiplication | |
jsconv | neon | FEAT_JSCVT — JavaScript conversion instruction |
lse | FEAT_LSE — Large System Extension | |
lor | FEAT_LOR — Limited Ordering Regions extension | |
mte | FEAT_MTE & FEAT_MTE2 — Memory Tagging Extension | |
neon | FEAT_FP & FEAT_AdvSIMD — Floating Point and Advanced SIMD extension | |
pan | FEAT_PAN — Privileged Access-Never extension | |
paca | FEAT_PAuth — Pointer Authentication (address authentication) | |
pacg | FEAT_PAuth — Pointer Authentication (generic authentication) | |
pmuv3 | FEAT_PMUv3 — Performance Monitors extension (v3) | |
rand | FEAT_RNG — Random Number Generator | |
ras | FEAT_RAS & FEAT_RASv1p1 — Reliability, Availability and Serviceability extension | |
rcpc | FEAT_LRCPC — Release consistent Processor Consistent | |
rcpc2 | rcpc | FEAT_LRCPC2 — RcPc with immediate offsets |
rdm | FEAT_RDM — Rounding Double Multiply accumulate | |
sb | FEAT_SB — Speculation Barrier | |
sha2 | neon | FEAT_SHA1 & FEAT_SHA256 — Advanced SIMD SHA instructions |
sha3 | sha2 | FEAT_SHA512 & FEAT_SHA3 — Advanced SIMD SHA instructions |
sm4 | neon | FEAT_SM3 & FEAT_SM4 — Advanced SIMD SM3/4 instructions |
spe | FEAT_SPE — Statistical Profiling Extension | |
ssbs | FEAT_SSBS & FEAT_SSBS2 — Speculative Store Bypass Safe | |
sve | fp16 | FEAT_SVE — Scalable Vector Extension |
sve2 | sve | FEAT_SVE2 — Scalable Vector Extension 2 |
sve2-aes | sve2 , aes | FEAT_SVE_AES — SVE AES instructions |
sve2-sm4 | sve2 , sm4 | FEAT_SVE_SM4 — SVE SM4 instructions |
sve2-sha3 | sve2 , sha3 | FEAT_SVE_SHA3 — SVE SHA3 instructions |
sve2-bitperm | sve2 | FEAT_SVE_BitPerm — SVE Bit Permute |
tme | FEAT_TME — Transactional Memory Extension | |
vh | FEAT_VHE — Virtualization Host Extensions |
riscv32
or riscv64
This platform requires that #[target_feature]
is only applied to unsafe
functions.
Further documentation on these features can be found in their respective specification. Many specifications are described in the RISC-V ISA Manual or in another manual hosted on the RISC-V GitHub Account.
Feature | Implicitly Enables | Description |
---|---|---|
a | A — Atomic instructions | |
c | C — Compressed instructions | |
m | M — Integer Multiplication and Division instructions | |
zb | zba , zbc , zbs | Zb — Bit Manipulation instructions |
zba | Zba — Address Generation instructions | |
zbb | Zbb — Basic bit-manipulation | |
zbc | Zbc — Carry-less multiplication | |
zbkb | Zbkb — Bit Manipulation Instructions for Cryptography | |
zbkc | Zbkc — Carry-less multiplication for Cryptography | |
zbkx | Zbkx — Crossbar permutations | |
zbs | Zbs — Single-bit instructions | |
zk | zkn , zkr , zks , zkt , zbkb , zbkc , zkbx | Zk — Scalar Cryptography |
zkn | zknd , zkne , zknh , zbkb , zbkc , zkbx | Zkn — NIST Algorithm suite extension |
zknd | Zknd — NIST Suite: AES Decryption | |
zkne | Zkne — NIST Suite: AES Encryption | |
zknh | Zknh — NIST Suite: Hash Function Instructions | |
zkr | Zkr — Entropy Source Extension | |
zks | zksed , zksh , zbkb , zbkc , zkbx | Zks — ShangMi Algorithm Suite |
zksed | Zksed — ShangMi Suite: SM4 Block Cipher Instructions | |
zksh | Zksh — ShangMi Suite: SM3 Hash Function Instructions | |
zkt | Zkt — Data Independent Execution Latency Subset |
wasm32
or wasm64
#[target_feature]
may be used with both safe and
unsafe
functions on Wasm platforms. It is impossible to
cause undefined behavior via the #[target_feature]
attribute because
attempting to use instructions unsupported by the Wasm engine will fail at load
time without the risk of being interpreted in a way different from what the
compiler expected.
Feature | Implicitly Enables | Description |
---|---|---|
bulk-memory | WebAssembly bulk memory operations proposal | |
extended-const | WebAssembly extended const expressions proposal | |
mutable-globals | WebAssembly mutable global proposal | |
nontrapping-fptoint | WebAssembly non-trapping float-to-int conversion proposal | |
relaxed-simd | simd128 | WebAssembly relaxed simd proposal |
sign-ext | WebAssembly sign extension operators Proposal | |
simd128 | WebAssembly simd proposal |
Additional information
See the target_feature
conditional compilation option for selectively
enabling or disabling compilation of code based on compile-time settings. Note
that this option is not affected by the target_feature
attribute, and is
only driven by the features enabled for the entire crate.
See the is_x86_feature_detected
or is_aarch64_feature_detected
macros
in the standard library for runtime feature detection on these platforms.
Note:
rustc
has a default set of features enabled for each target and CPU. The CPU may be chosen with the-C target-cpu
flag. Individual features may be enabled or disabled for an entire crate with the-C target-feature
flag.
The track_caller
attribute
The track_caller
attribute may be applied to any function with "Rust"
ABI
with the exception of the entry point fn main
. When applied to functions and methods in
trait declarations, the attribute applies to all implementations. If the trait provides a
default implementation with the attribute, then the attribute also applies to override implementations.
When applied to a function in an extern
block the attribute must also be applied to any linked
implementations, otherwise undefined behavior results. When applied to a function which is made
available to an extern
block, the declaration in the extern
block must also have the attribute,
otherwise undefined behavior results.
Behavior
Applying the attribute to a function f
allows code within f
to get a hint of the Location
of
the “topmost” tracked call that led to f
’s invocation. At the point of observation, an
implementation behaves as if it walks up the stack from f
’s frame to find the nearest frame of an
unattributed function outer
, and it returns the Location
of the tracked call in outer
.
#![allow(unused)] fn main() { #[track_caller] fn f() { println!("{}", std::panic::Location::caller()); } }
Note:
core
providescore::panic::Location::caller
for observing caller locations. It wraps thecore::intrinsics::caller_location
intrinsic implemented byrustc
.
Note: because the resulting
Location
is a hint, an implementation may halt its walk up the stack early. See Limitations for important caveats.
Examples
When f
is called directly by calls_f
, code in f
observes its callsite within calls_f
:
#![allow(unused)] fn main() { #[track_caller] fn f() { println!("{}", std::panic::Location::caller()); } fn calls_f() { f(); // <-- f() prints this location } }
When f
is called by another attributed function g
which is in turn called by calls_g
, code in
both f
and g
observes g
’s callsite within calls_g
:
#![allow(unused)] fn main() { #[track_caller] fn f() { println!("{}", std::panic::Location::caller()); } #[track_caller] fn g() { println!("{}", std::panic::Location::caller()); f(); } fn calls_g() { g(); // <-- g() prints this location twice, once itself and once from f() } }
When g
is called by another attributed function h
which is in turn called by calls_h
, all code
in f
, g
, and h
observes h
’s callsite within calls_h
:
#![allow(unused)] fn main() { #[track_caller] fn f() { println!("{}", std::panic::Location::caller()); } #[track_caller] fn g() { println!("{}", std::panic::Location::caller()); f(); } #[track_caller] fn h() { println!("{}", std::panic::Location::caller()); g(); } fn calls_h() { h(); // <-- prints this location three times, once itself, once from g(), once from f() } }
And so on.
Limitations
This information is a hint and implementations are not required to preserve it.
In particular, coercing a function with #[track_caller]
to a function pointer creates a shim which
appears to observers to have been called at the attributed function’s definition site, losing actual
caller information across virtual calls. A common example of this coercion is the creation of a
trait object whose methods are attributed.
Note: The aforementioned shim for function pointers is necessary because
rustc
implementstrack_caller
in a codegen context by appending an implicit parameter to the function ABI, but this would be unsound for an indirect call because the parameter is not a part of the function’s type and a given function pointer type may or may not refer to a function with the attribute. The creation of a shim hides the implicit parameter from callers of the function pointer, preserving soundness.
The instruction_set
attribute
The instruction_set
attribute may be applied to a function to control which instruction set the function will be generated for.
This allows mixing more than one instruction set in a single program on CPU architectures that support it.
It uses the MetaListPath syntax, and a path comprised of the architecture family name and instruction set name.
It is a compilation error to use the instruction_set
attribute on a target that does not support it.
On ARM
For the ARMv4T
and ARMv5te
architectures, the following are supported:
arm::a32
— Generate the function as A32 “ARM” code.arm::t32
— Generate the function as T32 “Thumb” code.
#[instruction_set(arm::a32)]
fn foo_arm_code() {}
#[instruction_set(arm::t32)]
fn bar_thumb_code() {}
Using the instruction_set
attribute has the following effects:
- If the address of the function is taken as a function pointer, the low bit of the address will be set to 0 (arm) or 1 (thumb) depending on the instruction set.
- Any inline assembly in the function must use the specified instruction set instead of the target default.
Limits
The following attributes affect compile-time limits.
The recursion_limit
attribute
The recursion_limit
attribute may be applied at the crate level to set the
maximum depth for potentially infinitely-recursive compile-time operations
like macro expansion or auto-dereference. It uses the MetaNameValueStr
syntax to specify the recursion depth.
Note: The default in
rustc
is 128.
#![allow(unused)] #![recursion_limit = "4"] fn main() { macro_rules! a { () => { a!(1); }; (1) => { a!(2); }; (2) => { a!(3); }; (3) => { a!(4); }; (4) => { }; } // This fails to expand because it requires a recursion depth greater than 4. a!{} }
#![allow(unused)] #![recursion_limit = "1"] fn main() { // This fails because it requires two recursive steps to auto-dereference. (|_: &u8| {})(&&&1); }
The type_length_limit
attribute
Note: This limit is only enforced when the nightly
-Zenforce-type-length-limit
flag is active.For more information, see https://github.com/rust-lang/rust/pull/127670.
The type_length_limit
attribute limits the maximum number of type
substitutions made when constructing a concrete type during monomorphization.
It is applied at the crate level, and uses the MetaNameValueStr syntax
to set the limit based on the number of type substitutions.
Note: The default in
rustc
is 1048576.
#![type_length_limit = "4"]
fn f<T>(x: T) {}
// This fails to compile because monomorphizing to
// `f::<((((i32,), i32), i32), i32)>` requires more than 4 type elements.
f(((((1,), 2), 3), 4));
Type system attributes
The following attributes are used for changing how a type can be used.
The non_exhaustive
attribute
The non_exhaustive
attribute indicates that a type or variant may have
more fields or variants added in the future. It can be applied to
struct
s, enum
s, and enum
variants.
The non_exhaustive
attribute uses the MetaWord syntax and thus does not
take any inputs.
Within the defining crate, non_exhaustive
has no effect.
#![allow(unused)] fn main() { #[non_exhaustive] pub struct Config { pub window_width: u16, pub window_height: u16, } #[non_exhaustive] pub struct Token; #[non_exhaustive] pub struct Id(pub u64); #[non_exhaustive] pub enum Error { Message(String), Other, } pub enum Message { #[non_exhaustive] Send { from: u32, to: u32, contents: String }, #[non_exhaustive] Reaction(u32), #[non_exhaustive] Quit, } // Non-exhaustive structs can be constructed as normal within the defining crate. let config = Config { window_width: 640, window_height: 480 }; let token = Token; let id = Id(4); // Non-exhaustive structs can be matched on exhaustively within the defining crate. let Config { window_width, window_height } = config; let Token = token; let Id(id_number) = id; let error = Error::Other; let message = Message::Reaction(3); // Non-exhaustive enums can be matched on exhaustively within the defining crate. match error { Error::Message(ref s) => { }, Error::Other => { }, } match message { // Non-exhaustive variants can be matched on exhaustively within the defining crate. Message::Send { from, to, contents } => { }, Message::Reaction(id) => { }, Message::Quit => { }, } }
Outside of the defining crate, types annotated with non_exhaustive
have limitations that
preserve backwards compatibility when new fields or variants are added.
Non-exhaustive types cannot be constructed outside of the defining crate:
- Non-exhaustive variants (
struct
orenum
variant) cannot be constructed with a StructExpression (including with functional update syntax). - The implicitly defined same-named constant of a unit-like struct,
or the same-named constructor function of a tuple struct,
has a visibility no greater than
pub(crate)
. That is, if the struct’s visibility ispub
, then the constant or constructor’s visibility ispub(crate)
, and otherwise the visibility of the two items is the same (as is the case without#[non_exhaustive]
). enum
instances can be constructed.
The following examples of construction do not compile when outside the defining crate:
// These are types defined in an upstream crate that have been annotated as
// `#[non_exhaustive]`.
use upstream::{Config, Token, Id, Error, Message};
// Cannot construct an instance of `Config`; if new fields were added in
// a new version of `upstream` then this would fail to compile, so it is
// disallowed.
let config = Config { window_width: 640, window_height: 480 };
// Cannot construct an instance of `Token`; if new fields were added, then
// it would not be a unit-like struct any more, so the same-named constant
// created by it being a unit-like struct is not public outside the crate;
// this code fails to compile.
let token = Token;
// Cannot construct an instance of `Id`; if new fields were added, then
// its constructor function signature would change, so its constructor
// function is not public outside the crate; this code fails to compile.
let id = Id(5);
// Can construct an instance of `Error`; new variants being introduced would
// not result in this failing to compile.
let error = Error::Message("foo".to_string());
// Cannot construct an instance of `Message::Send` or `Message::Reaction`;
// if new fields were added in a new version of `upstream` then this would
// fail to compile, so it is disallowed.
let message = Message::Send { from: 0, to: 1, contents: "foo".to_string(), };
let message = Message::Reaction(0);
// Cannot construct an instance of `Message::Quit`; if this were converted to
// a tuple-variant `upstream` then this would fail to compile.
let message = Message::Quit;
There are limitations when matching on non-exhaustive types outside of the defining crate:
- When pattern matching on a non-exhaustive variant (
struct
orenum
variant), a StructPattern must be used which must include a..
. A tuple variant’s constructor’s visibility is reduced to be no greater thanpub(crate)
. - When pattern matching on a non-exhaustive
enum
, matching on a variant does not contribute towards the exhaustiveness of the arms.
The following examples of matching do not compile when outside the defining crate:
// These are types defined in an upstream crate that have been annotated as
// `#[non_exhaustive]`.
use upstream::{Config, Token, Id, Error, Message};
// Cannot match on a non-exhaustive enum without including a wildcard arm.
match error {
Error::Message(ref s) => { },
Error::Other => { },
// would compile with: `_ => {},`
}
// Cannot match on a non-exhaustive struct without a wildcard.
if let Ok(Config { window_width, window_height }) = config {
// would compile with: `..`
}
// Cannot match a non-exhaustive unit-like or tuple struct except by using
// braced struct syntax with a wildcard.
// This would compile as `let Token { .. } = token;`
let Token = token;
// This would compile as `let Id { 0: id_number, .. } = id;`
let Id(id_number) = id;
match message {
// Cannot match on a non-exhaustive struct enum variant without including a wildcard.
Message::Send { from, to, contents } => { },
// Cannot match on a non-exhaustive tuple or unit enum variant.
Message::Reaction(type) => { },
Message::Quit => { },
}
It’s also not allowed to cast non-exhaustive types from foreign crates.
use othercrate::NonExhaustiveEnum;
// Cannot cast a non-exhaustive enum outside of its defining crate.
let _ = NonExhaustiveEnum::default() as u8;
Non-exhaustive types are always considered inhabited in downstream crates.
Debugger attributes
The following attributes are used for enhancing the debugging experience when using third-party debuggers like GDB or WinDbg.
The debugger_visualizer
attribute
The debugger_visualizer
attribute can be used to embed a debugger visualizer file into the debug information.
This enables an improved debugger experience for displaying values in the debugger.
It uses the MetaListNameValueStr syntax to specify its inputs, and must be specified as a crate attribute.
Using debugger_visualizer
with Natvis
Natvis is an XML-based framework for Microsoft debuggers (such as Visual Studio and WinDbg) that uses declarative rules to customize the display of types. For detailed information on the Natvis format, refer to Microsoft’s Natvis documentation.
This attribute only supports embedding Natvis files on -windows-msvc
targets.
The path to the Natvis file is specified with the natvis_file
key, which is a path relative to the crate source file:
#![debugger_visualizer(natvis_file = "Rectangle.natvis")]
struct FancyRect {
x: f32,
y: f32,
dx: f32,
dy: f32,
}
fn main() {
let fancy_rect = FancyRect { x: 10.0, y: 10.0, dx: 5.0, dy: 5.0 };
println!("set breakpoint here");
}
and Rectangle.natvis
contains:
<?xml version="1.0" encoding="utf-8"?>
<AutoVisualizer xmlns="http://schemas.microsoft.com/vstudio/debugger/natvis/2010">
<Type Name="foo::FancyRect">
<DisplayString>({x},{y}) + ({dx}, {dy})</DisplayString>
<Expand>
<Synthetic Name="LowerLeft">
<DisplayString>({x}, {y})</DisplayString>
</Synthetic>
<Synthetic Name="UpperLeft">
<DisplayString>({x}, {y + dy})</DisplayString>
</Synthetic>
<Synthetic Name="UpperRight">
<DisplayString>({x + dx}, {y + dy})</DisplayString>
</Synthetic>
<Synthetic Name="LowerRight">
<DisplayString>({x + dx}, {y})</DisplayString>
</Synthetic>
</Expand>
</Type>
</AutoVisualizer>
When viewed under WinDbg, the fancy_rect
variable would be shown as follows:
> Variables:
> fancy_rect: (10.0, 10.0) + (5.0, 5.0)
> LowerLeft: (10.0, 10.0)
> UpperLeft: (10.0, 15.0)
> UpperRight: (15.0, 15.0)
> LowerRight: (15.0, 10.0)
Using debugger_visualizer
with GDB
GDB supports the use of a structured Python script, called a pretty printer, that describes how a type should be visualized in the debugger view. For detailed information on pretty printers, refer to GDB’s pretty printing documentation.
Embedded pretty printers are not automatically loaded when debugging a binary under GDB. There are two ways to enable auto-loading embedded pretty printers:
- Launch GDB with extra arguments to explicitly add a directory or binary to the auto-load safe path:
gdb -iex "add-auto-load-safe-path safe-path path/to/binary" path/to/binary
For more information, see GDB’s auto-loading documentation. - Create a file named
gdbinit
under$HOME/.config/gdb
(you may need to create the directory if it doesn’t already exist). Add the following line to that file:add-auto-load-safe-path path/to/binary
.
These scripts are embedded using the gdb_script_file
key, which is a path relative to the crate source file.
#![debugger_visualizer(gdb_script_file = "printer.py")]
struct Person {
name: String,
age: i32,
}
fn main() {
let bob = Person { name: String::from("Bob"), age: 10 };
println!("set breakpoint here");
}
and printer.py
contains:
import gdb
class PersonPrinter:
"Print a Person"
def __init__(self, val):
self.val = val
self.name = val["name"]
self.age = int(val["age"])
def to_string(self):
return "{} is {} years old.".format(self.name, self.age)
def lookup(val):
lookup_tag = val.type.tag
if lookup_tag is None:
return None
if "foo::Person" == lookup_tag:
return PersonPrinter(val)
return None
gdb.current_objfile().pretty_printers.append(lookup)
When the crate’s debug executable is passed into GDB1, print bob
will display:
"Bob" is 10 years old.
Note: This assumes you are using the rust-gdb
script which configures pretty-printers for standard library types like String
.
The collapse_debuginfo
attribute
The collapse_debuginfo
attribute controls whether code locations from a macro definition are collapsed into a single location associated with the macro’s call site,
when generating debuginfo for code calling this macro.
The attribute uses the MetaListIdents syntax to specify its inputs, and can only be applied to macro definitions.
Accepted options:
#[collapse_debuginfo(yes)]
— code locations in debuginfo are collapsed.#[collapse_debuginfo(no)]
— code locations in debuginfo are not collapsed.#[collapse_debuginfo(external)]
— code locations in debuginfo are collapsed only if the macro comes from a different crate.
The external
behavior is the default for macros that don’t have this attribute, unless they are built-in macros.
For built-in macros the default is yes
.
Note:
rustc
has a-C collapse-macro-debuginfo
CLI option to override both the default collapsing behavior and#[collapse_debuginfo]
attributes.
#![allow(unused)] fn main() { #[collapse_debuginfo(yes)] macro_rules! example { () => { println!("hello!"); }; } }
Statements and expressions
Rust is primarily an expression language. This means that most forms of value-producing or effect-causing evaluation are directed by the uniform syntax category of expressions. Each kind of expression can typically nest within each other kind of expression, and rules for evaluation of expressions involve specifying both the value produced by the expression and the order in which its sub-expressions are themselves evaluated.
In contrast, statements serve mostly to contain and explicitly sequence expression evaluation.
Statements
Syntax
Statement :
;
| Item
| LetStatement
| ExpressionStatement
| MacroInvocationSemi
A statement is a component of a block, which is in turn a component of an outer expression or function.
Rust has two kinds of statement: declaration statements and expression statements.
Declaration statements
A declaration statement is one that introduces one or more names into the enclosing statement block. The declared names may denote new variables or new items.
The two kinds of declaration statements are item declarations and let
statements.
Item declarations
An item declaration statement has a syntactic form identical to an item declaration within a module. Declaring an item within a statement block restricts its scope to the block containing the statement. The item is not given a canonical path nor are any sub-items it may declare. The exception to this is that associated items defined by implementations are still accessible in outer scopes as long as the item and, if applicable, trait are accessible. It is otherwise identical in meaning to declaring the item inside a module.
There is no implicit capture of the containing function’s generic parameters, parameters, and local variables.
For example, inner
may not access outer_var
.
#![allow(unused)] fn main() { fn outer() { let outer_var = true; fn inner() { /* outer_var is not in scope here */ } inner(); } }
let
statements
Syntax
LetStatement :
OuterAttribute*let
PatternNoTopAlt (:
Type )? (=
Expression † (else
BlockExpression) ? ) ?;
† When an
else
block is specified, the Expression must not be a LazyBooleanExpression, or end with a}
.
A let
statement introduces a new set of variables, given by a pattern.
The pattern is followed optionally by a type annotation and then either ends, or is followed by an initializer expression plus an optional else
block.
When no type annotation is given, the compiler will infer the type, or signal an error if insufficient type information is available for definite inference.
Any variables introduced by a variable declaration are visible from the point of declaration until the end of the enclosing block scope, except when they are shadowed by another variable declaration.
If an else
block is not present, the pattern must be irrefutable.
If an else
block is present, the pattern may be refutable.
If the pattern does not match (this requires it to be refutable), the else
block is executed.
The else
block must always diverge (evaluate to the never type).
#![allow(unused)] fn main() { let (mut v, w) = (vec![1, 2, 3], 42); // The bindings may be mut or const let Some(t) = v.pop() else { // Refutable patterns require an else block panic!(); // The else block must diverge }; let [u, v] = [v[0], v[1]] else { // This pattern is irrefutable, so the compiler // will lint as the else block is redundant. panic!(); }; }
Expression statements
Syntax
ExpressionStatement :
ExpressionWithoutBlock;
| ExpressionWithBlock;
?
An expression statement is one that evaluates an expression and ignores its result. As a rule, an expression statement’s purpose is to trigger the effects of evaluating its expression.
An expression that consists of only a block expression or control flow expression, if used in a context where a statement is permitted, can omit the trailing semicolon. This can cause an ambiguity between it being parsed as a standalone statement and as a part of another expression; in this case, it is parsed as a statement. The type of ExpressionWithBlock expressions when used as statements must be the unit type.
#![allow(unused)] fn main() { let mut v = vec![1, 2, 3]; v.pop(); // Ignore the element returned from pop if v.is_empty() { v.push(5); } else { v.remove(0); } // Semicolon can be omitted. [1]; // Separate expression statement, not an indexing expression. }
When the trailing semicolon is omitted, the result must be type ()
.
#![allow(unused)] fn main() { // bad: the block's type is i32, not () // Error: expected `()` because of default return type // if true { // 1 // } // good: the block's type is i32 if true { 1 } else { 2 }; }
Attributes on Statements
Statements accept outer attributes.
The attributes that have meaning on a statement are cfg
, and the lint check attributes.
Expressions
Syntax
Expression :
ExpressionWithoutBlock
| ExpressionWithBlockExpressionWithoutBlock :
OuterAttribute*†
(
LiteralExpression
| PathExpression
| OperatorExpression
| GroupedExpression
| ArrayExpression
| AwaitExpression
| IndexExpression
| TupleExpression
| TupleIndexingExpression
| StructExpression
| CallExpression
| MethodCallExpression
| FieldExpression
| ClosureExpression
| AsyncBlockExpression
| ContinueExpression
| BreakExpression
| RangeExpression
| ReturnExpression
| UnderscoreExpression
| MacroInvocation
)ExpressionWithBlock :
OuterAttribute*†
(
BlockExpression
| ConstBlockExpression
| UnsafeBlockExpression
| LoopExpression
| IfExpression
| IfLetExpression
| MatchExpression
)
An expression may have two roles: it always produces a value, and it may have effects (otherwise known as “side effects”). An expression evaluates to a value, and has effects during evaluation. Many expressions contain sub-expressions, called the operands of the expression. The meaning of each kind of expression dictates several things:
- Whether or not to evaluate the operands when evaluating the expression
- The order in which to evaluate the operands
- How to combine the operands’ values to obtain the value of the expression
In this way, the structure of expressions dictates the structure of execution. Blocks are just another kind of expression, so blocks, statements, expressions, and blocks again can recursively nest inside each other to an arbitrary depth.
Note: We give names to the operands of expressions so that we may discuss them, but these names are not stable and may be changed.
Expression precedence
The precedence of Rust operators and expressions is ordered as follows, going from strong to weak. Binary Operators at the same precedence level are grouped in the order given by their associativity.
Operator/Expression | Associativity |
---|---|
Paths | |
Method calls | |
Field expressions | left to right |
Function calls, array indexing | |
? | |
Unary - * ! & &mut | |
as | left to right |
* / % | left to right |
+ - | left to right |
<< >> | left to right |
& | left to right |
^ | left to right |
| | left to right |
== != < > <= >= | Require parentheses |
&& | left to right |
|| | left to right |
.. ..= | Require parentheses |
= += -= *= /= %= &= |= ^= <<= >>= | right to left |
return break closures |
Evaluation order of operands
The following list of expressions all evaluate their operands the same way, as described after the list. Other expressions either don’t take operands or evaluate them conditionally as described on their respective pages.
- Dereference expression
- Error propagation expression
- Negation expression
- Arithmetic and logical binary operators
- Comparison operators
- Type cast expression
- Grouped expression
- Array expression
- Await expression
- Index expression
- Tuple expression
- Tuple index expression
- Struct expression
- Call expression
- Method call expression
- Field expression
- Break expression
- Range expression
- Return expression
The operands of these expressions are evaluated prior to applying the effects of the expression. Expressions taking multiple operands are evaluated left to right as written in the source code.
Note: Which subexpressions are the operands of an expression is determined by expression precedence as per the previous section.
For example, the two next
method calls will always be called in the same order:
#![allow(unused)] fn main() { // Using vec instead of array to avoid references // since there is no stable owned array iterator // at the time this example was written. let mut one_two = vec![1, 2].into_iter(); assert_eq!( (1, 2), (one_two.next().unwrap(), one_two.next().unwrap()) ); }
Note: Since this is applied recursively, these expressions are also evaluated from innermost to outermost, ignoring siblings until there are no inner subexpressions.
Place Expressions and Value Expressions
Expressions are divided into two main categories: place expressions and value expressions; there is also a third, minor category of expressions called assignee expressions. Within each expression, operands may likewise occur in either place context or value context. The evaluation of an expression depends both on its own category and the context it occurs within.
A place expression is an expression that represents a memory location.
These expressions are paths which refer to local variables, static variables, dereferences (*expr
), array indexing expressions (expr[expr]
), field references (expr.f
) and parenthesized place expressions.
All other expressions are value expressions.
A value expression is an expression that represents an actual value.
The following contexts are place expression contexts:
- The left operand of a compound assignment expression.
- The operand of a unary borrow, raw borrow or dereference operator.
- The operand of a field expression.
- The indexed operand of an array indexing expression.
- The operand of any implicit borrow.
- The initializer of a let statement.
- The scrutinee of an
if let
,match
, orwhile let
expression. - The base of a functional update struct expression.
Note: Historically, place expressions were called lvalues and value expressions were called rvalues.
An assignee expression is an expression that appears in the left operand of an assignment expression. Explicitly, the assignee expressions are:
- Place expressions.
- Underscores.
- Tuples of assignee expressions.
- Slices of assignee expressions.
- Tuple structs of assignee expressions.
- Structs of assignee expressions (with optionally named fields).
- Unit structs.
Arbitrary parenthesisation is permitted inside assignee expressions.
Moved and copied types
When a place expression is evaluated in a value expression context, or is bound by value in a pattern, it denotes the value held in that memory location.
If the type of that value implements Copy
, then the value will be copied.
In the remaining situations, if that type is Sized
, then it may be possible to move the value.
Only the following place expressions may be moved out of:
- Variables which are not currently borrowed.
- Temporary values.
- Fields of a place expression which can be moved out of and don’t implement
Drop
. - The result of dereferencing an expression with type
Box<T>
and that can also be moved out of.
After moving out of a place expression that evaluates to a local variable, the location is deinitialized and cannot be read from again until it is reinitialized. In all other cases, trying to use a place expression in a value expression context is an error.
Mutability
For a place expression to be assigned to, mutably borrowed, implicitly mutably borrowed, or bound to a pattern containing ref mut
, it must be mutable.
We call these mutable place expressions.
In contrast, other place expressions are called immutable place expressions.
The following expressions can be mutable place expression contexts:
- Mutable variables which are not currently borrowed.
- Mutable
static
items. - Temporary values.
- Fields: this evaluates the subexpression in a mutable place expression context.
- Dereferences of a
*mut T
pointer. - Dereference of a variable, or field of a variable, with type
&mut T
. Note: This is an exception to the requirement of the next rule. - Dereferences of a type that implements
DerefMut
: this then requires that the value being dereferenced is evaluated in a mutable place expression context. - Array indexing of a type that implements
IndexMut
: this then evaluates the value being indexed, but not the index, in mutable place expression context.
Temporaries
When using a value expression in most place expression contexts, a temporary unnamed memory location is created and initialized to that value.
The expression evaluates to that location instead, except if promoted to a static
.
The drop scope of the temporary is usually the end of the enclosing statement.
Implicit Borrows
Certain expressions will treat an expression as a place expression by implicitly borrowing it.
For example, it is possible to compare two unsized slices for equality directly, because the ==
operator implicitly borrows its operands:
#![allow(unused)] fn main() { let c = [1, 2, 3]; let d = vec![1, 2, 3]; let a: &[i32]; let b: &[i32]; a = &c; b = &d; // ... *a == *b; // Equivalent form: ::std::cmp::PartialEq::eq(&*a, &*b); }
Implicit borrows may be taken in the following expressions:
- Left operand in method-call expressions.
- Left operand in field expressions.
- Left operand in call expressions.
- Left operand in array indexing expressions.
- Operand of the dereference operator (
*
). - Operands of comparison.
- Left operands of the compound assignment.
Overloading Traits
Many of the following operators and expressions can also be overloaded for other types using traits in std::ops
or std::cmp
.
These traits also exist in core::ops
and core::cmp
with the same names.
Expression Attributes
Outer attributes before an expression are allowed only in a few specific cases:
- Before an expression used as a statement.
- Elements of array expressions, tuple expressions, call expressions, and tuple-style struct expressions.
- The tail expression of block expressions.
They are never allowed before:
- Range expressions.
- Binary operator expressions (ArithmeticOrLogicalExpression, ComparisonExpression, LazyBooleanExpression, TypeCastExpression, AssignmentExpression, CompoundAssignmentExpression).
Literal expressions
Syntax
LiteralExpression :
CHAR_LITERAL
| STRING_LITERAL
| RAW_STRING_LITERAL
| BYTE_LITERAL
| BYTE_STRING_LITERAL
| RAW_BYTE_STRING_LITERAL
| C_STRING_LITERAL
| RAW_C_STRING_LITERAL
| INTEGER_LITERAL
| FLOAT_LITERAL
|true
|false
A literal expression is an expression consisting of a single token, rather than a sequence of tokens, that immediately and directly denotes the value it evaluates to, rather than referring to it by name or some other evaluation rule.
A literal is a form of constant expression, so is evaluated (primarily) at compile time.
Each of the lexical literal forms described earlier can make up a literal expression, as can the keywords true
and false
.
#![allow(unused)] fn main() { "hello"; // string type '5'; // character type 5; // integer type }
In the descriptions below, the string representation of a token is the sequence of characters from the input which matched the token’s production in a Lexer grammar snippet.
Note: this string representation never includes a character
U+000D
(CR) immediately followed byU+000A
(LF): this pair would have been previously transformed into a singleU+000A
(LF).
Escapes
The descriptions of textual literal expressions below make use of several forms of escape.
Each form of escape is characterised by:
- an escape sequence: a sequence of characters, which always begins with
U+005C
(\
) - an escaped value: either a single character or an empty sequence of characters
In the definitions of escapes below:
- An octal digit is any of the characters in the range [
0
-7
]. - A hexadecimal digit is any of the characters in the ranges [
0
-9
], [a
-f
], or [A
-F
].
Simple escapes
Each sequence of characters occurring in the first column of the following table is an escape sequence.
In each case, the escaped value is the character given in the corresponding entry in the second column.
Escape sequence | Escaped value |
---|---|
\0 | U+0000 (NUL) |
\t | U+0009 (HT) |
\n | U+000A (LF) |
\r | U+000D (CR) |
\" | U+0022 (QUOTATION MARK) |
\' | U+0027 (APOSTROPHE) |
\\ | U+005C (REVERSE SOLIDUS) |
8-bit escapes
The escape sequence consists of \x
followed by two hexadecimal digits.
The escaped value is the character whose Unicode scalar value is the result of interpreting the final two characters in the escape sequence as a hexadecimal integer, as if by u8::from_str_radix
with radix 16.
Note: the escaped value therefore has a Unicode scalar value in the range of
u8
.
7-bit escapes
The escape sequence consists of \x
followed by an octal digit then a hexadecimal digit.
The escaped value is the character whose Unicode scalar value is the result of interpreting the final two characters in the escape sequence as a hexadecimal integer, as if by u8::from_str_radix
with radix 16.
Unicode escapes
The escape sequence consists of \u{
, followed by a sequence of characters each of which is a hexadecimal digit or _
, followed by }
.
The escaped value is the character whose Unicode scalar value is the result of interpreting the hexadecimal digits contained in the escape sequence as a hexadecimal integer, as if by u32::from_str_radix
with radix 16.
Note: the permitted forms of a CHAR_LITERAL or STRING_LITERAL token ensure that there is such a character.
String continuation escapes
The escape sequence consists of \
followed immediately by U+000A
(LF), and all following whitespace characters before the next non-whitespace character.
For this purpose, the whitespace characters are U+0009
(HT), U+000A
(LF), U+000D
(CR), and U+0020
(SPACE).
The escaped value is an empty sequence of characters.
Note: The effect of this form of escape is that a string continuation skips following whitespace, including additional newlines. Thus
a
,b
andc
are equal:#![allow(unused)] fn main() { let a = "foobar"; let b = "foo\ bar"; let c = "foo\ bar"; assert_eq!(a, b); assert_eq!(b, c); }
Skipping additional newlines (as in example c) is potentially confusing and unexpected. This behavior may be adjusted in the future. Until a decision is made, it is recommended to avoid relying on skipping multiple newlines with line continuations. See this issue for more information.
Character literal expressions
A character literal expression consists of a single CHAR_LITERAL token.
The expression’s type is the primitive char
type.
The token must not have a suffix.
The token’s literal content is the sequence of characters following the first U+0027
('
) and preceding the last U+0027
('
) in the string representation of the token.
The literal expression’s represented character is derived from the literal content as follows:
-
If the literal content is one of the following forms of escape sequence, the represented character is the escape sequence’s escaped value:
-
Otherwise the represented character is the single character that makes up the literal content.
The expression’s value is the char
corresponding to the represented character’s Unicode scalar value.
Note: the permitted forms of a CHAR_LITERAL token ensure that these rules always produce a single character.
Examples of character literal expressions:
#![allow(unused)] fn main() { 'R'; // R '\''; // ' '\x52'; // R '\u{00E6}'; // LATIN SMALL LETTER AE (U+00E6) }
String literal expressions
A string literal expression consists of a single STRING_LITERAL or RAW_STRING_LITERAL token.
The expression’s type is a shared reference (with static
lifetime) to the primitive str
type.
That is, the type is &'static str
.
The token must not have a suffix.
The token’s literal content is the sequence of characters following the first U+0022
("
) and preceding the last U+0022
("
) in the string representation of the token.
The literal expression’s represented string is a sequence of characters derived from the literal content as follows:
-
If the token is a STRING_LITERAL, each escape sequence of any of the following forms occurring in the literal content is replaced by the escape sequence’s escaped value.
These replacements take place in left-to-right order. For example, the token
"\\x41"
is converted to the characters\
x
4
1
. -
If the token is a RAW_STRING_LITERAL, the represented string is identical to the literal content.
The expression’s value is a reference to a statically allocated str
containing the UTF-8 encoding of the represented string.
Examples of string literal expressions:
#![allow(unused)] fn main() { "foo"; r"foo"; // foo "\"foo\""; r#""foo""#; // "foo" "foo #\"# bar"; r##"foo #"# bar"##; // foo #"# bar "\x52"; "R"; r"R"; // R "\\x52"; r"\x52"; // \x52 }
Byte literal expressions
A byte literal expression consists of a single BYTE_LITERAL token.
The expression’s type is the primitive u8
type.
The token must not have a suffix.
The token’s literal content is the sequence of characters following the first U+0027
('
) and preceding the last U+0027
('
) in the string representation of the token.
The literal expression’s represented character is derived from the literal content as follows:
-
If the literal content is one of the following forms of escape sequence, the represented character is the escape sequence’s escaped value:
-
Otherwise the represented character is the single character that makes up the literal content.
The expression’s value is the represented character’s Unicode scalar value.
Note: the permitted forms of a BYTE_LITERAL token ensure that these rules always produce a single character, whose Unicode scalar value is in the range of
u8
.
Examples of byte literal expressions:
#![allow(unused)] fn main() { b'R'; // 82 b'\''; // 39 b'\x52'; // 82 b'\xA0'; // 160 }
Byte string literal expressions
A byte string literal expression consists of a single BYTE_STRING_LITERAL or RAW_BYTE_STRING_LITERAL token.
The expression’s type is a shared reference (with static
lifetime) to an array whose element type is u8
.
That is, the type is &'static [u8; N]
, where N
is the number of bytes in the represented string described below.
The token must not have a suffix.
The token’s literal content is the sequence of characters following the first U+0022
("
) and preceding the last U+0022
("
) in the string representation of the token.
The literal expression’s represented string is a sequence of characters derived from the literal content as follows:
-
If the token is a BYTE_STRING_LITERAL, each escape sequence of any of the following forms occurring in the literal content is replaced by the escape sequence’s escaped value.
These replacements take place in left-to-right order. For example, the token
b"\\x41"
is converted to the characters\
x
4
1
. -
If the token is a RAW_BYTE_STRING_LITERAL, the represented string is identical to the literal content.
The expression’s value is a reference to a statically allocated array containing the Unicode scalar values of the characters in the represented string, in the same order.
Note: the permitted forms of BYTE_STRING_LITERAL and RAW_BYTE_STRING_LITERAL tokens ensure that these rules always produce array element values in the range of
u8
.
Examples of byte string literal expressions:
#![allow(unused)] fn main() { b"foo"; br"foo"; // foo b"\"foo\""; br#""foo""#; // "foo" b"foo #\"# bar"; br##"foo #"# bar"##; // foo #"# bar b"\x52"; b"R"; br"R"; // R b"\\x52"; br"\x52"; // \x52 }
C string literal expressions
A C string literal expression consists of a single C_STRING_LITERAL or RAW_C_STRING_LITERAL token.
The expression’s type is a shared reference (with static
lifetime) to the standard library CStr type.
That is, the type is &'static core::ffi::CStr
.
The token must not have a suffix.
The token’s literal content is the sequence of characters following the first "
and preceding the last "
in the string representation of the token.
The literal expression’s represented bytes are a sequence of bytes derived from the literal content as follows:
-
If the token is a C_STRING_LITERAL, the literal content is treated as a sequence of items, each of which is either a single Unicode character other than
\
or an escape. The sequence of items is converted to a sequence of bytes as follows:- Each single Unicode character contributes its UTF-8 representation.
- Each simple escape contributes the Unicode scalar value of its escaped value.
- Each 8-bit escape contributes a single byte containing the Unicode scalar value of its escaped value.
- Each unicode escape contributes the UTF-8 representation of its escaped value.
- Each string continuation escape contributes no bytes.
-
If the token is a RAW_C_STRING_LITERAL, the represented bytes are the UTF-8 encoding of the literal content.
Note: the permitted forms of C_STRING_LITERAL and RAW_C_STRING_LITERAL tokens ensure that the represented bytes never include a null byte.
The expression’s value is a reference to a statically allocated CStr whose array of bytes contains the represented bytes followed by a null byte.
Examples of C string literal expressions:
#![allow(unused)] fn main() { c"foo"; cr"foo"; // foo c"\"foo\""; cr#""foo""#; // "foo" c"foo #\"# bar"; cr##"foo #"# bar"##; // foo #"# bar c"\x52"; c"R"; cr"R"; // R c"\\x52"; cr"\x52"; // \x52 c"æ"; // LATIN SMALL LETTER AE (U+00E6) c"\u{00E6}"; // LATIN SMALL LETTER AE (U+00E6) c"\xC3\xA6"; // LATIN SMALL LETTER AE (U+00E6) c"\xE6".to_bytes(); // [230] c"\u{00E6}".to_bytes(); // [195, 166] }
Integer literal expressions
An integer literal expression consists of a single INTEGER_LITERAL token.
If the token has a suffix, the suffix must be the name of one of the primitive integer types: u8
, i8
, u16
, i16
, u32
, i32
, u64
, i64
, u128
, i128
, usize
, or isize
, and the expression has that type.
If the token has no suffix, the expression’s type is determined by type inference:
-
If an integer type can be uniquely determined from the surrounding program context, the expression has that type.
-
If the program context under-constrains the type, it defaults to the signed 32-bit integer
i32
. -
If the program context over-constrains the type, it is considered a static type error.
Examples of integer literal expressions:
#![allow(unused)] fn main() { 123; // type i32 123i32; // type i32 123u32; // type u32 123_u32; // type u32 let a: u64 = 123; // type u64 0xff; // type i32 0xff_u8; // type u8 0o70; // type i32 0o70_i16; // type i16 0b1111_1111_1001_0000; // type i32 0b1111_1111_1001_0000i64; // type i64 0usize; // type usize }
The value of the expression is determined from the string representation of the token as follows:
-
An integer radix is chosen by inspecting the first two characters of the string, as follows:
0b
indicates radix 20o
indicates radix 80x
indicates radix 16- otherwise the radix is 10.
-
If the radix is not 10, the first two characters are removed from the string.
-
Any suffix is removed from the string.
-
Any underscores are removed from the string.
-
The string is converted to a
u128
value as if byu128::from_str_radix
with the chosen radix. If the value does not fit inu128
, it is a compiler error. -
The
u128
value is converted to the expression’s type via a numeric cast.
Note: The final cast will truncate the value of the literal if it does not fit in the expression’s type.
rustc
includes a lint check namedoverflowing_literals
, defaulting todeny
, which rejects expressions where this occurs.
Note:
-1i8
, for example, is an application of the negation operator to the literal expression1i8
, not a single integer literal expression. See Overflow for notes on representing the most negative value for a signed type.
Floating-point literal expressions
A floating-point literal expression has one of two forms:
- a single FLOAT_LITERAL token
- a single INTEGER_LITERAL token which has a suffix and no radix indicator
If the token has a suffix, the suffix must be the name of one of the primitive floating-point types: f32
or f64
, and the expression has that type.
If the token has no suffix, the expression’s type is determined by type inference:
-
If a floating-point type can be uniquely determined from the surrounding program context, the expression has that type.
-
If the program context under-constrains the type, it defaults to
f64
. -
If the program context over-constrains the type, it is considered a static type error.
Examples of floating-point literal expressions:
#![allow(unused)] fn main() { 123.0f64; // type f64 0.1f64; // type f64 0.1f32; // type f32 12E+99_f64; // type f64 5f32; // type f32 let x: f64 = 2.; // type f64 }
The value of the expression is determined from the string representation of the token as follows:
-
Any suffix is removed from the string.
-
Any underscores are removed from the string.
-
The string is converted to the expression’s type as if by
f32::from_str
orf64::from_str
.
Note:
-1.0
, for example, is an application of the negation operator to the literal expression1.0
, not a single floating-point literal expression.
Note:
inf
andNaN
are not literal tokens. Thef32::INFINITY
,f64::INFINITY
,f32::NAN
, andf64::NAN
constants can be used instead of literal expressions. Inrustc
, a literal large enough to be evaluated as infinite will trigger theoverflowing_literals
lint check.
Boolean literal expressions
A boolean literal expression consists of one of the keywords true
or false
.
The expression’s type is the primitive boolean type, and its value is:
- true if the keyword is
true
- false if the keyword is
false
Path expressions
Syntax
PathExpression :
PathInExpression
| QualifiedPathInExpression
A path used as an expression context denotes either a local variable or an item.
Path expressions that resolve to local or static variables are place expressions, other paths are value expressions.
Using a static mut
variable requires an unsafe
block.
#![allow(unused)] fn main() { mod globals { pub static STATIC_VAR: i32 = 5; pub static mut STATIC_MUT_VAR: i32 = 7; } let local_var = 3; local_var; globals::STATIC_VAR; unsafe { globals::STATIC_MUT_VAR }; let some_constructor = Some::<i32>; let push_integer = Vec::<i32>::push; let slice_reverse = <[i32]>::reverse; }
Evaluation of associated constants is handled the same way as const
blocks.
Block expressions
Syntax
BlockExpression :
{
InnerAttribute*
Statements?
}
Statements :
Statement+
| Statement+ ExpressionWithoutBlock
| ExpressionWithoutBlock
A block expression, or block, is a control flow expression and anonymous namespace scope for items and variable declarations.
As a control flow expression, a block sequentially executes its component non-item declaration statements and then its final optional expression.
As an anonymous namespace scope, item declarations are only in scope inside the block itself and variables declared by let
statements are in scope from the next statement until the end of the block.
See the scopes chapter for more details.
The syntax for a block is {
, then any inner attributes, then any number of statements, then an optional expression, called the final operand, and finally a }
.
Statements are usually required to be followed by a semicolon, with two exceptions:
- Item declaration statements do not need to be followed by a semicolon.
- Expression statements usually require a following semicolon except if its outer expression is a flow control expression.
Furthermore, extra semicolons between statements are allowed, but these semicolons do not affect semantics.
When evaluating a block expression, each statement, except for item declaration statements, is executed sequentially. Then the final operand is executed, if given.
The type of a block is the type of the final operand, or ()
if the final operand is omitted.
#![allow(unused)] fn main() { fn fn_call() {} let _: () = { fn_call(); }; let five: i32 = { fn_call(); 5 }; assert_eq!(5, five); }
Note: As a control flow expression, if a block expression is the outer expression of an expression statement, the expected type is
()
unless it is followed immediately by a semicolon.
Blocks are always value expressions and evaluate the last operand in value expression context.
Note: This can be used to force moving a value if really needed. For example, the following example fails on the call to
consume_self
because the struct was moved out ofs
in the block expression.#![allow(unused)] fn main() { struct Struct; impl Struct { fn consume_self(self) {} fn borrow_self(&self) {} } fn move_by_block_expression() { let s = Struct; // Move the value out of `s` in the block expression. (&{ s }).borrow_self(); // Fails to execute because `s` is moved out of. s.consume_self(); } }
async
blocks
Syntax
AsyncBlockExpression :
async
move
? BlockExpression
An async block is a variant of a block expression which evaluates to a future. The final expression of the block, if present, determines the result value of the future.
Executing an async block is similar to executing a closure expression:
its immediate effect is to produce and return an anonymous type.
Whereas closures return a type that implements one or more of the std::ops::Fn
traits, however, the type returned for an async block implements the std::future::Future
trait.
The actual data format for this type is unspecified.
Note: The future type that rustc generates is roughly equivalent to an enum with one variant per
await
point, where each variant stores the data needed to resume from its corresponding point.
Edition differences: Async blocks are only available beginning with Rust 2018.
Capture modes
Async blocks capture variables from their environment using the same capture modes as closures.
Like closures, when written async { .. }
the capture mode for each variable will be inferred from the content of the block.
async move { .. }
blocks however will move all referenced variables into the resulting future.
Async context
Because async blocks construct a future, they define an async context which can in turn contain await
expressions.
Async contexts are established by async blocks as well as the bodies of async functions, whose semantics are defined in terms of async blocks.
Control-flow operators
Async blocks act like a function boundary, much like closures.
Therefore, the ?
operator and return
expressions both affect the output of the future, not the enclosing function or other context.
That is, return <expr>
from within an async block will return the result of <expr>
as the output of the future.
Similarly, if <expr>?
propagates an error, that error is propagated as the result of the future.
Finally, the break
and continue
keywords cannot be used to branch out from an async block.
Therefore the following is illegal:
#![allow(unused)] fn main() { loop { async move { break; // error[E0267]: `break` inside of an `async` block } } }
const
blocks
Syntax
ConstBlockExpression :
const
BlockExpression
A const block is a variant of a block expression whose body evaluates at compile-time instead of at runtime.
Const blocks allows you to define a constant value without having to define new constant items, and thus they are also sometimes referred as inline consts. It also supports type inference so there is no need to specify the type, unlike constant items.
Const blocks have the ability to reference generic parameters in scope, unlike free constant items. They are desugared to constant items with generic parameters in scope (similar to associated constants, but without a trait or type they are associated with). For example, this code:
#![allow(unused)] fn main() { fn foo<T>() -> usize { const { std::mem::size_of::<T>() + 1 } } }
is equivalent to:
#![allow(unused)] fn main() { fn foo<T>() -> usize { { struct Const<T>(T); impl<T> Const<T> { const CONST: usize = std::mem::size_of::<T>() + 1; } Const::<T>::CONST } } }
If the const block expression is executed at runtime, then the constant is guaranteed to be evaluated, even if its return value is ignored:
#![allow(unused)] fn main() { fn foo<T>() -> usize { // If this code ever gets executed, then the assertion has definitely // been evaluated at compile-time. const { assert!(std::mem::size_of::<T>() > 0); } // Here we can have unsafe code relying on the type being non-zero-sized. /* ... */ 42 } }
If the const block expression is not executed at runtime, it may or may not be evaluated:
#![allow(unused)] fn main() { if false { // The panic may or may not occur when the program is built. const { panic!(); } } }
unsafe
blocks
Syntax
UnsafeBlockExpression :
unsafe
BlockExpression
See unsafe
blocks for more information on when to use unsafe
.
A block of code can be prefixed with the unsafe
keyword to permit unsafe operations.
Examples:
#![allow(unused)] fn main() { unsafe { let b = [13u8, 17u8]; let a = &b[0] as *const u8; assert_eq!(*a, 13); assert_eq!(*a.offset(1), 17); } unsafe fn an_unsafe_fn() -> i32 { 10 } let a = unsafe { an_unsafe_fn() }; }
Labelled block expressions
Labelled block expressions are documented in the Loops and other breakable expressions section.
Attributes on block expressions
Inner attributes are allowed directly after the opening brace of a block expression in the following situations:
- Function and method bodies.
- Loop bodies (
loop
,while
,while let
, andfor
). - Block expressions used as a statement.
- Block expressions as elements of array expressions, tuple expressions, call expressions, and tuple-style struct expressions.
- A block expression as the tail expression of another block expression.
The attributes that have meaning on a block expression are cfg
and the lint check attributes.
For example, this function returns true
on unix platforms and false
on other platforms.
#![allow(unused)] fn main() { fn is_unix_platform() -> bool { #[cfg(unix)] { true } #[cfg(not(unix))] { false } } }
Operator expressions
Syntax
OperatorExpression :
BorrowExpression
| DereferenceExpression
| ErrorPropagationExpression
| NegationExpression
| ArithmeticOrLogicalExpression
| ComparisonExpression
| LazyBooleanExpression
| TypeCastExpression
| AssignmentExpression
| CompoundAssignmentExpression
Operators are defined for built in types by the Rust language.
Many of the following operators can also be overloaded using traits in std::ops
or std::cmp
.
Overflow
Integer operators will panic when they overflow when compiled in debug mode.
The -C debug-assertions
and -C overflow-checks
compiler flags can be used to control this more directly.
The following things are considered to be overflow:
- When
+
,*
or binary-
create a value greater than the maximum value, or less than the minimum value that can be stored. - Applying unary
-
to the most negative value of any signed integer type, unless the operand is a literal expression (or a literal expression standing alone inside one or more grouped expressions). - Using
/
or%
, where the left-hand argument is the smallest integer of a signed integer type and the right-hand argument is-1
. These checks occur even when-C overflow-checks
is disabled, for legacy reasons. - Using
<<
or>>
where the right-hand argument is greater than or equal to the number of bits in the type of the left-hand argument, or is negative.
Note: The exception for literal expressions behind unary
-
means that forms such as-128_i8
orlet j: i8 = -(128)
never cause a panic and have the expected value of -128.In these cases, the literal expression already has the most negative value for its type (for example,
128_i8
has the value -128) because integer literals are truncated to their type per the description in Integer literal expressions.Negation of these most negative values leaves the value unchanged due to two’s complement overflow conventions.
In
rustc
, these most negative expressions are also ignored by theoverflowing_literals
lint check.
Borrow operators
Syntax
BorrowExpression :
(&
|&&
) Expression
| (&
|&&
)mut
Expression
| (&
|&&
)raw
const
Expression
| (&
|&&
)raw
mut
Expression
The &
(shared borrow) and &mut
(mutable borrow) operators are unary prefix operators.
When applied to a place expression, this expressions produces a reference (pointer) to the location that the value refers to.
The memory location is also placed into a borrowed state for the duration of the reference.
For a shared borrow (&
), this implies that the place may not be mutated, but it may be read or shared again.
For a mutable borrow (&mut
), the place may not be accessed in any way until the borrow expires.
&mut
evaluates its operand in a mutable place expression context.
If the &
or &mut
operators are applied to a value expression, then a temporary value is created.
These operators cannot be overloaded.
#![allow(unused)] fn main() { { // a temporary with value 7 is created that lasts for this scope. let shared_reference = &7; } let mut array = [-2, 3, 9]; { // Mutably borrows `array` for this scope. // `array` may only be used through `mutable_reference`. let mutable_reference = &mut array; } }
Even though &&
is a single token (the lazy ‘and’ operator), when used in the context of borrow expressions it works as two borrows:
#![allow(unused)] fn main() { // same meanings: let a = && 10; let a = & & 10; // same meanings: let a = &&&& mut 10; let a = && && mut 10; let a = & & & & mut 10; }
Raw borrow operators
&raw const
and &raw mut
are the raw borrow operators.
The operand expression of these operators is evaluated in place expression context.
&raw const expr
then creates a const raw pointer of type *const T
to the given place, and &raw mut expr
creates a mutable raw pointer of type *mut T
.
The raw borrow operators must be used instead of a borrow operator whenever the place expression could evaluate to a place that is not properly aligned or does not store a valid value as determined by its type, or whenever creating a reference would introduce incorrect aliasing assumptions. In those situations, using a borrow operator would cause undefined behavior by creating an invalid reference, but a raw pointer may still be constructed.
The following is an example of creating a raw pointer to an unaligned place through a packed
struct:
#![allow(unused)] fn main() { #[repr(packed)] struct Packed { f1: u8, f2: u16, } let packed = Packed { f1: 1, f2: 2 }; // `&packed.f2` would create an unaligned reference, and thus be undefined behavior! let raw_f2 = &raw const packed.f2; assert_eq!(unsafe { raw_f2.read_unaligned() }, 2); }
The following is an example of creating a raw pointer to a place that does not contain a valid value:
#![allow(unused)] fn main() { use std::mem::MaybeUninit; struct Demo { field: bool, } let mut uninit = MaybeUninit::<Demo>::uninit(); // `&uninit.as_mut().field` would create a reference to an uninitialized `bool`, // and thus be undefined behavior! let f1_ptr = unsafe { &raw mut (*uninit.as_mut_ptr()).field }; unsafe { f1_ptr.write(true); } let init = unsafe { uninit.assume_init() }; }
The dereference operator
Syntax
DereferenceExpression :
*
Expression
The *
(dereference) operator is also a unary prefix operator.
When applied to a pointer it denotes the pointed-to location.
If the expression is of type &mut T
or *mut T
, and is either a local variable, a (nested) field of a local variable or is a mutable place expression, then the resulting memory location can be assigned to.
Dereferencing a raw pointer requires unsafe
.
On non-pointer types *x
is equivalent to *std::ops::Deref::deref(&x)
in an immutable place expression context and *std::ops::DerefMut::deref_mut(&mut x)
in a mutable place expression context.
#![allow(unused)] fn main() { let x = &7; assert_eq!(*x, 7); let y = &mut 9; *y = 11; assert_eq!(*y, 11); }
The question mark operator
Syntax
ErrorPropagationExpression :
Expression?
The question mark operator (?
) unwraps valid values or returns erroneous values, propagating them to the calling function.
It is a unary postfix operator that can only be applied to the types Result<T, E>
and Option<T>
.
When applied to values of the Result<T, E>
type, it propagates errors.
If the value is Err(e)
, then it will return Err(From::from(e))
from the enclosing function or closure.
If applied to Ok(x)
, then it will unwrap the value to evaluate to x
.
#![allow(unused)] fn main() { use std::num::ParseIntError; fn try_to_parse() -> Result<i32, ParseIntError> { let x: i32 = "123".parse()?; // x = 123 let y: i32 = "24a".parse()?; // returns an Err() immediately Ok(x + y) // Doesn't run. } let res = try_to_parse(); println!("{:?}", res); assert!(res.is_err()) }
When applied to values of the Option<T>
type, it propagates None
s.
If the value is None
, then it will return None
.
If applied to Some(x)
, then it will unwrap the value to evaluate to x
.
#![allow(unused)] fn main() { fn try_option_some() -> Option<u8> { let val = Some(1)?; Some(val) } assert_eq!(try_option_some(), Some(1)); fn try_option_none() -> Option<u8> { let val = None?; Some(val) } assert_eq!(try_option_none(), None); }
?
cannot be overloaded.
Negation operators
Syntax
NegationExpression :
-
Expression
|!
Expression
These are the last two unary operators. This table summarizes the behavior of them on primitive types and which traits are used to overload these operators for other types. Remember that signed integers are always represented using two’s complement. The operands of all of these operators are evaluated in value expression context so are moved or copied.
Symbol | Integer | bool | Floating Point | Overloading Trait |
---|---|---|---|---|
- | Negation* | Negation | std::ops::Neg | |
! | Bitwise NOT | Logical NOT | std::ops::Not |
* Only for signed integer types.
Here are some example of these operators
#![allow(unused)] fn main() { let x = 6; assert_eq!(-x, -6); assert_eq!(!x, -7); assert_eq!(true, !false); }
Arithmetic and Logical Binary Operators
Syntax
ArithmeticOrLogicalExpression :
Expression+
Expression
| Expression-
Expression
| Expression*
Expression
| Expression/
Expression
| Expression%
Expression
| Expression&
Expression
| Expression|
Expression
| Expression^
Expression
| Expression<<
Expression
| Expression>>
Expression
Binary operators expressions are all written with infix notation. This table summarizes the behavior of arithmetic and logical binary operators on primitive types and which traits are used to overload these operators for other types. Remember that signed integers are always represented using two’s complement. The operands of all of these operators are evaluated in value expression context so are moved or copied.
Symbol | Integer | bool | Floating Point | Overloading Trait | Overloading Compound Assignment Trait |
---|---|---|---|---|---|
+ | Addition | Addition | std::ops::Add | std::ops::AddAssign | |
- | Subtraction | Subtraction | std::ops::Sub | std::ops::SubAssign | |
* | Multiplication | Multiplication | std::ops::Mul | std::ops::MulAssign | |
/ | Division*† | Division | std::ops::Div | std::ops::DivAssign | |
% | Remainder**† | Remainder | std::ops::Rem | std::ops::RemAssign | |
& | Bitwise AND | Logical AND | std::ops::BitAnd | std::ops::BitAndAssign | |
| | Bitwise OR | Logical OR | std::ops::BitOr | std::ops::BitOrAssign | |
^ | Bitwise XOR | Logical XOR | std::ops::BitXor | std::ops::BitXorAssign | |
<< | Left Shift | std::ops::Shl | std::ops::ShlAssign | ||
>> | Right Shift*** | std::ops::Shr | std::ops::ShrAssign |
* Integer division rounds towards zero.
** Rust uses a remainder defined with truncating division. Given remainder = dividend % divisor
, the remainder will have the same sign as the dividend.
*** Arithmetic right shift on signed integer types, logical right shift on unsigned integer types.
† For integer types, division by zero panics.
Here are examples of these operators being used.
#![allow(unused)] fn main() { assert_eq!(3 + 6, 9); assert_eq!(5.5 - 1.25, 4.25); assert_eq!(-5 * 14, -70); assert_eq!(14 / 3, 4); assert_eq!(100 % 7, 2); assert_eq!(0b1010 & 0b1100, 0b1000); assert_eq!(0b1010 | 0b1100, 0b1110); assert_eq!(0b1010 ^ 0b1100, 0b110); assert_eq!(13 << 3, 104); assert_eq!(-10 >> 2, -3); }
Comparison Operators
Syntax
ComparisonExpression :
Expression==
Expression
| Expression!=
Expression
| Expression>
Expression
| Expression<
Expression
| Expression>=
Expression
| Expression<=
Expression
Comparison operators are also defined both for primitive types and many types in the standard library.
Parentheses are required when chaining comparison operators. For example, the expression a == b == c
is invalid and may be written as (a == b) == c
.
Unlike arithmetic and logical operators, the traits for overloading these operators are used more generally to show how a type may be compared and will likely be assumed to define actual comparisons by functions that use these traits as bounds. Many functions and macros in the standard library can then use that assumption (although not to ensure safety). Unlike the arithmetic and logical operators above, these operators implicitly take shared borrows of their operands, evaluating them in place expression context:
#![allow(unused)] fn main() { let a = 1; let b = 1; a == b; // is equivalent to ::std::cmp::PartialEq::eq(&a, &b); }
This means that the operands don’t have to be moved out of.
Symbol | Meaning | Overloading method |
---|---|---|
== | Equal | std::cmp::PartialEq::eq |
!= | Not equal | std::cmp::PartialEq::ne |
> | Greater than | std::cmp::PartialOrd::gt |
< | Less than | std::cmp::PartialOrd::lt |
>= | Greater than or equal to | std::cmp::PartialOrd::ge |
<= | Less than or equal to | std::cmp::PartialOrd::le |
Here are examples of the comparison operators being used.
#![allow(unused)] fn main() { assert!(123 == 123); assert!(23 != -12); assert!(12.5 > 12.2); assert!([1, 2, 3] < [1, 3, 4]); assert!('A' <= 'B'); assert!("World" >= "Hello"); }
Lazy boolean operators
Syntax
LazyBooleanExpression :
Expression||
Expression
| Expression&&
Expression
The operators ||
and &&
may be applied to operands of boolean type.
The ||
operator denotes logical ‘or’, and the &&
operator denotes logical ‘and’.
They differ from |
and &
in that the right-hand operand is only evaluated when the left-hand operand does not already determine the result of the expression.
That is, ||
only evaluates its right-hand operand when the left-hand operand evaluates to false
, and &&
only when it evaluates to true
.
#![allow(unused)] fn main() { let x = false || true; // true let y = false && panic!(); // false, doesn't evaluate `panic!()` }
Type cast expressions
Syntax
TypeCastExpression :
Expressionas
TypeNoBounds
A type cast expression is denoted with the binary operator as
.
Executing an as
expression casts the value on the left-hand side to the type on the right-hand side.
An example of an as
expression:
#![allow(unused)] fn main() { fn sum(values: &[f64]) -> f64 { 0.0 } fn len(values: &[f64]) -> i32 { 0 } fn average(values: &[f64]) -> f64 { let sum: f64 = sum(values); let size: f64 = len(values) as f64; sum / size } }
as
can be used to explicitly perform coercions, as well as the following additional casts.
Any cast that does not fit either a coercion rule or an entry in the table is a compiler error.
Here *T
means either *const T
or *mut T
. m
stands for optional mut
in
reference types and mut
or const
in pointer types.
Type of e | U | Cast performed by e as U |
---|---|---|
Integer or Float type | Integer or Float type | Numeric cast |
Enumeration | Integer type | Enum cast |
bool or char | Integer type | Primitive to integer cast |
u8 | char | u8 to char cast |
*T | *V where V: Sized * | Pointer to pointer cast |
*T where T: Sized | Integer type | Pointer to address cast |
Integer type | *V where V: Sized | Address to pointer cast |
&m₁ T | *m₂ T ** | Reference to pointer cast |
&m₁ [T; n] | *m₂ T ** | Array to pointer cast |
Function item | Function pointer | Function item to function pointer cast |
Function item | *V where V: Sized | Function item to pointer cast |
Function item | Integer | Function item to address cast |
Function pointer | *V where V: Sized | Function pointer to pointer cast |
Function pointer | Integer | Function pointer to address cast |
Closure *** | Function pointer | Closure to function pointer cast |
* or T
and V
are compatible unsized types, e.g., both slices, both the same trait object.
** only when m₁
is mut
or m₂
is const
. Casting mut
reference to
const
pointer is allowed.
*** only for closures that do not capture (close over) any local variables
Semantics
Numeric cast
- Casting between two integers of the same size (e.g. i32 -> u32) is a no-op (Rust uses 2’s complement for negative values of fixed integers)
- Casting from a larger integer to a smaller integer (e.g. u32 -> u8) will truncate
- Casting from a smaller integer to a larger integer (e.g. u8 -> u32) will
- zero-extend if the source is unsigned
- sign-extend if the source is signed
- Casting from a float to an integer will round the float towards zero
NaN
will return0
- Values larger than the maximum integer value, including
INFINITY
, will saturate to the maximum value of the integer type. - Values smaller than the minimum integer value, including
NEG_INFINITY
, will saturate to the minimum value of the integer type.
- Casting from an integer to float will produce the closest possible float *
- if necessary, rounding is according to
roundTiesToEven
mode *** - on overflow, infinity (of the same sign as the input) is produced
- note: with the current set of numeric types, overflow can only happen
on
u128 as f32
for values greater or equal tof32::MAX + (0.5 ULP)
- if necessary, rounding is according to
- Casting from an f32 to an f64 is perfect and lossless
- Casting from an f64 to an f32 will produce the closest possible f32 **
- if necessary, rounding is according to
roundTiesToEven
mode *** - on overflow, infinity (of the same sign as the input) is produced
- if necessary, rounding is according to
* if integer-to-float casts with this rounding mode and overflow behavior are not supported natively by the hardware, these casts will likely be slower than expected.
** if f64-to-f32 casts with this rounding mode and overflow behavior are not supported natively by the hardware, these casts will likely be slower than expected.
*** as defined in IEEE 754-2008 §4.3.1: pick the nearest floating point number, preferring the one with an even least significant digit if exactly halfway between two floating point numbers.
Enum cast
Casts an enum to its discriminant, then uses a numeric cast if needed. Casting is limited to the following kinds of enumerations:
- Unit-only enums
- Field-less enums without explicit discriminants, or where only unit-variants have explicit discriminants
Primitive to integer cast
false
casts to0
,true
casts to1
char
casts to the value of the code point, then uses a numeric cast if needed.
u8
to char
cast
Casts to the char
with the corresponding code point.
Pointer to address cast
Casting from a raw pointer to an integer produces the machine address of the referenced memory.
If the integer type is smaller than the pointer type, the address may be truncated; using usize
avoids this.
Address to pointer cast
Casting from an integer to a raw pointer interprets the integer as a memory address and produces a pointer referencing that memory.
Warning: This interacts with the Rust memory model, which is still under development. A pointer obtained from this cast may suffer additional restrictions even if it is bitwise equal to a valid pointer. Dereferencing such a pointer may be undefined behavior if aliasing rules are not followed.
A trivial example of sound address arithmetic:
#![allow(unused)] fn main() { let mut values: [i32; 2] = [1, 2]; let p1: *mut i32 = values.as_mut_ptr(); let first_address = p1 as usize; let second_address = first_address + 4; // 4 == size_of::<i32>() let p2 = second_address as *mut i32; unsafe { *p2 += 1; } assert_eq!(values[1], 3); }
Pointer-to-pointer cast
*const T
/ *mut T
can be cast to *const U
/ *mut U
with the following behavior:
-
If
T
andU
are both sized, the pointer is returned unchanged. -
If
T
andU
are both unsized, the pointer is also returned unchanged. In particular, the metadata is preserved exactly.For instance, a cast from
*const [T]
to*const [U]
preserves the number of elements. Note that, as a consequence, such casts do not necessarily preserve the size of the pointer’s referent (e.g., casting*const [u16]
to*const [u8]
will result in a raw pointer which refers to an object of half the size of the original). The same holds forstr
and any compound type whose unsized tail is a slice type, such asstruct Foo(i32, [u8])
or(u64, Foo)
. -
If
T
is unsized andU
is sized, the cast discards all metadata that completes the wide pointerT
and produces a thin pointerU
consisting of the data part of the unsized pointer.
Assignment expressions
Syntax
AssignmentExpression :
Expression=
Expression
An assignment expression moves a value into a specified place.
An assignment expression consists of a mutable assignee expression, the assignee operand, followed by an equals sign (=
) and a value expression, the assigned value operand.
In its most basic form, an assignee expression is a place expression, and we discuss this case first.
The more general case of destructuring assignment is discussed below, but this case always decomposes into sequential assignments to place expressions, which may be considered the more fundamental case.
Basic assignments
Evaluating assignment expressions begins by evaluating its operands. The assigned value operand is evaluated first, followed by the assignee expression. For destructuring assignment, subexpressions of the assignee expression are evaluated left-to-right.
Note: This is different than other expressions in that the right operand is evaluated before the left one.
It then has the effect of first dropping the value at the assigned place, unless the place is an uninitialized local variable or an uninitialized field of a local variable. Next it either copies or moves the assigned value to the assigned place.
An assignment expression always produces the unit value.
Example:
#![allow(unused)] fn main() { let mut x = 0; let y = 0; x = y; }
Destructuring assignments
Destructuring assignment is a counterpart to destructuring pattern matches for variable declaration, permitting assignment to complex values, such as tuples or structs. For instance, we may swap two mutable variables:
#![allow(unused)] fn main() { let (mut a, mut b) = (0, 1); // Swap `a` and `b` using destructuring assignment. (b, a) = (a, b); }
In contrast to destructuring declarations using let
, patterns may not appear on the left-hand side of an assignment due to syntactic ambiguities.
Instead, a group of expressions that correspond to patterns are designated to be assignee expressions, and permitted on the left-hand side of an assignment.
Assignee expressions are then desugared to pattern matches followed by sequential assignment.
The desugared patterns must be irrefutable: in particular, this means that only slice patterns whose length is known at compile-time, and the trivial slice [..]
, are permitted for destructuring assignment.
The desugaring method is straightforward, and is illustrated best by example.
#![allow(unused)] fn main() { struct Struct { x: u32, y: u32 } let (mut a, mut b) = (0, 0); (a, b) = (3, 4); [a, b] = [3, 4]; Struct { x: a, y: b } = Struct { x: 3, y: 4}; // desugars to: { let (_a, _b) = (3, 4); a = _a; b = _b; } { let [_a, _b] = [3, 4]; a = _a; b = _b; } { let Struct { x: _a, y: _b } = Struct { x: 3, y: 4}; a = _a; b = _b; } }
Identifiers are not forbidden from being used multiple times in a single assignee expression.
Underscore expressions and empty range expressions may be used to ignore certain values, without binding them.
Note that default binding modes do not apply for the desugared expression.
Compound assignment expressions
Syntax
CompoundAssignmentExpression :
Expression+=
Expression
| Expression-=
Expression
| Expression*=
Expression
| Expression/=
Expression
| Expression%=
Expression
| Expression&=
Expression
| Expression|=
Expression
| Expression^=
Expression
| Expression<<=
Expression
| Expression>>=
Expression
Compound assignment expressions combine arithmetic and logical binary operators with assignment expressions.
For example:
#![allow(unused)] fn main() { let mut x = 5; x += 1; assert!(x == 6); }
The syntax of compound assignment is a mutable place expression, the assigned operand, then one of the operators followed by an =
as a single token (no whitespace), and then a value expression, the modifying operand.
Unlike other place operands, the assigned place operand must be a place expression. Attempting to use a value expression is a compiler error rather than promoting it to a temporary.
Evaluation of compound assignment expressions depends on the types of the operators.
If both types are primitives, then the modifying operand will be evaluated first followed by the assigned operand. It will then set the value of the assigned operand’s place to the value of performing the operation of the operator with the values of the assigned operand and modifying operand.
Note: This is different than other expressions in that the right operand is evaluated before the left one.
Otherwise, this expression is syntactic sugar for calling the function of the overloading compound assignment trait of the operator (see the table earlier in this chapter). A mutable borrow of the assigned operand is automatically taken.
For example, the following expression statements in example
are equivalent:
#![allow(unused)] fn main() { struct Addable; use std::ops::AddAssign; impl AddAssign<Addable> for Addable { /* */ fn add_assign(&mut self, other: Addable) {} } fn example() { let (mut a1, a2) = (Addable, Addable); a1 += a2; let (mut a1, a2) = (Addable, Addable); AddAssign::add_assign(&mut a1, a2); } }
Like assignment expressions, compound assignment expressions always produce the unit value.
Warning: The evaluation order of operands swaps depending on the types of the operands: with primitive types the right-hand side will get evaluated first, while with non-primitive types the left-hand side will get evaluated first. Try not to write code that depends on the evaluation order of operands in compound assignment expressions. See this test for an example of using this dependency.
Grouped expressions
Syntax
GroupedExpression :
(
Expression)
A parenthesized expression wraps a single expression, evaluating to that expression.
The syntax for a parenthesized expression is a (
, then an expression, called the enclosed operand, and then a )
.
Parenthesized expressions evaluate to the value of the enclosed operand. Unlike other expressions, parenthesized expressions are both place expressions and value expressions. When the enclosed operand is a place expression, it is a place expression and when the enclosed operand is a value expression, it is a value expression.
Parentheses can be used to explicitly modify the precedence order of subexpressions within an expression.
An example of a parenthesized expression:
#![allow(unused)] fn main() { let x: i32 = 2 + 3 * 4; // not parenthesized let y: i32 = (2 + 3) * 4; // parenthesized assert_eq!(x, 14); assert_eq!(y, 20); }
An example of a necessary use of parentheses is when calling a function pointer that is a member of a struct:
#![allow(unused)] fn main() { struct A { f: fn() -> &'static str } impl A { fn f(&self) -> &'static str { "The method f" } } let a = A{f: || "The field f"}; assert_eq!( a.f (), "The method f"); assert_eq!((a.f)(), "The field f"); }
Array and array index expressions
Array expressions
Syntax
ArrayExpression :
[
ArrayElements?]
ArrayElements :
Expression (,
Expression )*,
?
| Expression;
Expression
Array expressions construct arrays. Array expressions come in two forms.
The first form lists out every value in the array. The syntax for this form is a comma-separated list of expressions of uniform type enclosed in square brackets. This produces an array containing each of these values in the order they are written.
The syntax for the second form is two expressions separated by a semicolon (;
) enclosed in square brackets.
The expression before the ;
is called the repeat operand.
The expression after the ;
is called the length operand.
It must have type usize
and be a constant expression, such as a literal or a constant item.
An array expression of this form creates an array with the length of the value of the length operand with each element being a copy of the repeat operand.
That is, [a; b]
creates an array containing b
copies of the value of a
.
If the length operand has a value greater than 1 then this requires that the type of the repeat operand is Copy
or that it must be a path to a constant item.
When the repeat operand is a constant item, it is evaluated the length operand’s value times.
If that value is 0
, then the constant item is not evaluated at all.
For expressions that are not a constant item, it is evaluated exactly once, and then the result is copied the length operand’s value times.
#![allow(unused)] fn main() { [1, 2, 3, 4]; ["a", "b", "c", "d"]; [0; 128]; // array with 128 zeros [0u8, 0u8, 0u8, 0u8,]; [[1, 0, 0], [0, 1, 0], [0, 0, 1]]; // 2D array const EMPTY: Vec<i32> = Vec::new(); [EMPTY; 2]; }
Array and slice indexing expressions
Syntax
IndexExpression :
Expression[
Expression]
Array and slice-typed values can be indexed by writing a square-bracket-enclosed expression of type usize
(the index) after them.
When the array is mutable, the resulting memory location can be assigned to.
For other types an index expression a[b]
is equivalent to *std::ops::Index::index(&a, b)
, or *std::ops::IndexMut::index_mut(&mut a, b)
in a mutable place expression context.
Just as with methods, Rust will also insert dereference operations on a
repeatedly to find an implementation.
Indices are zero-based for arrays and slices. Array access is a constant expression, so bounds can be checked at compile-time with a constant index value. Otherwise a check will be performed at run-time that will put the thread in a panicked state if it fails.
#![allow(unused)] fn main() { // lint is deny by default. #![warn(unconditional_panic)] ([1, 2, 3, 4])[2]; // Evaluates to 3 let b = [[1, 0, 0], [0, 1, 0], [0, 0, 1]]; b[1][2]; // multidimensional array indexing let x = (["a", "b"])[10]; // warning: index out of bounds let n = 10; let y = (["a", "b"])[n]; // panics let arr = ["a", "b"]; arr[10]; // warning: index out of bounds }
The array index expression can be implemented for types other than arrays and slices by implementing the Index and IndexMut traits.
Tuple and tuple indexing expressions
Tuple expressions
Syntax
TupleExpression :
(
TupleElements?)
TupleElements :
( Expression,
)+ Expression?
A tuple expression constructs tuple values.
The syntax for tuple expressions is a parenthesized, comma separated list of expressions, called the tuple initializer operands. 1-ary tuple expressions require a comma after their tuple initializer operand to be disambiguated with a parenthetical expression.
Tuple expressions are a value expression that evaluate into a newly constructed value of a tuple type.
The number of tuple initializer operands is the arity of the constructed tuple.
Tuple expressions without any tuple initializer operands produce the unit tuple.
For other tuple expressions, the first written tuple initializer operand initializes the field 0
and subsequent operands initializes the next highest field.
For example, in the tuple expression ('a', 'b', 'c')
, 'a'
initializes the value of the field 0
, 'b'
field 1
, and 'c'
field 2
.
Examples of tuple expressions and their types:
Expression | Type |
---|---|
() | () (unit) |
(0.0, 4.5) | (f64, f64) |
("x".to_string(), ) | (String, ) |
("a", 4usize, true) | (&'static str, usize, bool) |
Tuple indexing expressions
Syntax
TupleIndexingExpression :
Expression.
TUPLE_INDEX
A tuple indexing expression accesses fields of tuples and tuple structs.
The syntax for a tuple index expression is an expression, called the tuple operand, then a .
, then finally a tuple index.
The syntax for the tuple index is a decimal literal with no leading zeros, underscores, or suffix.
For example 0
and 2
are valid tuple indices but not 01
, 0_
, nor 0i32
.
The type of the tuple operand must be a tuple type or a tuple struct. The tuple index must be a name of a field of the type of the tuple operand.
Evaluation of tuple index expressions has no side effects beyond evaluation of its tuple operand. As a place expression, it evaluates to the location of the field of the tuple operand with the same name as the tuple index.
Examples of tuple indexing expressions:
#![allow(unused)] fn main() { // Indexing a tuple let pair = ("a string", 2); assert_eq!(pair.1, 2); // Indexing a tuple struct struct Point(f32, f32); let point = Point(1.0, 0.0); assert_eq!(point.0, 1.0); assert_eq!(point.1, 0.0); }
Note: Unlike field access expressions, tuple index expressions can be the function operand of a call expression as it cannot be confused with a method call since method names cannot be numbers.
Note: Although arrays and slices also have elements, you must use an array or slice indexing expression or a slice pattern to access their elements.
Struct expressions
Syntax
StructExpression :
StructExprStruct
| StructExprTuple
| StructExprUnitStructExprStruct :
PathInExpression{
(StructExprFields | StructBase)?}
StructExprFields :
StructExprField (,
StructExprField)* (,
StructBase |,
?)StructExprField :
OuterAttribute *
(
IDENTIFIER
| (IDENTIFIER | TUPLE_INDEX):
Expression
)StructBase :
..
ExpressionStructExprTuple :
PathInExpression(
( Expression (,
Expression)*,
? )?
)
StructExprUnit : PathInExpression
A struct expression creates a struct, enum, or union value. It consists of a path to a struct, enum variant, or union item followed by the values for the fields of the item. There are three forms of struct expressions: struct, tuple, and unit.
The following are examples of struct expressions:
#![allow(unused)] fn main() { struct Point { x: f64, y: f64 } struct NothingInMe { } struct TuplePoint(f64, f64); mod game { pub struct User<'a> { pub name: &'a str, pub age: u32, pub score: usize } } struct Cookie; fn some_fn<T>(t: T) {} Point {x: 10.0, y: 20.0}; NothingInMe {}; TuplePoint(10.0, 20.0); TuplePoint { 0: 10.0, 1: 20.0 }; // Results in the same value as the above line let u = game::User {name: "Joe", age: 35, score: 100_000}; some_fn::<Cookie>(Cookie); }
Field struct expression
A struct expression with fields enclosed in curly braces allows you to specify the value for each individual field in any order. The field name is separated from its value with a colon.
A value of a union type can only be created using this syntax, and it must specify exactly one field.
Functional update syntax
A struct expression that constructs a value of a struct type can terminate with the syntax ..
followed by an expression to denote a functional update.
The expression following ..
(the base) must have the same struct type as the new struct type being formed.
The entire expression uses the given values for the fields that were specified and moves or copies the remaining fields from the base expression. As with all struct expressions, all of the fields of the struct must be visible, even those not explicitly named.
#![allow(unused)] fn main() { struct Point3d { x: i32, y: i32, z: i32 } let mut base = Point3d {x: 1, y: 2, z: 3}; let y_ref = &mut base.y; Point3d {y: 0, z: 10, .. base}; // OK, only base.x is accessed drop(y_ref); }
Struct expressions with curly braces can’t be used directly in a loop or if expression’s head, or in the scrutinee of an if let or match expression. However, struct expressions can be used in these situations if they are within another expression, for example inside parentheses.
The field names can be decimal integer values to specify indices for constructing tuple structs. This can be used with base structs to fill out the remaining indices not specified:
#![allow(unused)] fn main() { struct Color(u8, u8, u8); let c1 = Color(0, 0, 0); // Typical way of creating a tuple struct. let c2 = Color{0: 255, 1: 127, 2: 0}; // Specifying fields by index. let c3 = Color{1: 0, ..c2}; // Fill out all other fields using a base struct. }
Struct field init shorthand
When initializing a data structure (struct, enum, union) with named (but not numbered) fields, it is allowed to write fieldname
as a shorthand for fieldname: fieldname
.
This allows a compact syntax with less duplication.
For example:
#![allow(unused)] fn main() { struct Point3d { x: i32, y: i32, z: i32 } let x = 0; let y_value = 0; let z = 0; Point3d { x: x, y: y_value, z: z }; Point3d { x, y: y_value, z }; }
Tuple struct expression
A struct expression with fields enclosed in parentheses constructs a tuple struct. Though it is listed here as a specific expression for completeness, it is equivalent to a call expression to the tuple struct’s constructor. For example:
#![allow(unused)] fn main() { struct Position(i32, i32, i32); Position(0, 0, 0); // Typical way of creating a tuple struct. let c = Position; // `c` is a function that takes 3 arguments. let pos = c(8, 6, 7); // Creates a `Position` value. }
Unit struct expression
A unit struct expression is just the path to a unit struct item. This refers to the unit struct’s implicit constant of its value. The unit struct value can also be constructed with a fieldless struct expression. For example:
#![allow(unused)] fn main() { struct Gamma; let a = Gamma; // Gamma unit value. let b = Gamma{}; // Exact same value as `a`. }
Call expressions
Syntax
CallExpression :
Expression(
CallParams?)
CallParams :
Expression (,
Expression )*,
?
A call expression calls a function.
The syntax of a call expression is an expression, called the function operand, followed by a parenthesized comma-separated list of expression, called the argument operands.
If the function eventually returns, then the expression completes.
For non-function types, the expression f(...)
uses the method on one of the std::ops::Fn
, std::ops::FnMut
or std::ops::FnOnce
traits, which differ in whether they take the type by reference, mutable reference, or take ownership respectively.
An automatic borrow will be taken if needed.
The function operand will also be automatically dereferenced as required.
Some examples of call expressions:
#![allow(unused)] fn main() { fn add(x: i32, y: i32) -> i32 { 0 } let three: i32 = add(1i32, 2i32); let name: &'static str = (|| "Rust")(); }
Disambiguating Function Calls
All function calls are sugar for a more explicit fully-qualified syntax. Function calls may need to be fully qualified, depending on the ambiguity of a call in light of in-scope items.
Note: In the past, the terms “Unambiguous Function Call Syntax”, “Universal Function Call Syntax”, or “UFCS”, have been used in documentation, issues, RFCs, and other community writings. However, these terms lack descriptive power and potentially confuse the issue at hand. We mention them here for searchability’s sake.
Several situations often occur which result in ambiguities about the receiver or referent of method or associated function calls. These situations may include:
- Multiple in-scope traits define methods with the same name for the same types
- Auto-
deref
is undesirable; for example, distinguishing between methods on a smart pointer itself and the pointer’s referent - Methods which take no arguments, like
default()
, and return properties of a type, likesize_of()
To resolve the ambiguity, the programmer may refer to their desired method or function using more specific paths, types, or traits.
For example,
trait Pretty { fn print(&self); } trait Ugly { fn print(&self); } struct Foo; impl Pretty for Foo { fn print(&self) {} } struct Bar; impl Pretty for Bar { fn print(&self) {} } impl Ugly for Bar { fn print(&self) {} } fn main() { let f = Foo; let b = Bar; // we can do this because we only have one item called `print` for `Foo`s f.print(); // more explicit, and, in the case of `Foo`, not necessary Foo::print(&f); // if you're not into the whole brevity thing <Foo as Pretty>::print(&f); // b.print(); // Error: multiple 'print' found // Bar::print(&b); // Still an error: multiple `print` found // necessary because of in-scope items defining `print` <Bar as Pretty>::print(&b); }
Refer to RFC 132 for further details and motivations.
Method-call expressions
Syntax
MethodCallExpression :
Expression.
PathExprSegment(
CallParams?)
A method call consists of an expression (the receiver) followed by a single dot, an expression path segment, and a parenthesized expression-list.
Method calls are resolved to associated methods on specific traits, either statically dispatching to a method if the exact self
-type of the left-hand-side is known, or dynamically dispatching if the left-hand-side expression is an indirect trait object.
#![allow(unused)] fn main() { let pi: Result<f32, _> = "3.14".parse(); let log_pi = pi.unwrap_or(1.0).log(2.72); assert!(1.14 < log_pi && log_pi < 1.15) }
When looking up a method call, the receiver may be automatically dereferenced or borrowed in order to call a method. This requires a more complex lookup process than for other functions, since there may be a number of possible methods to call. The following procedure is used:
The first step is to build a list of candidate receiver types.
Obtain these by repeatedly dereferencing the receiver expression’s type, adding each type encountered to the list, then finally attempting an unsized coercion at the end, and adding the result type if that is successful.
Then, for each candidate T
, add &T
and &mut T
to the list immediately after T
.
For instance, if the receiver has type Box<[i32;2]>
, then the candidate types will be Box<[i32;2]>
, &Box<[i32;2]>
, &mut Box<[i32;2]>
, [i32; 2]
(by dereferencing), &[i32; 2]
, &mut [i32; 2]
, [i32]
(by unsized coercion), &[i32]
, and finally &mut [i32]
.
Then, for each candidate type T
, search for a visible method with a receiver of that type in the following places:
T
’s inherent methods (methods implemented directly onT
).- Any of the methods provided by a visible trait implemented by
T
. IfT
is a type parameter, methods provided by trait bounds onT
are looked up first. Then all remaining methods in scope are looked up.
Note: the lookup is done for each type in order, which can occasionally lead to surprising results. The below code will print “In trait impl!”, because
&self
methods are looked up first, the trait method is found before the struct’s&mut self
method is found.struct Foo {} trait Bar { fn bar(&self); } impl Foo { fn bar(&mut self) { println!("In struct impl!") } } impl Bar for Foo { fn bar(&self) { println!("In trait impl!") } } fn main() { let mut f = Foo{}; f.bar(); }
If this results in multiple possible candidates, then it is an error, and the receiver must be converted to an appropriate receiver type to make the method call.
This process does not take into account the mutability or lifetime of the receiver, or whether a method is unsafe
.
Once a method is looked up, if it can’t be called for one (or more) of those reasons, the result is a compiler error.
If a step is reached where there is more than one possible method, such as where generic methods or traits are considered the same, then it is a compiler error. These cases require a disambiguating function call syntax for method and function invocation.
Edition differences: Before the 2021 edition, during the search for visible methods, if the candidate receiver type is an array type, methods provided by the standard library
IntoIterator
trait are ignored.The edition used for this purpose is determined by the token representing the method name.
This special case may be removed in the future.
Warning: For trait objects, if there is an inherent method of the same name as a trait method, it will give a compiler error when trying to call the method in a method call expression. Instead, you can call the method using disambiguating function call syntax, in which case it calls the trait method, not the inherent method. There is no way to call the inherent method. Just don’t define inherent methods on trait objects with the same name as a trait method and you’ll be fine.
Field access expressions
Syntax
FieldExpression :
Expression.
IDENTIFIER
A field expression is a place expression that evaluates to the location of a field of a struct or union. When the operand is mutable, the field expression is also mutable.
The syntax for a field expression is an expression, called the container operand, then a .
, and finally an identifier.
Field expressions cannot be followed by a parenthetical comma-separated list of expressions, as that is instead parsed as a method call expression.
That is, they cannot be the function operand of a call expression.
Note: Wrap the field expression in a parenthesized expression to use it in a call expression.
#![allow(unused)] fn main() { struct HoldsCallable<F: Fn()> { callable: F } let holds_callable = HoldsCallable { callable: || () }; // Invalid: Parsed as calling the method "callable" // holds_callable.callable(); // Valid (holds_callable.callable)(); }
Examples:
mystruct.myfield;
foo().x;
(Struct {a: 10, b: 20}).a;
(mystruct.function_field)() // Call expression containing a field expression
Automatic dereferencing
If the type of the container operand implements Deref
or DerefMut
depending on whether the operand is mutable, it is automatically dereferenced as many times as necessary to make the field access possible.
This process is also called autoderef for short.
Borrowing
The fields of a struct or a reference to a struct are treated as separate entities when borrowing.
If the struct does not implement Drop
and is stored in a local variable, this also applies to moving out of each of its fields.
This also does not apply if automatic dereferencing is done though user-defined types other than Box
.
#![allow(unused)] fn main() { struct A { f1: String, f2: String, f3: String } let mut x: A; x = A { f1: "f1".to_string(), f2: "f2".to_string(), f3: "f3".to_string() }; let a: &mut String = &mut x.f1; // x.f1 borrowed mutably let b: &String = &x.f2; // x.f2 borrowed immutably let c: &String = &x.f2; // Can borrow again let d: String = x.f3; // Move out of x.f3 }
Closure expressions
Syntax
ClosureExpression :
move
?
(||
||
ClosureParameters?|
)
(Expression |->
TypeNoBounds BlockExpression)ClosureParameters :
ClosureParam (,
ClosureParam)*,
?ClosureParam :
OuterAttribute* PatternNoTopAlt (:
Type )?
A closure expression, also known as a lambda expression or a lambda, defines a closure type and evaluates to a value of that type.
The syntax for a closure expression is an optional move
keyword, then a pipe-symbol-delimited (|
) comma-separated list of patterns, called the closure parameters each optionally followed by a :
and a type, then an optional ->
and type, called the return type, and then an expression, called the closure body operand.
The optional type after each pattern is a type annotation for the pattern.
If there is a return type, the closure body must be a block.
A closure expression denotes a function that maps a list of parameters onto the expression that follows the parameters.
Just like a let
binding, the closure parameters are irrefutable patterns, whose type annotation is optional and will be inferred from context if not given.
Each closure expression has a unique, anonymous type.
Significantly, closure expressions capture their environment, which regular function definitions do not.
Without the move
keyword, the closure expression infers how it captures each variable from its environment, preferring to capture by shared reference, effectively borrowing all outer variables mentioned inside the closure’s body.
If needed the compiler will infer that instead mutable references should be taken, or that the values should be moved or copied (depending on their type) from the environment.
A closure can be forced to capture its environment by copying or moving values by prefixing it with the move
keyword.
This is often used to ensure that the closure’s lifetime is 'static
.
Closure trait implementations
Which traits the closure type implement depends on how variables are captured and the types of the captured variables.
See the call traits and coercions chapter for how and when a closure implements Fn
, FnMut
, and FnOnce
.
The closure type implements Send
and Sync
if the type of every captured variable also implements the trait.
Example
In this example, we define a function ten_times
that takes a higher-order function argument, and we then call it with a closure expression as an argument, followed by a closure expression that moves values from its environment.
#![allow(unused)] fn main() { fn ten_times<F>(f: F) where F: Fn(i32) { for index in 0..10 { f(index); } } ten_times(|j| println!("hello, {}", j)); // With type annotations ten_times(|j: i32| -> () { println!("hello, {}", j) }); let word = "konnichiwa".to_owned(); ten_times(move |j| println!("{}, {}", word, j)); }
Attributes on closure parameters
Attributes on closure parameters follow the same rules and restrictions as regular function parameters.
Loops and other breakable expressions
Syntax
LoopExpression :
LoopLabel? (
InfiniteLoopExpression
| PredicateLoopExpression
| PredicatePatternLoopExpression
| IteratorLoopExpression
| LabelBlockExpression
)
Rust supports five loop expressions:
- A
loop
expression denotes an infinite loop. - A
while
expression loops until a predicate is false. - A
while let
expression tests a pattern. - A
for
expression extracts values from an iterator, looping until the iterator is empty. - A labelled block expression runs a loop exactly once, but allows exiting the loop early with
break
.
All five types of loop support break
expressions, and labels.
All except labelled block expressions support continue
expressions.
Only loop
and labelled block expressions support evaluation to non-trivial values.
Infinite loops
Syntax
InfiniteLoopExpression :
loop
BlockExpression
A loop
expression repeats execution of its body continuously:
loop { println!("I live."); }
.
A loop
expression without an associated break
expression is diverging and has type !
.
A loop
expression containing associated break
expression(s) may terminate, and must have type compatible with the value of the break
expression(s).
Predicate loops
Syntax
PredicateLoopExpression :
while
Expressionexcept struct expression BlockExpression
A while
loop begins by evaluating the boolean loop conditional operand.
If the loop conditional operand evaluates to true
, the loop body block executes, then control returns to the loop conditional operand.
If the loop conditional expression evaluates to false
, the while
expression completes.
An example:
#![allow(unused)] fn main() { let mut i = 0; while i < 10 { println!("hello"); i = i + 1; } }
Predicate pattern loops
Syntax
PredicatePatternLoopExpression :
while
let
Pattern=
Scrutineeexcept lazy boolean operator expression BlockExpression
A while let
loop is semantically similar to a while
loop but in place of a condition expression it expects the keyword let
followed by a pattern, an =
, a scrutinee expression and a block expression.
If the value of the scrutinee matches the pattern, the loop body block executes then control returns to the pattern matching statement.
Otherwise, the while expression completes.
#![allow(unused)] fn main() { let mut x = vec![1, 2, 3]; while let Some(y) = x.pop() { println!("y = {}", y); } while let _ = 5 { println!("Irrefutable patterns are always true"); break; } }
A while let
loop is equivalent to a loop
expression containing a match
expression as follows.
'label: while let PATS = EXPR {
/* loop body */
}
is equivalent to
'label: loop {
match EXPR {
PATS => { /* loop body */ },
_ => break,
}
}
Multiple patterns may be specified with the |
operator.
This has the same semantics as with |
in match
expressions:
#![allow(unused)] fn main() { let mut vals = vec![2, 3, 1, 2, 2]; while let Some(v @ 1) | Some(v @ 2) = vals.pop() { // Prints 2, 2, then 1 println!("{}", v); } }
As is the case in if let
expressions, the scrutinee cannot be a lazy boolean operator expression.
Iterator loops
Syntax
IteratorLoopExpression :
for
Patternin
Expressionexcept struct expression BlockExpression
A for
expression is a syntactic construct for looping over elements provided by an implementation of std::iter::IntoIterator
.
If the iterator yields a value, that value is matched against the irrefutable pattern, the body of the loop is executed, and then control returns to the head of the for
loop.
If the iterator is empty, the for
expression completes.
An example of a for
loop over the contents of an array:
#![allow(unused)] fn main() { let v = &["apples", "cake", "coffee"]; for text in v { println!("I like {}.", text); } }
An example of a for loop over a series of integers:
#![allow(unused)] fn main() { let mut sum = 0; for n in 1..11 { sum += n; } assert_eq!(sum, 55); }
A for
loop is equivalent to a loop
expression containing a match
expression as follows:
'label: for PATTERN in iter_expr {
/* loop body */
}
is equivalent to
{
let result = match IntoIterator::into_iter(iter_expr) {
mut iter => 'label: loop {
let mut next;
match Iterator::next(&mut iter) {
Option::Some(val) => next = val,
Option::None => break,
};
let PATTERN = next;
let () = { /* loop body */ };
},
};
result
}
IntoIterator
, Iterator
, and Option
are always the standard library items here, not whatever those names resolve to in the current scope.
The variable names next
, iter
, and val
are for exposition only, they do not actually have names the user can type.
Note: that the outer
match
is used to ensure that any temporary values initer_expr
don’t get dropped before the loop is finished.next
is declared before being assigned because it results in types being inferred correctly more often.
Loop labels
Syntax
LoopLabel :
LIFETIME_OR_LABEL:
A loop expression may optionally have a label. The label is written as a lifetime preceding the loop expression, as in 'foo: loop { break 'foo; }
, 'bar: while false {}
, 'humbug: for _ in 0..0 {}
.
If a label is present, then labeled break
and continue
expressions nested within this loop may exit out of this loop or return control to its head.
See break expressions and continue expressions.
Labels follow the hygiene and shadowing rules of local variables. For example, this code will print “outer loop”:
#![allow(unused)] fn main() { 'a: loop { 'a: loop { break 'a; } print!("outer loop"); break 'a; } }
'_
is not a valid loop label.
break
expressions
Syntax
BreakExpression :
break
LIFETIME_OR_LABEL? Expression?
When break
is encountered, execution of the associated loop body is immediately terminated, for example:
#![allow(unused)] fn main() { let mut last = 0; for x in 1..100 { if x > 12 { break; } last = x; } assert_eq!(last, 12); }
A break
expression is normally associated with the innermost loop
, for
or while
loop enclosing the break
expression,
but a label can be used to specify which enclosing loop is affected.
Example:
#![allow(unused)] fn main() { 'outer: loop { while true { break 'outer; } } }
A break
expression is only permitted in the body of a loop, and has one of the forms break
, break 'label
or (see below) break EXPR
or break 'label EXPR
.
Labelled block expressions
Syntax
LabelBlockExpression :
BlockExpression
Labelled block expressions are exactly like block expressions, except that they allow using break
expressions within the block.
Unlike loops, break
expressions within a labelled block expression must have a label (i.e. the label is not optional).
Similarly, labelled block expressions must begin with a label.
#![allow(unused)] fn main() { fn do_thing() {} fn condition_not_met() -> bool { true } fn do_next_thing() {} fn do_last_thing() {} let result = 'block: { do_thing(); if condition_not_met() { break 'block 1; } do_next_thing(); if condition_not_met() { break 'block 2; } do_last_thing(); 3 }; }
continue
expressions
Syntax
ContinueExpression :
continue
LIFETIME_OR_LABEL?
When continue
is encountered, the current iteration of the associated loop body is immediately terminated, returning control to the loop head.
In the case of a while
loop, the head is the conditional expression controlling the loop.
In the case of a for
loop, the head is the call-expression controlling the loop.
Like break
, continue
is normally associated with the innermost enclosing loop, but continue 'label
may be used to specify the loop affected.
A continue
expression is only permitted in the body of a loop.
break
and loop values
When associated with a loop
, a break expression may be used to return a value from that loop, via one of the forms break EXPR
or break 'label EXPR
, where EXPR
is an expression whose result is returned from the loop
.
For example:
#![allow(unused)] fn main() { let (mut a, mut b) = (1, 1); let result = loop { if b > 10 { break b; } let c = a + b; a = b; b = c; }; // first number in Fibonacci sequence over 10: assert_eq!(result, 13); }
In the case a loop
has an associated break
, it is not considered diverging, and the loop
must have a type compatible with each break
expression.
break
without an expression is considered identical to break
with expression ()
.
Range expressions
Syntax
RangeExpression :
RangeExpr
| RangeFromExpr
| RangeToExpr
| RangeFullExpr
| RangeInclusiveExpr
| RangeToInclusiveExprRangeExpr :
Expression..
ExpressionRangeFromExpr :
Expression..
RangeToExpr :
..
ExpressionRangeFullExpr :
..
RangeInclusiveExpr :
Expression..=
ExpressionRangeToInclusiveExpr :
..=
Expression
The ..
and ..=
operators will construct an object of one of the std::ops::Range
(or core::ops::Range
) variants, according to the following table:
Production | Syntax | Type | Range |
---|---|---|---|
RangeExpr | start.. end | std::ops::Range | start ≤ x < end |
RangeFromExpr | start.. | std::ops::RangeFrom | start ≤ x |
RangeToExpr | .. end | std::ops::RangeTo | x < end |
RangeFullExpr | .. | std::ops::RangeFull | - |
RangeInclusiveExpr | start..= end | std::ops::RangeInclusive | start ≤ x ≤ end |
RangeToInclusiveExpr | ..= end | std::ops::RangeToInclusive | x ≤ end |
Examples:
#![allow(unused)] fn main() { 1..2; // std::ops::Range 3..; // std::ops::RangeFrom ..4; // std::ops::RangeTo ..; // std::ops::RangeFull 5..=6; // std::ops::RangeInclusive ..=7; // std::ops::RangeToInclusive }
The following expressions are equivalent.
#![allow(unused)] fn main() { let x = std::ops::Range {start: 0, end: 10}; let y = 0..10; assert_eq!(x, y); }
Ranges can be used in for
loops:
#![allow(unused)] fn main() { for i in 1..11 { println!("{}", i); } }
if
and if let
expressions
if
expressions
Syntax
IfExpression :
if
Expressionexcept struct expression BlockExpression
(else
( BlockExpression | IfExpression | IfLetExpression ) )?
An if
expression is a conditional branch in program control.
The syntax of an if
expression is a condition operand, followed by a consequent block, any number of else if
conditions and blocks, and an optional trailing else
block.
The condition operands must have the boolean type.
If a condition operand evaluates to true
, the consequent block is executed and any subsequent else if
or else
block is skipped.
If a condition operand evaluates to false
, the consequent block is skipped and any subsequent else if
condition is evaluated.
If all if
and else if
conditions evaluate to false
then any else
block is executed.
An if expression evaluates to the same value as the executed block, or ()
if no block is evaluated.
An if
expression must have the same type in all situations.
#![allow(unused)] fn main() { let x = 3; if x == 4 { println!("x is four"); } else if x == 3 { println!("x is three"); } else { println!("x is something else"); } let y = if 12 * 15 > 150 { "Bigger" } else { "Smaller" }; assert_eq!(y, "Bigger"); }
if let
expressions
Syntax
IfLetExpression :
if
let
Pattern=
Scrutineeexcept lazy boolean operator expression BlockExpression
(else
( BlockExpression | IfExpression | IfLetExpression ) )?
An if let
expression is semantically similar to an if
expression but in place of a condition operand it expects the keyword let
followed by a pattern, an =
and a scrutinee operand.
If the value of the scrutinee matches the pattern, the corresponding block will execute.
Otherwise, flow proceeds to the following else
block if it exists.
Like if
expressions, if let
expressions have a value determined by the block that is evaluated.
#![allow(unused)] fn main() { let dish = ("Ham", "Eggs"); // this body will be skipped because the pattern is refuted if let ("Bacon", b) = dish { println!("Bacon is served with {}", b); } else { // This block is evaluated instead. println!("No bacon will be served"); } // this body will execute if let ("Ham", b) = dish { println!("Ham is served with {}", b); } if let _ = 5 { println!("Irrefutable patterns are always true"); } }
if
and if let
expressions can be intermixed:
#![allow(unused)] fn main() { let x = Some(3); let a = if let Some(1) = x { 1 } else if x == Some(2) { 2 } else if let Some(y) = x { y } else { -1 }; assert_eq!(a, 3); }
An if let
expression is equivalent to a match
expression as follows:
if let PATS = EXPR {
/* body */
} else {
/*else */
}
is equivalent to
match EXPR {
PATS => { /* body */ },
_ => { /* else */ }, // () if there is no else
}
Multiple patterns may be specified with the |
operator. This has the same semantics as with |
in match
expressions:
#![allow(unused)] fn main() { enum E { X(u8), Y(u8), Z(u8), } let v = E::Y(12); if let E::X(n) | E::Y(n) = v { assert_eq!(n, 12); } }
The expression cannot be a lazy boolean operator expression. Use of a lazy boolean operator is ambiguous with a planned feature change of the language (the implementation of if-let chains - see eRFC 2947). When lazy boolean operator expression is desired, this can be achieved by using parenthesis as below:
// Before...
if let PAT = EXPR && EXPR { .. }
// After...
if let PAT = ( EXPR && EXPR ) { .. }
// Before...
if let PAT = EXPR || EXPR { .. }
// After...
if let PAT = ( EXPR || EXPR ) { .. }
match
expressions
Syntax
MatchExpression :
match
Scrutinee{
InnerAttribute*
MatchArms?
}
Scrutinee :
Expressionexcept struct expressionMatchArms :
( MatchArm=>
( ExpressionWithoutBlock,
| ExpressionWithBlock,
? ) )*
MatchArm=>
Expression,
?MatchArm :
OuterAttribute* Pattern MatchArmGuard?MatchArmGuard :
if
Expression
A match
expression branches on a pattern.
The exact form of matching that occurs depends on the pattern.
A match
expression has a scrutinee expression, which is the value to compare to the patterns.
The scrutinee expression and the patterns must have the same type.
A match
behaves differently depending on whether or not the scrutinee expression is a place expression or value expression.
If the scrutinee expression is a value expression, it is first evaluated into a temporary location, and the resulting value is sequentially compared to the patterns in the arms until a match is found.
The first arm with a matching pattern is chosen as the branch target of the match
, any variables bound by the pattern are assigned to local variables in the arm’s block, and control enters the block.
When the scrutinee expression is a place expression, the match does not allocate a temporary location; however, a by-value binding may copy or move from the memory location. When possible, it is preferable to match on place expressions, as the lifetime of these matches inherits the lifetime of the place expression rather than being restricted to the inside of the match.
An example of a match
expression:
#![allow(unused)] fn main() { let x = 1; match x { 1 => println!("one"), 2 => println!("two"), 3 => println!("three"), 4 => println!("four"), 5 => println!("five"), _ => println!("something else"), } }
Variables bound within the pattern are scoped to the match guard and the arm’s expression. The binding mode (move, copy, or reference) depends on the pattern.
Multiple match patterns may be joined with the |
operator.
Each pattern will be tested in left-to-right sequence until a successful match is found.
#![allow(unused)] fn main() { let x = 9; let message = match x { 0 | 1 => "not many", 2 ..= 9 => "a few", _ => "lots" }; assert_eq!(message, "a few"); // Demonstration of pattern match order. struct S(i32, i32); match S(1, 2) { S(z @ 1, _) | S(_, z @ 2) => assert_eq!(z, 1), _ => panic!(), } }
Note: The
2..=9
is a Range Pattern, not a Range Expression. Thus, only those types of ranges supported by range patterns can be used in match arms.
Every binding in each |
separated pattern must appear in all of the patterns in the arm.
Every binding of the same name must have the same type, and have the same binding mode.
Match guards
Match arms can accept match guards to further refine the criteria for matching a case.
Pattern guards appear after the pattern and consist of a bool
-typed expression following the if
keyword.
When the pattern matches successfully, the pattern guard expression is executed.
If the expression evaluates to true, the pattern is successfully matched against.
Otherwise, the next pattern, including other matches with the |
operator in the same arm, is tested.
#![allow(unused)] fn main() { let maybe_digit = Some(0); fn process_digit(i: i32) { } fn process_other(i: i32) { } let message = match maybe_digit { Some(x) if x < 10 => process_digit(x), Some(x) => process_other(x), None => panic!(), }; }
Note: Multiple matches using the
|
operator can cause the pattern guard and the side effects it has to execute multiple times. For example:#![allow(unused)] fn main() { use std::cell::Cell; let i : Cell<i32> = Cell::new(0); match 1 { 1 | _ if { i.set(i.get() + 1); false } => {} _ => {} } assert_eq!(i.get(), 2); }
A pattern guard may refer to the variables bound within the pattern they follow. Before evaluating the guard, a shared reference is taken to the part of the scrutinee the variable matches on. While evaluating the guard, this shared reference is then used when accessing the variable. Only when the guard evaluates to true is the value moved, or copied, from the scrutinee into the variable. This allows shared borrows to be used inside guards without moving out of the scrutinee in case guard fails to match. Moreover, by holding a shared reference while evaluating the guard, mutation inside guards is also prevented.
Attributes on match arms
Outer attributes are allowed on match arms.
The only attributes that have meaning on match arms are cfg
and the lint check attributes.
Inner attributes are allowed directly after the opening brace of the match expression in the same expression contexts as attributes on block expressions.
return
expressions
Syntax
ReturnExpression :
return
Expression?
Return expressions are denoted with the keyword return
.
Evaluating a return
expression moves its argument into the designated output location for the current function call, destroys the current function activation frame, and transfers control to the caller frame.
An example of a return
expression:
#![allow(unused)] fn main() { fn max(a: i32, b: i32) -> i32 { if a > b { return a; } return b; } }
Await expressions
Syntax
AwaitExpression :
Expression.
await
An await
expression is a syntactic construct for suspending a computation
provided by an implementation of std::future::IntoFuture
until the given
future is ready to produce a value.
The syntax for an await expression is an expression with a type that implements the IntoFuture
trait, called the future operand, then the token .
, and then the await
keyword.
Await expressions are legal only within an async context, like an async fn
or an async
block.
More specifically, an await expression has the following effect.
- Create a future by calling
IntoFuture::into_future
on the future operand. - Evaluate the future to a future
tmp
; - Pin
tmp
usingPin::new_unchecked
; - This pinned future is then polled by calling the
Future::poll
method and passing it the current task context; - If the call to
poll
returnsPoll::Pending
, then the future returnsPoll::Pending
, suspending its state so that, when the surrounding async context is re-polled,execution returns to step 3; - Otherwise the call to
poll
must have returnedPoll::Ready
, in which case the value contained in thePoll::Ready
variant is used as the result of theawait
expression itself.
Edition differences: Await expressions are only available beginning with Rust 2018.
Task context
The task context refers to the Context
which was supplied to the current async context when the async context itself was polled.
Because await
expressions are only legal in an async context, there must be some task context available.
Approximate desugaring
Effectively, an await expression is roughly equivalent to the following non-normative desugaring:
match operand.into_future() {
mut pinned => loop {
let mut pin = unsafe { Pin::new_unchecked(&mut pinned) };
match Pin::future::poll(Pin::borrow(&mut pin), &mut current_context) {
Poll::Ready(r) => break r,
Poll::Pending => yield Poll::Pending,
}
}
}
where the yield
pseudo-code returns Poll::Pending
and, when re-invoked, resumes execution from that point.
The variable current_context
refers to the context taken from the async environment.
_
expressions
Syntax
UnderscoreExpression :
_
Underscore expressions, denoted with the symbol _
, are used to signify a
placeholder in a destructuring assignment. They may only appear in the left-hand
side of an assignment.
Note that this is distinct from the wildcard pattern.
Examples of _
expressions:
#![allow(unused)] fn main() { let p = (1, 2); let mut a = 0; (_, a) = p; struct Position { x: u32, y: u32, } Position { x: a, y: _ } = Position{ x: 2, y: 3 }; // unused result, assignment to `_` used to declare intent and remove a warning _ = 2 + 2; // triggers unused_must_use warning // 2 + 2; // equivalent technique using a wildcard pattern in a let-binding let _ = 2 + 2; }
Patterns
Syntax
Pattern :
|
? PatternNoTopAlt (|
PatternNoTopAlt )*PatternNoTopAlt :
PatternWithoutRange
| RangePatternPatternWithoutRange :
LiteralPattern
| IdentifierPattern
| WildcardPattern
| RestPattern
| ReferencePattern
| StructPattern
| TupleStructPattern
| TuplePattern
| GroupedPattern
| SlicePattern
| PathPattern
| MacroInvocation
Patterns are used to match values against structures and to, optionally, bind variables to values inside these structures. They are also used in variable declarations and parameters for functions and closures.
The pattern in the following example does four things:
- Tests if
person
has thecar
field filled with something. - Tests if the person’s
age
field is between 13 and 19, and binds its value to theperson_age
variable. - Binds a reference to the
name
field to the variableperson_name
. - Ignores the rest of the fields of
person
. The remaining fields can have any value and are not bound to any variables.
#![allow(unused)] fn main() { struct Car; struct Computer; struct Person { name: String, car: Option<Car>, computer: Option<Computer>, age: u8, } let person = Person { name: String::from("John"), car: Some(Car), computer: None, age: 15, }; if let Person { car: Some(_), age: person_age @ 13..=19, name: ref person_name, .. } = person { println!("{} has a car and is {} years old.", person_name, person_age); } }
Patterns are used in:
let
declarations- Function and closure parameters
match
expressionsif let
expressionswhile let
expressionsfor
expressions
Destructuring
Patterns can be used to destructure structs, enums, and tuples.
Destructuring breaks up a value into its component pieces.
The syntax used is almost the same as when creating such values.
In a pattern whose scrutinee expression has a struct
, enum
or tuple
type, a placeholder (_
) stands in for a single data field, whereas a wildcard ..
stands in for all the remaining fields of a particular variant.
When destructuring a data structure with named (but not numbered) fields, it is allowed to write fieldname
as a shorthand for fieldname: fieldname
.
#![allow(unused)] fn main() { enum Message { Quit, WriteString(String), Move { x: i32, y: i32 }, ChangeColor(u8, u8, u8), } let message = Message::Quit; match message { Message::Quit => println!("Quit"), Message::WriteString(write) => println!("{}", &write), Message::Move{ x, y: 0 } => println!("move {} horizontally", x), Message::Move{ .. } => println!("other move"), Message::ChangeColor { 0: red, 1: green, 2: _ } => { println!("color change, red: {}, green: {}", red, green); } }; }
Refutability
A pattern is said to be refutable when it has the possibility of not being matched by the value it is being matched against. Irrefutable patterns, on the other hand, always match the value they are being matched against. Examples:
#![allow(unused)] fn main() { let (x, y) = (1, 2); // "(x, y)" is an irrefutable pattern if let (a, 3) = (1, 2) { // "(a, 3)" is refutable, and will not match panic!("Shouldn't reach here"); } else if let (a, 4) = (3, 4) { // "(a, 4)" is refutable, and will match println!("Matched ({}, 4)", a); } }
Literal patterns
Syntax
LiteralPattern :
true
|false
| CHAR_LITERAL
| BYTE_LITERAL
| STRING_LITERAL
| RAW_STRING_LITERAL
| BYTE_STRING_LITERAL
| RAW_BYTE_STRING_LITERAL
| C_STRING_LITERAL
| RAW_C_STRING_LITERAL
|-
? INTEGER_LITERAL
|-
? FLOAT_LITERAL
Literal patterns match exactly the same value as what is created by the literal. Since negative numbers are not literals, literal patterns also accept an optional minus sign before the literal, which acts like the negation operator.
C string and raw C string literals are accepted in literal patterns, but &CStr
doesn’t implement structural equality (#[derive(Eq, PartialEq)]
) and therefore
any such match
on a &CStr
will be rejected with a type error.
Literal patterns are always refutable.
Examples:
#![allow(unused)] fn main() { for i in -2..5 { match i { -1 => println!("It's minus one"), 1 => println!("It's a one"), 2|4 => println!("It's either a two or a four"), _ => println!("Matched none of the arms"), } } }
Identifier patterns
Syntax
IdentifierPattern :
ref
?mut
? IDENTIFIER (@
PatternNoTopAlt ) ?
Identifier patterns bind the value they match to a variable in the value namespace.
The identifier must be unique within the pattern.
The variable will shadow any variables of the same name in scope.
The scope of the new binding depends on the context of where the pattern is used (such as a let
binding or a match
arm).
Patterns that consist of only an identifier, possibly with a mut
, match any value and bind it to that identifier.
This is the most commonly used pattern in variable declarations and parameters for functions and closures.
#![allow(unused)] fn main() { let mut variable = 10; fn sum(x: i32, y: i32) -> i32 { x + y } }
To bind the matched value of a pattern to a variable, use the syntax variable @ subpattern
.
For example, the following binds the value 2 to e
(not the entire range: the range here is a range subpattern).
#![allow(unused)] fn main() { let x = 2; match x { e @ 1 ..= 5 => println!("got a range element {}", e), _ => println!("anything"), } }
By default, identifier patterns bind a variable to a copy of or move from the matched value depending on whether the matched value implements Copy
.
This can be changed to bind to a reference by using the ref
keyword, or to a mutable reference using ref mut
. For example:
#![allow(unused)] fn main() { let a = Some(10); match a { None => (), Some(value) => (), } match a { None => (), Some(ref value) => (), } }
In the first match expression, the value is copied (or moved).
In the second match, a reference to the same memory location is bound to the variable value.
This syntax is needed because in destructuring subpatterns the &
operator can’t be applied to the value’s fields.
For example, the following is not valid:
#![allow(unused)] fn main() { struct Person { name: String, age: u8, } let value = Person { name: String::from("John"), age: 23 }; if let Person { name: &person_name, age: 18..=150 } = value { } }
To make it valid, write the following:
#![allow(unused)] fn main() { struct Person { name: String, age: u8, } let value = Person { name: String::from("John"), age: 23 }; if let Person {name: ref person_name, age: 18..=150 } = value { } }
Thus, ref
is not something that is being matched against.
Its objective is exclusively to make the matched binding a reference, instead of potentially copying or moving what was matched.
Path patterns take precedence over identifier patterns.
It is an error if ref
or ref mut
is specified and the identifier shadows a constant.
Identifier patterns are irrefutable if the @
subpattern is irrefutable or the subpattern is not specified.
Binding modes
To service better ergonomics, patterns operate in different binding modes in order to make it easier to bind references to values.
When a reference value is matched by a non-reference pattern, it will be automatically treated as a ref
or ref mut
binding.
Example:
#![allow(unused)] fn main() { let x: &Option<i32> = &Some(3); if let Some(y) = x { // y was converted to `ref y` and its type is &i32 } }
Non-reference patterns include all patterns except bindings, wildcard patterns (_
), const
patterns of reference types, and reference patterns.
If a binding pattern does not explicitly have ref
, ref mut
, or mut
, then it uses the default binding mode to determine how the variable is bound.
The default binding mode starts in “move” mode which uses move semantics.
When matching a pattern, the compiler starts from the outside of the pattern and works inwards.
Each time a reference is matched using a non-reference pattern, it will automatically dereference the value and update the default binding mode.
References will set the default binding mode to ref
.
Mutable references will set the mode to ref mut
unless the mode is already ref
in which case it remains ref
.
If the automatically dereferenced value is still a reference, it is dereferenced and this process repeats.
Move bindings and reference bindings can be mixed together in the same pattern. Doing so will result in partial move of the object bound to and the object cannot be used afterwards. This applies only if the type cannot be copied.
In the example below, name
is moved out of person
.
Trying to use person
as a whole or person.name
would result in an error because of partial move.
Example:
#![allow(unused)] fn main() { struct Person { name: String, age: u8, } let person = Person{ name: String::from("John"), age: 23 }; // `name` is moved from person and `age` referenced let Person { name, ref age } = person; }
Wildcard pattern
Syntax
WildcardPattern :
_
The wildcard pattern (an underscore symbol) matches any value.
It is used to ignore values when they don’t matter.
Inside other patterns it matches a single data field (as opposed to the ..
which matches the remaining fields).
Unlike identifier patterns, it does not copy, move or borrow the value it matches.
Examples:
#![allow(unused)] fn main() { let x = 20; let (a, _) = (10, x); // the x is always matched by _ assert_eq!(a, 10); // ignore a function/closure param let real_part = |a: f64, _: f64| { a }; // ignore a field from a struct struct RGBA { r: f32, g: f32, b: f32, a: f32, } let color = RGBA{r: 0.4, g: 0.1, b: 0.9, a: 0.5}; let RGBA{r: red, g: green, b: blue, a: _} = color; assert_eq!(color.r, red); assert_eq!(color.g, green); assert_eq!(color.b, blue); // accept any Some, with any value let x = Some(10); if let Some(_) = x {} }
The wildcard pattern is always irrefutable.
Rest patterns
Syntax
RestPattern :
..
The rest pattern (the ..
token) acts as a variable-length pattern which matches zero or more elements that haven’t been matched already before and after.
It may only be used in tuple, tuple struct, and slice patterns, and may only appear once as one of the elements in those patterns.
It is also allowed in an identifier pattern for slice patterns only.
The rest pattern is always irrefutable.
Examples:
#![allow(unused)] fn main() { let words = vec!["a", "b", "c"]; let slice = &words[..]; match slice { [] => println!("slice is empty"), [one] => println!("single element {}", one), [head, tail @ ..] => println!("head={} tail={:?}", head, tail), } match slice { // Ignore everything but the last element, which must be "!". [.., "!"] => println!("!!!"), // `start` is a slice of everything except the last element, which must be "z". [start @ .., "z"] => println!("starts with: {:?}", start), // `end` is a slice of everything but the first element, which must be "a". ["a", end @ ..] => println!("ends with: {:?}", end), // 'whole' is the entire slice and `last` is the final element whole @ [.., last] => println!("the last element of {:?} is {}", whole, last), rest => println!("{:?}", rest), } if let [.., penultimate, _] = slice { println!("next to last is {}", penultimate); } let tuple = (1, 2, 3, 4, 5); // Rest patterns may also be used in tuple and tuple struct patterns. match tuple { (1, .., y, z) => println!("y={} z={}", y, z), (.., 5) => println!("tail must be 5"), (..) => println!("matches everything else"), } }
Range patterns
Syntax
RangePattern :
RangeInclusivePattern
| RangeFromPattern
| RangeToInclusivePattern
| ObsoleteRangePatternRangeExclusivePattern :
RangePatternBound..
RangePatternBoundRangeInclusivePattern :
RangePatternBound..=
RangePatternBoundRangeFromPattern :
RangePatternBound..
RangeToInclusivePattern :
..=
RangePatternBoundObsoleteRangePattern :
RangePatternBound...
RangePatternBoundRangePatternBound :
CHAR_LITERAL
| BYTE_LITERAL
|-
? INTEGER_LITERAL
|-
? FLOAT_LITERAL
| PathExpression
Range patterns match scalar values within the range defined by their bounds.
They comprise a sigil (one of ..
, ..=
, or ...
) and a bound on one or both sides.
A bound on the left of the sigil is a lower bound.
A bound on the right is an upper bound.
A range pattern with both a lower and upper bound will match all values between and including both of its bounds.
It is written as its lower bound, followed by ..
for end-exclusive or ..=
for end-inclusive, followed by its upper bound.
The type of the range pattern is the type unification of its upper and lower bounds.
For example, a pattern 'm'..='p'
will match only the values 'm'
, 'n'
, 'o'
, and 'p'
.
Similarly, 'm'..'p'
will match only 'm'
, 'n'
and 'o'
, specifically not including 'p'
.
The lower bound cannot be greater than the upper bound.
That is, in a..=b
, a ≤ b must be the case.
For example, it is an error to have a range pattern 10..=0
.
A range pattern with only a lower bound will match any value greater than or equal to the lower bound.
It is written as its lower bound followed by ..
, and has the same type as its lower bound.
For example, 1..
will match 1, 9, or 9001, or 9007199254740991 (if it is of an appropriate size), but not 0, and not negative numbers for signed integers.
A range pattern with only an upper bound matches any value less than or equal to the upper bound.
It is written as ..=
followed by its upper bound, and has the same type as its upper bound.
For example, ..=10
will match 10, 1, 0, and for signed integer types, all negative values.
Range patterns with only one bound cannot be used as the top-level pattern for subpatterns in slice patterns.
The bounds is written as one of:
- A character, byte, integer, or float literal.
- A
-
followed by an integer or float literal. - A path
If the bounds is written as a path, after macro resolution, the path must resolve to a constant item of the type char
, an integer type, or a float type.
The type and value of the bounds is dependent upon how it is written out.
If the bounds is a path, the pattern has the type and value of the constant the path resolves to.
For float range patterns, the constant may not be a NaN
.
If it is a literal, it has the type and value of the corresponding literal expression.
If is a literal preceded by a -
, it has the same type as the corresponding literal expression and the value of negating the value of the corresponding literal expression.
Examples:
#![allow(unused)] fn main() { let c = 'f'; let valid_variable = match c { 'a'..='z' => true, 'A'..='Z' => true, 'α'..='ω' => true, _ => false, }; let ph = 10; println!("{}", match ph { 0..7 => "acid", 7 => "neutral", 8..=14 => "base", _ => unreachable!(), }); let uint: u32 = 5; match uint { 0 => "zero!", 1.. => "positive number!", }; // using paths to constants: const TROPOSPHERE_MIN : u8 = 6; const TROPOSPHERE_MAX : u8 = 20; const STRATOSPHERE_MIN : u8 = TROPOSPHERE_MAX + 1; const STRATOSPHERE_MAX : u8 = 50; const MESOSPHERE_MIN : u8 = STRATOSPHERE_MAX + 1; const MESOSPHERE_MAX : u8 = 85; let altitude = 70; println!("{}", match altitude { TROPOSPHERE_MIN..=TROPOSPHERE_MAX => "troposphere", STRATOSPHERE_MIN..=STRATOSPHERE_MAX => "stratosphere", MESOSPHERE_MIN..=MESOSPHERE_MAX => "mesosphere", _ => "outer space, maybe", }); pub mod binary { pub const MEGA : u64 = 1024*1024; pub const GIGA : u64 = 1024*1024*1024; } let n_items = 20_832_425; let bytes_per_item = 12; if let size @ binary::MEGA..=binary::GIGA = n_items * bytes_per_item { println!("It fits and occupies {} bytes", size); } trait MaxValue { const MAX: u64; } impl MaxValue for u8 { const MAX: u64 = (1 << 8) - 1; } impl MaxValue for u16 { const MAX: u64 = (1 << 16) - 1; } impl MaxValue for u32 { const MAX: u64 = (1 << 32) - 1; } // using qualified paths: println!("{}", match 0xfacade { 0 ..= <u8 as MaxValue>::MAX => "fits in a u8", 0 ..= <u16 as MaxValue>::MAX => "fits in a u16", 0 ..= <u32 as MaxValue>::MAX => "fits in a u32", _ => "too big", }); }
Range patterns for fix-width integer and char
types are irrefutable when they span the entire set of possible values of a type.
For example, 0u8..=255u8
is irrefutable.
The range of values for an integer type is the closed range from its minimum to maximum value.
The range of values for a char
type are precisely those ranges containing all Unicode Scalar Values: '\u{0000}'..='\u{D7FF}'
and '\u{E000}'..='\u{10FFFF}'
.
Edition differences: Before the 2021 edition, range patterns with both a lower and upper bound may also be written using
...
in place of..=
, with the same meaning.
Reference patterns
Syntax
ReferencePattern :
(&
|&&
)mut
? PatternWithoutRange
Reference patterns dereference the pointers that are being matched and, thus, borrow them.
For example, these two matches on x: &i32
are equivalent:
#![allow(unused)] fn main() { let int_reference = &3; let a = match *int_reference { 0 => "zero", _ => "some" }; let b = match int_reference { &0 => "zero", _ => "some" }; assert_eq!(a, b); }
The grammar production for reference patterns has to match the token &&
to match a reference to a reference because it is a token by itself, not two &
tokens.
Adding the mut
keyword dereferences a mutable reference. The mutability must match the mutability of the reference.
Reference patterns are always irrefutable.
Struct patterns
Syntax
StructPattern :
PathInExpression{
StructPatternElements ?
}
StructPatternElements :
StructPatternFields (,
|,
StructPatternEtCetera)?
| StructPatternEtCeteraStructPatternFields :
StructPatternField (,
StructPatternField) *StructPatternField :
OuterAttribute *
(
TUPLE_INDEX:
Pattern
| IDENTIFIER:
Pattern
|ref
?mut
? IDENTIFIER
)StructPatternEtCetera :
OuterAttribute *
..
Struct patterns match struct, enum, and union values that match all criteria defined by its subpatterns. They are also used to destructure a struct, enum, or union value.
On a struct pattern, the fields are referenced by name, index (in the case of tuple structs) or ignored by use of ..
:
#![allow(unused)] fn main() { struct Point { x: u32, y: u32, } let s = Point {x: 1, y: 1}; match s { Point {x: 10, y: 20} => (), Point {y: 10, x: 20} => (), // order doesn't matter Point {x: 10, ..} => (), Point {..} => (), } struct PointTuple ( u32, u32, ); let t = PointTuple(1, 2); match t { PointTuple {0: 10, 1: 20} => (), PointTuple {1: 10, 0: 20} => (), // order doesn't matter PointTuple {0: 10, ..} => (), PointTuple {..} => (), } enum Message { Quit, Move { x: i32, y: i32 }, } let m = Message::Quit; match m { Message::Quit => (), Message::Move {x: 10, y: 20} => (), Message::Move {..} => (), } }
If ..
is not used, a struct pattern used to match a struct is required to specify all fields:
#![allow(unused)] fn main() { struct Struct { a: i32, b: char, c: bool, } let mut struct_value = Struct{a: 10, b: 'X', c: false}; match struct_value { Struct{a: 10, b: 'X', c: false} => (), Struct{a: 10, b: 'X', ref c} => (), Struct{a: 10, b: 'X', ref mut c} => (), Struct{a: 10, b: 'X', c: _} => (), Struct{a: _, b: _, c: _} => (), } }
A struct pattern used to match a union must specify exactly one field (see Pattern matching on unions).
The ref
and/or mut
IDENTIFIER syntax matches any value and binds it to a variable with the same name as the given field.
#![allow(unused)] fn main() { struct Struct { a: i32, b: char, c: bool, } let struct_value = Struct{a: 10, b: 'X', c: false}; let Struct{a: x, b: y, c: z} = struct_value; // destructure all fields }
A struct pattern is refutable if the PathInExpression resolves to a constructor of an enum with more than one variant, or one of its subpatterns is refutable.
Tuple struct patterns
Syntax
TupleStructPattern :
PathInExpression(
TupleStructItems?)
Tuple struct patterns match tuple struct and enum values that match all criteria defined by its subpatterns. They are also used to destructure a tuple struct or enum value.
A tuple struct pattern is refutable if the PathInExpression resolves to a constructor of an enum with more than one variant, or one of its subpatterns is refutable.
Tuple patterns
Syntax
TuplePattern :
(
TuplePatternItems?)
TuplePatternItems :
Pattern,
| RestPattern
| Pattern (,
Pattern)+,
?
Tuple patterns match tuple values that match all criteria defined by its subpatterns. They are also used to destructure a tuple.
The form (..)
with a single RestPattern is a special form that does not require a comma, and matches a tuple of any size.
The tuple pattern is refutable when one of its subpatterns is refutable.
An example of using tuple patterns:
#![allow(unused)] fn main() { let pair = (10, "ten"); let (a, b) = pair; assert_eq!(a, 10); assert_eq!(b, "ten"); }
Grouped patterns
Syntax
GroupedPattern :
(
Pattern)
Enclosing a pattern in parentheses can be used to explicitly control the precedence of compound patterns.
For example, a reference pattern next to a range pattern such as &0..=5
is ambiguous and is not allowed, but can be expressed with parentheses.
#![allow(unused)] fn main() { let int_reference = &3; match int_reference { &(0..=5) => (), _ => (), } }
Slice patterns
Syntax
SlicePattern :
[
SlicePatternItems?]
Slice patterns can match both arrays of fixed size and slices of dynamic size.
#![allow(unused)] fn main() { // Fixed size let arr = [1, 2, 3]; match arr { [1, _, _] => "starts with one", [a, b, c] => "starts with something else", }; }
#![allow(unused)] fn main() { // Dynamic size let v = vec![1, 2, 3]; match v[..] { [a, b] => { /* this arm will not apply because the length doesn't match */ } [a, b, c] => { /* this arm will apply */ } _ => { /* this wildcard is required, since the length is not known statically */ } }; }
Slice patterns are irrefutable when matching an array as long as each element is irrefutable.
When matching a slice, it is irrefutable only in the form with a single ..
rest pattern or identifier pattern with the ..
rest pattern as a subpattern.
Within a slice, a range pattern without both lower and upper bound must be enclosed in parentheses, as in (a..)
, to clarify it is intended to match against a single slice element.
A range pattern with both lower and upper bound, like a..=b
, is not required to be enclosed in parentheses.
Path patterns
Syntax
PathPattern :
PathExpression
Path patterns are patterns that refer either to constant values or to structs or enum variants that have no fields.
Unqualified path patterns can refer to:
- enum variants
- structs
- constants
- associated constants
Qualified path patterns can only refer to associated constants.
Path patterns are irrefutable when they refer to structs or an enum variant when the enum has only one variant or a constant whose type is irrefutable. They are refutable when they refer to refutable constants or enum variants for enums with multiple variants.
Constant patterns
When a constant C
of type T
is used as a pattern, we first check that T: PartialEq
.
Furthermore we require that the value of C
has (recursive) structural equality, which is defined recursively as follows:
- Integers as well as
str
,bool
andchar
values always have structural equality. - Tuples, arrays, and slices have structural equality if all their fields/elements have structural equality.
(In particular,
()
and[]
always have structural equality.) - References have structural equality if the value they point to has structural equality.
- A value of
struct
orenum
type has structural equality if itsPartialEq
instance is derived via#[derive(PartialEq)]
, and all fields (for enums: of the active variant) have structural equality. - A raw pointer has structural equality if it was defined as a constant integer (and then cast/transmuted).
- A float value has structural equality if it is not a
NaN
. - Nothing else has structural equality.
In particular, the value of C
must be known at pattern-building time (which is pre-monomorphization).
This means that associated consts that involve generic parameters cannot be used as patterns.
After ensuring all conditions are met, the constant value is translated into a pattern, and now behaves exactly as-if that pattern had been written directly.
In particular, it fully participates in exhaustiveness checking.
(For raw pointers, constants are the only way to write such patterns. Only _
is ever considered exhaustive for these types.)
Or-patterns
Or-patterns are patterns that match on one of two or more sub-patterns (for example A | B | C
).
They can nest arbitrarily.
Syntactically, or-patterns are allowed in any of the places where other patterns are allowed (represented by the Pattern production), with the exceptions of let
-bindings and function and closure arguments (represented by the PatternNoTopAlt production).
Static semantics
-
Given a pattern
p | q
at some depth for some arbitrary patternsp
andq
, the pattern is considered ill-formed if:- the type inferred for
p
does not unify with the type inferred forq
, or - the same set of bindings are not introduced in
p
andq
, or - the type of any two bindings with the same name in
p
andq
do not unify with respect to types or binding modes.
Unification of types is in all instances aforementioned exact and implicit type coercions do not apply.
- the type inferred for
-
When type checking an expression
match e_s { a_1 => e_1, ... a_n => e_n }
, for each match arma_i
which contains a pattern of formp_i | q_i
, the patternp_i | q_i
is considered ill formed if, at the depthd
where it exists the fragment ofe_s
at depthd
, the type of the expression fragment does not unify withp_i | q_i
. -
With respect to exhaustiveness checking, a pattern
p | q
is considered to coverp
as well asq
. For some constructorc(x, ..)
the distributive law applies such thatc(p | q, ..rest)
covers the same set of value asc(p, ..rest) | c(q, ..rest)
does. This can be applied recursively until there are no more nested patterns of formp | q
other than those that exist at the top level.Note that by “constructor” we do not refer to tuple struct patterns, but rather we refer to a pattern for any product type. This includes enum variants, tuple structs, structs with named fields, arrays, tuples, and slices.
Dynamic semantics
- The dynamic semantics of pattern matching a scrutinee expression
e_s
against a patternc(p | q, ..rest)
at depthd
wherec
is some constructor,p
andq
are arbitrary patterns, andrest
is optionally any remaining potential factors inc
, is defined as being the same as that ofc(p, ..rest) | c(q, ..rest)
.
Precedence with other undelimited patterns
As shown elsewhere in this chapter, there are several types of patterns that are syntactically undelimited, including identifier patterns, reference patterns, and or-patterns.
Or-patterns always have the lowest-precedence.
This allows us to reserve syntactic space for a possible future type ascription feature and also to reduce ambiguity.
For example, x @ A(..) | B(..)
will result in an error that x
is not bound in all patterns.
&A(x) | B(x)
will result in a type mismatch between x
in the different subpatterns.
Type system
Types
Every variable, item, and value in a Rust program has a type. The type of a value defines the interpretation of the memory holding it and the operations that may be performed on the value.
Built-in types are tightly integrated into the language, in nontrivial ways that are not possible to emulate in user-defined types. User-defined types have limited capabilities.
The list of types is:
- Primitive types:
- Sequence types:
- User-defined types:
- Function types:
- Pointer types:
- Trait types:
Type expressions
Syntax
Type :
TypeNoBounds
| ImplTraitType
| TraitObjectTypeTypeNoBounds :
ParenthesizedType
| ImplTraitTypeOneBound
| TraitObjectTypeOneBound
| TypePath
| TupleType
| NeverType
| RawPointerType
| ReferenceType
| ArrayType
| SliceType
| InferredType
| QualifiedPathInType
| BareFunctionType
| MacroInvocation
A type expression as defined in the Type grammar rule above is the syntax for referring to a type. It may refer to:
- Sequence types (tuple, array, slice).
- Type paths which can reference:
- Pointer types (reference, raw pointer, function pointer).
- The inferred type which asks the compiler to determine the type.
- Parentheses which are used for disambiguation.
- Trait types: Trait objects and impl trait.
- The never type.
- Macros which expand to a type expression.
Parenthesized types
ParenthesizedType :
(
Type)
In some situations the combination of types may be ambiguous. Use parentheses
around a type to avoid ambiguity. For example, the +
operator for type
boundaries within a reference type is unclear where the
boundary applies, so the use of parentheses is required. Grammar rules that
require this disambiguation use the TypeNoBounds rule instead of
Type.
#![allow(unused)] fn main() { use std::any::Any; type T<'a> = &'a (dyn Any + Send); }
Recursive types
Nominal types — structs, enumerations, and unions — may be
recursive. That is, each enum
variant or struct
or union
field may
refer, directly or indirectly, to the enclosing enum
or struct
type
itself. Such recursion has restrictions:
- Recursive types must include a nominal type in the recursion (not mere type
aliases, or other structural types such as arrays or tuples). So
type Rec = &'static [Rec]
is not allowed. - The size of a recursive type must be finite; in other words the recursive fields of the type must be pointer types.
An example of a recursive type and its use:
#![allow(unused)] fn main() { enum List<T> { Nil, Cons(T, Box<List<T>>) } let a: List<i32> = List::Cons(7, Box::new(List::Cons(13, Box::new(List::Nil)))); }
Boolean type
#![allow(unused)] fn main() { let b: bool = true; }
The boolean type or bool is a primitive data type that can take on one of two values, called true and false.
Values of this type may be created using a literal expression using the
keywords true
and false
corresponding to the value of the same name.
This type is a part of the language prelude with the name bool
.
An object with the boolean type has a size and alignment of 1 each. The
value false has the bit pattern 0x00
and the value true has the bit pattern
0x01
. It is undefined behavior for an object with the boolean type to have
any other bit pattern.
The boolean type is the type of many operands in various expressions:
- The condition operand in if expressions and while expressions
- The operands in lazy boolean operator expressions
Note: The boolean type acts similarly to but is not an enumerated type. In practice, this mostly means that constructors are not associated to the type (e.g.
bool::true
).
Like all primitives, the boolean type implements the
traits Clone
, Copy
, Sized
,
Send
, and Sync
.
Note: See the standard library docs for library operations.
Operations on boolean values
When using certain operator expressions with aboolean type for its operands, they evaluate using the rules of boolean logic.
Logical not
b | !b |
---|---|
true | false |
false | true |
Logical or
a | b | a | b |
---|---|---|
true | true | true |
true | false | true |
false | true | true |
false | false | false |
Logical and
a | b | a & b |
---|---|---|
true | true | true |
true | false | false |
false | true | false |
false | false | false |
Logical xor
a | b | a ^ b |
---|---|---|
true | true | false |
true | false | true |
false | true | true |
false | false | false |
Comparisons
a | b | a == b |
---|---|---|
true | true | true |
true | false | false |
false | true | false |
false | false | true |
a | b | a > b |
---|---|---|
true | true | false |
true | false | true |
false | true | false |
false | false | false |
a != b
is the same as!(a == b)
a >= b
is the same asa == b | a > b
a < b
is the same as!(a >= b)
a <= b
is the same asa == b | a < b
Bit validity
The single byte of a bool
is guaranteed to be initialized (in other words,
transmute::<bool, u8>(...)
is always sound – but since some bit patterns
are invalid bool
s, the inverse is not always sound).
Numeric types
Integer types
The unsigned integer types consist of:
Type | Minimum | Maximum |
---|---|---|
u8 | 0 | 28-1 |
u16 | 0 | 216-1 |
u32 | 0 | 232-1 |
u64 | 0 | 264-1 |
u128 | 0 | 2128-1 |
The signed two’s complement integer types consist of:
Type | Minimum | Maximum |
---|---|---|
i8 | -(27) | 27-1 |
i16 | -(215) | 215-1 |
i32 | -(231) | 231-1 |
i64 | -(263) | 263-1 |
i128 | -(2127) | 2127-1 |
Floating-point types
The IEEE 754-2008 “binary32” and “binary64” floating-point types are f32
and
f64
, respectively.
Machine-dependent integer types
The usize
type is an unsigned integer type with the same number of bits as the
platform’s pointer type. It can represent every memory address in the process.
The isize
type is a signed integer type with the same number of bits as the
platform’s pointer type. The theoretical upper bound on object and array size
is the maximum isize
value. This ensures that isize
can be used to calculate
differences between pointers into an object or array and can address every byte
within an object along with one byte past the end.
usize
and isize
are at least 16-bits wide.
Note: Many pieces of Rust code may assume that pointers,
usize
, andisize
are either 32-bit or 64-bit. As a consequence, 16-bit pointer support is limited and may require explicit care and acknowledgment from a library to support.
Bit validity
For every numeric type, T
, the bit validity of T
is equivalent to the bit
validity of [u8; size_of::<T>()]
. An uninitialized byte is not a valid u8
.
Textual types
The types char
and str
hold textual data.
A value of type char
is a Unicode scalar value (i.e. a code point that is
not a surrogate), represented as a 32-bit unsigned word in the 0x0000 to 0xD7FF
or 0xE000 to 0x10FFFF range. It is immediate undefined behavior to create a
char
that falls outside this range. A [char]
is effectively a UCS-4 / UTF-32
string of length 1.
A value of type str
is represented the same way as [u8]
, a slice of
8-bit unsigned bytes. However, the Rust standard library makes extra assumptions
about str
: methods working on str
assume and ensure that the data in there
is valid UTF-8. Calling a str
method with a non-UTF-8 buffer can cause
undefined behavior now or in the future.
Since str
is a dynamically sized type, it can only be instantiated through a
pointer type, such as &str
.
Layout and bit validity
char
is guaranteed to have the same size and alignment as u32
on all platforms.
Every byte of a char
is guaranteed to be initialized (in other words,
transmute::<char, [u8; size_of::<char>()]>(...)
is always sound – but since
some bit patterns are invalid char
s, the inverse is not always sound).
Never type
Syntax
NeverType :!
The never type !
is a type with no values, representing the result of
computations that never complete. Expressions of type !
can be coerced into
any other type.
The !
type can only appear in function return types presently,
indicating it is a diverging function that never returns.
#![allow(unused)] fn main() { fn foo() -> ! { panic!("This call never returns."); } }
#![allow(unused)] fn main() { unsafe extern "C" { pub safe fn no_return_extern_func() -> !; } }
Tuple types
Tuple types are a family of structural types1 for heterogeneous lists of other types.
The syntax for a tuple type is a parenthesized, comma-separated list of types. 1-ary tuples require a comma after their element type to be disambiguated with a parenthesized type.
A tuple type has a number of fields equal to the length of the list of types.
This number of fields determines the arity of the tuple.
A tuple with n
fields is called an n-ary tuple.
For example, a tuple with 2 fields is a 2-ary tuple.
Fields of tuples are named using increasing numeric names matching their position in the list of types.
The first field is 0
.
The second field is 1
.
And so on.
The type of each field is the type of the same position in the tuple’s list of types.
For convenience and historical reasons, the tuple type with no fields (()
) is often called unit or the unit type.
Its one value is also called unit or the unit value.
Some examples of tuple types:
()
(unit)(i32,)
(1-ary tuple)(f64, f64)
(String, i32)
(i32, String)
(different type from the previous example)(i32, f64, Vec<String>, Option<bool>)
Values of this type are constructed using a tuple expression. Furthermore, various expressions will produce the unit value if there is no other meaningful value for it to evaluate to. Tuple fields can be accessed by either a tuple index expression or pattern matching.
Structural types are always equivalent if their internal types are equivalent. For a nominal version of tuples, see tuple structs.
Array types
Syntax
ArrayType :
[
Type;
Expression]
An array is a fixed-size sequence of N
elements of type T
. The array type
is written as [T; N]
. The size is a constant expression that evaluates to a
usize
.
Examples:
#![allow(unused)] fn main() { // A stack-allocated array let array: [i32; 3] = [1, 2, 3]; // A heap-allocated array, coerced to a slice let boxed_array: Box<[i32]> = Box::new([1, 2, 3]); }
All elements of arrays are always initialized, and access to an array is always bounds-checked in safe methods and operators.
Note: The
Vec<T>
standard library type provides a heap-allocated resizable array type.
Slice types
Syntax
SliceType :
[
Type]
A slice is a dynamically sized type representing a ‘view’ into a sequence of
elements of type T
. The slice type is written as [T]
.
Slice types are generally used through pointer types. For example:
&[T]
: a ‘shared slice’, often just called a ‘slice’. It doesn’t own the data it points to; it borrows it.&mut [T]
: a ‘mutable slice’. It mutably borrows the data it points to.Box<[T]>
: a ‘boxed slice’
Examples:
#![allow(unused)] fn main() { // A heap-allocated array, coerced to a slice let boxed_array: Box<[i32]> = Box::new([1, 2, 3]); // A (shared) slice into an array let slice: &[i32] = &boxed_array[..]; }
All elements of slices are always initialized, and access to a slice is always bounds-checked in safe methods and operators.
Struct types
A struct
type is a heterogeneous product of other types, called the
fields of the type.1
New instances of a struct
can be constructed with a struct expression.
The memory layout of a struct
is undefined by default to allow for compiler
optimizations like field reordering, but it can be fixed with the
repr
attribute. In either case, fields may be given in any order in a
corresponding struct expression; the resulting struct
value will always
have the same memory layout.
The fields of a struct
may be qualified by visibility modifiers, to allow
access to data in a struct outside a module.
A tuple struct type is just like a struct type, except that the fields are anonymous.
A unit-like struct type is like a struct type, except that it has no fields. The one value constructed by the associated struct expression is the only value that inhabits such a type.
struct
types are analogous to struct
types in C, the
record types of the ML family, or the struct types of the Lisp family.
Enumerated types
An enumerated type is a nominal, heterogeneous disjoint union type, denoted
by the name of an enum
item. 1
An enum
item declares both the type and a number of variants, each of
which is independently named and has the syntax of a struct, tuple struct or
unit-like struct.
New instances of an enum
can be constructed with a struct expression.
Any enum
value consumes as much memory as the largest variant for its
corresponding enum
type, as well as the size needed to store a discriminant.
Enum types cannot be denoted structurally as types, but must be denoted by
named reference to an enum
item.
The enum
type is analogous to a data
constructor declaration in
Haskell, or a pick ADT in Limbo.
Union types
A union type is a nominal, heterogeneous C-like union, denoted by the name of
a union
item.
Unions have no notion of an “active field”. Instead, every union access
transmutes parts of the content of the union to the type of the accessed field.
Since transmutes can cause unexpected or undefined behaviour, unsafe
is
required to read from a union field. Union field types are also restricted to a
subset of types which ensures that they never need dropping. See the item
documentation for further details.
The memory layout of a union
is undefined by default (in particular, fields do
not have to be at offset 0), but the #[repr(...)]
attribute can be used to
fix a layout.
Function item types
When referred to, a function item, or the constructor of a tuple-like struct or enum variant, yields a zero-sized value of its function item type. That type explicitly identifies the function - its name, its type arguments, and its early-bound lifetime arguments (but not its late-bound lifetime arguments, which are only assigned when the function is called) - so the value does not need to contain an actual function pointer, and no indirection is needed when the function is called.
There is no syntax that directly refers to a function item type, but the
compiler will display the type as something like fn(u32) -> i32 {fn_name}
in
error messages.
Because the function item type explicitly identifies the function, the item types of different functions - different items, or the same item with different generics - are distinct, and mixing them will create a type error:
#![allow(unused)] fn main() { fn foo<T>() { } let x = &mut foo::<i32>; *x = foo::<u32>; //~ ERROR mismatched types }
However, there is a coercion from function items to function pointers with
the same signature, which is triggered not only when a function item is used
when a function pointer is directly expected, but also when different function
item types with the same signature meet in different arms of the same if
or
match
:
#![allow(unused)] fn main() { let want_i32 = false; fn foo<T>() { } // `foo_ptr_1` has function pointer type `fn()` here let foo_ptr_1: fn() = foo::<i32>; // ... and so does `foo_ptr_2` - this type-checks. let foo_ptr_2 = if want_i32 { foo::<i32> } else { foo::<u32> }; }
All function items implement Fn
, FnMut
, FnOnce
, Copy
,
Clone
, Send
, and Sync
.
Closure types
A closure expression produces a closure value with a unique, anonymous type that cannot be written out. A closure type is approximately equivalent to a struct which contains the captured variables. For instance, the following closure:
#![allow(unused)] fn main() { fn f<F : FnOnce() -> String> (g: F) { println!("{}", g()); } let mut s = String::from("foo"); let t = String::from("bar"); f(|| { s += &t; s }); // Prints "foobar". }
generates a closure type roughly like the following:
struct Closure<'a> {
s : String,
t : &'a String,
}
impl<'a> FnOnce<()> for Closure<'a> {
type Output = String;
fn call_once(self) -> String {
self.s += &*self.t;
self.s
}
}
so that the call to f
works as if it were:
f(Closure{s: s, t: &t});
Capture modes
The compiler prefers to capture a closed-over variable by immutable borrow, followed by unique immutable borrow (see below), by mutable borrow, and finally by move. It will pick the first choice of these that is compatible with how the captured variable is used inside the closure body. The compiler does not take surrounding code into account, such as the lifetimes of involved variables, or of the closure itself.
If the move
keyword is used, then all captures are by move or, for Copy
types, by copy, regardless of whether a borrow would work. The move
keyword is
usually used to allow the closure to outlive the captured values, such as if the
closure is being returned or used to spawn a new thread.
Composite types such as structs, tuples, and enums are always captured entirely, not by individual fields. It may be necessary to borrow into a local variable in order to capture a single field:
#![allow(unused)] fn main() { use std::collections::HashSet; struct SetVec { set: HashSet<u32>, vec: Vec<u32> } impl SetVec { fn populate(&mut self) { let vec = &mut self.vec; self.set.iter().for_each(|&n| { vec.push(n); }) } } }
If, instead, the closure were to use self.vec
directly, then it would attempt
to capture self
by mutable reference. But since self.set
is already
borrowed to iterate over, the code would not compile.
Unique immutable borrows in captures
Captures can occur by a special kind of borrow called a unique immutable borrow, which cannot be used anywhere else in the language and cannot be written out explicitly. It occurs when modifying the referent of a mutable reference, as in the following example:
#![allow(unused)] fn main() { let mut b = false; let x = &mut b; { let mut c = || { *x = true; }; // The following line is an error: // let y = &x; c(); } let z = &x; }
In this case, borrowing x
mutably is not possible, because x
is not mut
.
But at the same time, borrowing x
immutably would make the assignment illegal,
because a & &mut
reference might not be unique, so it cannot safely be used to
modify a value. So a unique immutable borrow is used: it borrows x
immutably,
but like a mutable borrow, it must be unique. In the above example, uncommenting
the declaration of y
will produce an error because it would violate the
uniqueness of the closure’s borrow of x
; the declaration of z is valid because
the closure’s lifetime has expired at the end of the block, releasing the borrow.
Call traits and coercions
Closure types all implement FnOnce
, indicating that they can be called once
by consuming ownership of the closure. Additionally, some closures implement
more specific call traits:
-
A closure which does not move out of any captured variables implements
FnMut
, indicating that it can be called by mutable reference. -
A closure which does not mutate or move out of any captured variables implements
Fn
, indicating that it can be called by shared reference.
Note:
move
closures may still implementFn
orFnMut
, even though they capture variables by move. This is because the traits implemented by a closure type are determined by what the closure does with captured values, not how it captures them.
Non-capturing closures are closures that don’t capture anything from their
environment. They can be coerced to function pointers (e.g., fn()
)
with the matching signature.
#![allow(unused)] fn main() { let add = |x, y| x + y; let mut x = add(5,7); type Binop = fn(i32, i32) -> i32; let bo: Binop = add; x = bo(5,7); }
Other traits
All closure types implement Sized
. Additionally, closure types implement the
following traits if allowed to do so by the types of the captures it stores:
The rules for Send
and Sync
match those for normal struct types, while
Clone
and Copy
behave as if derived. For Clone
, the order of
cloning of the captured variables is left unspecified.
Because captures are often by reference, the following general rules arise:
- A closure is
Sync
if all captured variables areSync
. - A closure is
Send
if all variables captured by non-unique immutable reference areSync
, and all values captured by unique immutable or mutable reference, copy, or move areSend
. - A closure is
Clone
orCopy
if it does not capture any values by unique immutable or mutable reference, and if all values it captures by copy or move areClone
orCopy
, respectively.
Pointer types
All pointers are explicit first-class values. They can be moved or copied, stored into data structs, and returned from functions.
References (&
and &mut
)
Syntax
ReferenceType :
&
Lifetime?mut
? TypeNoBounds
Shared references (&
)
Shared references point to memory which is owned by some other value.
When a shared reference to a value is created, it prevents direct mutation of the value.
Interior mutability provides an exception for this in certain circumstances.
As the name suggests, any number of shared references to a value may exist.
A shared reference type is written &type
, or &'a type
when you need to specify an explicit lifetime.
Copying a reference is a “shallow” operation:
it involves only copying the pointer itself, that is, pointers are Copy
.
Releasing a reference has no effect on the value it points to, but referencing of a temporary value will keep it alive during the scope of the reference itself.
Mutable references (&mut
)
Mutable references point to memory which is owned by some other value.
A mutable reference type is written &mut type
or &'a mut type
.
A mutable reference (that hasn’t been borrowed) is the only way to access the value it points to, so is not Copy
.
Raw pointers (*const
and *mut
)
Syntax
RawPointerType :
*
(mut
|const
) TypeNoBounds
Raw pointers are pointers without safety or liveness guarantees.
Raw pointers are written as *const T
or *mut T
.
For example *const i32
means a raw pointer to a 32-bit integer.
Copying or dropping a raw pointer has no effect on the lifecycle of any other value.
Dereferencing a raw pointer is an unsafe
operation.
This can also be used to convert a raw pointer to a reference by reborrowing it (&*
or &mut *
).
Raw pointers are generally discouraged;
they exist to support interoperability with foreign code, and writing performance-critical or low-level functions.
When comparing raw pointers they are compared by their address, rather than by what they point to. When comparing raw pointers to dynamically sized types they also have their additional data compared.
Raw pointers can be created directly using &raw const
for *const
pointers and &raw mut
for *mut
pointers.
Smart Pointers
The standard library contains additional ‘smart pointer’ types beyond references and raw pointers.
Bit validity
Despite pointers and references being similar to usize
s in the machine code emitted on most platforms,
the semantics of transmuting a reference or pointer type to a non-pointer type is currently undecided.
Thus, it may not be valid to transmute a pointer or reference type, P
, to a [u8; size_of::<P>()]
.
For thin raw pointers (i.e., for P = *const T
or P = *mut T
for T: Sized
),
the inverse direction (transmuting from an integer or array of integers to P
) is always valid.
However, the pointer produced via such a transmutation may not be dereferenced (not even if T
has size zero).
Function pointer types
Syntax
BareFunctionType :
ForLifetimes? FunctionTypeQualifiersfn
(
FunctionParametersMaybeNamedVariadic?)
BareFunctionReturnType?FunctionTypeQualifiers:
unsafe
? (extern
Abi?)?BareFunctionReturnType:
->
TypeNoBoundsFunctionParametersMaybeNamedVariadic :
MaybeNamedFunctionParameters | MaybeNamedFunctionParametersVariadicMaybeNamedFunctionParameters :
MaybeNamedParam (,
MaybeNamedParam )*,
?MaybeNamedParam :
OuterAttribute* ( ( IDENTIFIER |_
):
)? TypeMaybeNamedFunctionParametersVariadic :
( MaybeNamedParam,
)* MaybeNamedParam,
OuterAttribute*...
Function pointer types, written using the fn
keyword, refer to a function
whose identity is not necessarily known at compile-time. They can be created
via a coercion from both function items and non-capturing closures.
The unsafe
qualifier indicates that the type’s value is an unsafe
function, and the extern
qualifier indicates it is an extern function.
Variadic parameters can only be specified with extern
function types with
the "C"
or "cdecl"
calling convention.
An example where Binop
is defined as a function pointer type:
#![allow(unused)] fn main() { fn add(x: i32, y: i32) -> i32 { x + y } let mut x = add(5,7); type Binop = fn(i32, i32) -> i32; let bo: Binop = add; x = bo(5,7); }
Attributes on function pointer parameters
Attributes on function pointer parameters follow the same rules and restrictions as regular function parameters.
Trait objects
Syntax
TraitObjectType :
dyn
? TypeParamBoundsTraitObjectTypeOneBound :
dyn
? TraitBound
A trait object is an opaque value of another type that implements a set of traits. The set of traits is made up of an object safe base trait plus any number of auto traits.
Trait objects implement the base trait, its auto traits, and any supertraits of the base trait.
Trait objects are written as the keyword dyn
followed by a set of trait
bounds, but with the following restrictions on the trait bounds. All traits
except the first trait must be auto traits, there may not be more than one
lifetime, and opt-out bounds (e.g. ?Sized
) are not allowed. Furthermore,
paths to traits may be parenthesized.
For example, given a trait Trait
, the following are all trait objects:
dyn Trait
dyn Trait + Send
dyn Trait + Send + Sync
dyn Trait + 'static
dyn Trait + Send + 'static
dyn Trait +
dyn 'static + Trait
.dyn (Trait)
Edition differences: Before the 2021 edition, the
dyn
keyword may be omitted.Note: For clarity, it is recommended to always use the
dyn
keyword on your trait objects unless your codebase supports compiling with Rust 1.26 or lower.
Edition differences: In the 2015 edition, if the first bound of the trait object is a path that starts with
::
, then thedyn
will be treated as a part of the path. The first path can be put in parenthesis to get around this. As such, if you want a trait object with the trait::your_module::Trait
, you should write it asdyn (::your_module::Trait)
.Beginning in the 2018 edition,
dyn
is a true keyword and is not allowed in paths, so the parentheses are not necessary.
Two trait object types alias each other if the base traits alias each other and
if the sets of auto traits are the same and the lifetime bounds are the same.
For example, dyn Trait + Send + UnwindSafe
is the same as
dyn Trait + UnwindSafe + Send
.
Due to the opaqueness of which concrete type the value is of, trait objects are
dynamically sized types. Like all
DSTs, trait objects are used
behind some type of pointer; for example &dyn SomeTrait
or
Box<dyn SomeTrait>
. Each instance of a pointer to a trait object includes:
- a pointer to an instance of a type
T
that implementsSomeTrait
- a virtual method table, often just called a vtable, which contains, for
each method of
SomeTrait
and its supertraits thatT
implements, a pointer toT
’s implementation (i.e. a function pointer).
The purpose of trait objects is to permit “late binding” of methods. Calling a method on a trait object results in virtual dispatch at runtime: that is, a function pointer is loaded from the trait object vtable and invoked indirectly. The actual implementation for each vtable entry can vary on an object-by-object basis.
An example of a trait object:
trait Printable { fn stringify(&self) -> String; } impl Printable for i32 { fn stringify(&self) -> String { self.to_string() } } fn print(a: Box<dyn Printable>) { println!("{}", a.stringify()); } fn main() { print(Box::new(10) as Box<dyn Printable>); }
In this example, the trait Printable
occurs as a trait object in both the
type signature of print
, and the cast expression in main
.
Trait Object Lifetime Bounds
Since a trait object can contain references, the lifetimes of those references
need to be expressed as part of the trait object. This lifetime is written as
Trait + 'a
. There are defaults that allow this lifetime to usually be
inferred with a sensible choice.
Impl trait
Syntax
ImplTraitType :impl
TypeParamBoundsImplTraitTypeOneBound :
impl
TraitBound
impl Trait
provides ways to specify unnamed but concrete types that
implement a specific trait.
It can appear in two sorts of places: argument position (where it can act as an anonymous type parameter to functions), and return position (where it can act as an abstract return type).
#![allow(unused)] fn main() { trait Trait {} impl Trait for () {} // argument position: anonymous type parameter fn foo(arg: impl Trait) { } // return position: abstract return type fn bar() -> impl Trait { } }
Anonymous type parameters
Note: This is often called “impl Trait in argument position”. (The term “parameter” is more correct here, but “impl Trait in argument position” is the phrasing used during the development of this feature, and it remains in parts of the implementation.)
Functions can use impl
followed by a set of trait bounds to declare a parameter as having an anonymous type.
The caller must provide a type that satisfies the bounds declared by the anonymous type parameter, and the function can only use the methods available through the trait bounds of the anonymous type parameter.
For example, these two forms are almost equivalent:
#![allow(unused)] fn main() { trait Trait {} // generic type parameter fn with_generic_type<T: Trait>(arg: T) { } // impl Trait in argument position fn with_impl_trait(arg: impl Trait) { } }
That is, impl Trait
in argument position is syntactic sugar for a generic type parameter like <T: Trait>
, except that the type is anonymous and doesn’t appear in the GenericParams list.
Note: For function parameters, generic type parameters and
impl Trait
are not exactly equivalent. With a generic parameter such as<T: Trait>
, the caller has the option to explicitly specify the generic argument forT
at the call site using GenericArgs, for example,foo::<usize>(1)
. Changing a parameter from either one to the other can constitute a breaking change for the callers of a function, since this changes the number of generic arguments.
Abstract return types
Note: This is often called “impl Trait in return position”.
Functions can use impl Trait
to return an abstract return type.
These types stand in for another concrete type where the caller may only use the methods declared by the specified Trait
.
Each possible return value from the function must resolve to the same concrete type.
impl Trait
in return position allows a function to return an unboxed abstract type.
This is particularly useful with closures and iterators.
For example, closures have a unique, un-writable type.
Previously, the only way to return a closure from a function was to use a trait object:
#![allow(unused)] fn main() { fn returns_closure() -> Box<dyn Fn(i32) -> i32> { Box::new(|x| x + 1) } }
This could incur performance penalties from heap allocation and dynamic dispatch.
It wasn’t possible to fully specify the type of the closure, only to use the Fn
trait.
That means that the trait object is necessary.
However, with impl Trait
, it is possible to write this more simply:
#![allow(unused)] fn main() { fn returns_closure() -> impl Fn(i32) -> i32 { |x| x + 1 } }
which also avoids the drawbacks of using a boxed trait object.
Similarly, the concrete types of iterators could become very complex, incorporating the types of all previous iterators in a chain.
Returning impl Iterator
means that a function only exposes the Iterator
trait as a bound on its return type, instead of explicitly specifying all of the other iterator types involved.
Return-position impl Trait
in traits and trait implementations
Functions in traits may also use impl Trait
as a syntax for an anonymous associated type.
Every impl Trait
in the return type of an associated function in a trait is desugared to an anonymous associated type. The return type that appears in the implementation’s function signature is used to determine the value of the associated type.
Differences between generics and impl Trait
in return position
In argument position, impl Trait
is very similar in semantics to a generic type parameter.
However, there are significant differences between the two in return position.
With impl Trait
, unlike with a generic type parameter, the function chooses the return type, and the caller cannot choose the return type.
The function:
#![allow(unused)] fn main() { trait Trait {} fn foo<T: Trait>() -> T { // ... panic!() } }
allows the caller to determine the return type, T
, and the function returns that type.
The function:
#![allow(unused)] fn main() { trait Trait {} impl Trait for () {} fn foo() -> impl Trait { // ... } }
doesn’t allow the caller to determine the return type.
Instead, the function chooses the return type, but only promises that it will implement Trait
.
Limitations
impl Trait
can only appear as a parameter or return type of a non-extern
function.
It cannot be the type of a let
binding, field type, or appear inside a type alias.
Type parameters
Within the body of an item that has type parameter declarations, the names of its type parameters are types:
#![allow(unused)] fn main() { fn to_vec<A: Clone>(xs: &[A]) -> Vec<A> { if xs.is_empty() { return vec![]; } let first: A = xs[0].clone(); let mut rest: Vec<A> = to_vec(&xs[1..]); rest.insert(0, first); rest } }
Here, first
has type A
, referring to to_vec
’s A
type parameter; and
rest
has type Vec<A>
, a vector with element type A
.
Inferred type
Syntax
InferredType :_
The inferred type asks the compiler to infer the type if possible based on the surrounding information available. It cannot be used in item signatures. It is often used in generic arguments:
#![allow(unused)] fn main() { let x: Vec<_> = (0..10).collect(); }
Dynamically Sized Types
Most types have a fixed size that is known at compile time and implement the
trait Sized
. A type with a size that is known only at run-time is
called a dynamically sized type (DST) or, informally, an unsized type.
Slices and trait objects are two examples of DSTs. Such types can only be used in certain cases:
- Pointer types to DSTs are
sized but have twice the size of pointers to sized types
- Pointers to slices also store the number of elements of the slice.
- Pointers to trait objects also store a pointer to a vtable.
- DSTs can be provided as
type arguments to generic type parameters having the special
?Sized
bound. They can also be used for associated type definitions when the corresponding associated type declaration has a?Sized
bound. By default, any type parameter or associated type has aSized
bound, unless it is relaxed using?Sized
. - Traits may be implemented for DSTs.
Unlike with generic type parameters,
Self: ?Sized
is the default in trait definitions. - Structs may contain a DST as the last field; this makes the struct itself a DST.
Note: variables, function parameters, const items, and static items must be
Sized
.
Type Layout
The layout of a type is its size, alignment, and the relative offsets of its fields. For enums, how the discriminant is laid out and interpreted is also part of type layout.
Type layout can be changed with each compilation. Instead of trying to document exactly what is done, we only document what is guaranteed today.
Size and Alignment
All values have an alignment and size.
The alignment of a value specifies what addresses are valid to store the value
at. A value of alignment n
must only be stored at an address that is a
multiple of n. For example, a value with an alignment of 2 must be stored at an
even address, while a value with an alignment of 1 can be stored at any address.
Alignment is measured in bytes, and must be at least 1, and always a power of 2.
The alignment of a value can be checked with the align_of_val
function.
The size of a value is the offset in bytes between successive elements in an
array with that item type including alignment padding. The size of a value is
always a multiple of its alignment. Note that some types are zero-sized; 0 is
considered a multiple of any alignment (for example, on some platforms, the type
[u16; 0]
has size 0 and alignment 2). The size of a value can be checked with
the size_of_val
function.
Types where all values have the same size and alignment, and both are known at
compile time, implement the Sized
trait and can be checked with the
size_of
and align_of
functions. Types that are not Sized
are known
as dynamically sized types. Since all values of a Sized
type share the same
size and alignment, we refer to those shared values as the size of the type and
the alignment of the type respectively.
Primitive Data Layout
The size of most primitives is given in this table.
Type | size_of::<Type>() |
---|---|
bool | 1 |
u8 / i8 | 1 |
u16 / i16 | 2 |
u32 / i32 | 4 |
u64 / i64 | 8 |
u128 / i128 | 16 |
usize / isize | See below |
f32 | 4 |
f64 | 8 |
char | 4 |
usize
and isize
have a size big enough to contain every address on the
target platform. For example, on a 32 bit target, this is 4 bytes, and on a 64
bit target, this is 8 bytes.
The alignment of primitives is platform-specific.
In most cases, their alignment is equal to their size, but it may be less.
In particular, i128
and u128
are often aligned to 4 or 8 bytes even though
their size is 16, and on many 32-bit platforms, i64
, u64
, and f64
are only
aligned to 4 bytes, not 8.
Pointers and References Layout
Pointers and references have the same layout. Mutability of the pointer or reference does not change the layout.
Pointers to sized types have the same size and alignment as usize
.
Pointers to unsized types are sized. The size and alignment is guaranteed to be at least equal to the size and alignment of a pointer.
Note: Though you should not rely on this, all pointers to DSTs are currently twice the size of the size of
usize
and have the same alignment.
Array Layout
An array of [T; N]
has a size of size_of::<T>() * N
and the same alignment
of T
. Arrays are laid out so that the zero-based nth
element of the array
is offset from the start of the array by n * size_of::<T>()
bytes.
Slice Layout
Slices have the same layout as the section of the array they slice.
Note: This is about the raw
[T]
type, not pointers (&[T]
,Box<[T]>
, etc.) to slices.
str
Layout
String slices are a UTF-8 representation of characters that have the same layout as slices of type [u8]
.
Tuple Layout
Tuples are laid out according to the Rust
representation.
The exception to this is the unit tuple (()
), which is guaranteed as a
zero-sized type to have a size of 0 and an alignment of 1.
Trait Object Layout
Trait objects have the same layout as the value the trait object is of.
Note: This is about the raw trait object types, not pointers (
&dyn Trait
,Box<dyn Trait>
, etc.) to trait objects.
Closure Layout
Closures have no layout guarantees.
Representations
All user-defined composite types (struct
s, enum
s, and union
s) have a
representation that specifies what the layout is for the type. The possible
representations for a type are:
Rust
(default)C
- The primitive representations
transparent
The representation of a type can be changed by applying the repr
attribute
to it. The following example shows a struct with a C
representation.
#![allow(unused)] fn main() { #[repr(C)] struct ThreeInts { first: i16, second: i8, third: i32 } }
The alignment may be raised or lowered with the align
and packed
modifiers
respectively. They alter the representation specified in the attribute.
If no representation is specified, the default one is altered.
#![allow(unused)] fn main() { // Default representation, alignment lowered to 2. #[repr(packed(2))] struct PackedStruct { first: i16, second: i8, third: i32 } // C representation, alignment raised to 8 #[repr(C, align(8))] struct AlignedStruct { first: i16, second: i8, third: i32 } }
Note: As a consequence of the representation being an attribute on the item, the representation does not depend on generic parameters. Any two types with the same name have the same representation. For example,
Foo<Bar>
andFoo<Baz>
both have the same representation.
The representation of a type can change the padding between fields, but does
not change the layout of the fields themselves. For example, a struct with a
C
representation that contains a struct Inner
with the default
representation will not change the layout of Inner
.
The Rust
Representation
The Rust
representation is the default representation for nominal types
without a repr
attribute. Using this representation explicitly through a
repr
attribute is guaranteed to be the same as omitting the attribute
entirely.
The only data layout guarantees made by this representation are those required for soundness. They are:
- The fields are properly aligned.
- The fields do not overlap.
- The alignment of the type is at least the maximum alignment of its fields.
Formally, the first guarantee means that the offset of any field is divisible by that field’s alignment. The second guarantee means that the fields can be ordered such that the offset plus the size of any field is less than or equal to the offset of the next field in the ordering. The ordering does not have to be the same as the order in which the fields are specified in the declaration of the type.
Be aware that the second guarantee does not imply that the fields have distinct addresses: zero-sized types may have the same address as other fields in the same struct.
There are no other guarantees of data layout made by this representation.
The C
Representation
The C
representation is designed for dual purposes. One purpose is for
creating types that are interoperable with the C Language. The second purpose is
to create types that you can soundly perform operations on that rely on data
layout such as reinterpreting values as a different type.
Because of this dual purpose, it is possible to create types that are not useful for interfacing with the C programming language.
This representation can be applied to structs, unions, and enums. The exception
is zero-variant enums for which the C
representation is an error.
#[repr(C)]
Structs
The alignment of the struct is the alignment of the most-aligned field in it.
The size and offset of fields is determined by the following algorithm.
Start with a current offset of 0 bytes.
For each field in declaration order in the struct, first determine the size and alignment of the field. If the current offset is not a multiple of the field’s alignment, then add padding bytes to the current offset until it is a multiple of the field’s alignment. The offset for the field is what the current offset is now. Then increase the current offset by the size of the field.
Finally, the size of the struct is the current offset rounded up to the nearest multiple of the struct’s alignment.
Here is this algorithm described in pseudocode.
/// Returns the amount of padding needed after `offset` to ensure that the
/// following address will be aligned to `alignment`.
fn padding_needed_for(offset: usize, alignment: usize) -> usize {
let misalignment = offset % alignment;
if misalignment > 0 {
// round up to next multiple of `alignment`
alignment - misalignment
} else {
// already a multiple of `alignment`
0
}
}
struct.alignment = struct.fields().map(|field| field.alignment).max();
let current_offset = 0;
for field in struct.fields_in_declaration_order() {
// Increase the current offset so that it's a multiple of the alignment
// of this field. For the first field, this will always be zero.
// The skipped bytes are called padding bytes.
current_offset += padding_needed_for(current_offset, field.alignment);
struct[field].offset = current_offset;
current_offset += field.size;
}
struct.size = current_offset + padding_needed_for(current_offset, struct.alignment);
Warning: This pseudocode uses a naive algorithm that ignores overflow issues for
the sake of clarity. To perform memory layout computations in actual code, use
Layout
.
Note: This algorithm can produce zero-sized structs. In C, an empty struct declaration like
struct Foo { }
is illegal. However, both gcc and clang support options to enable such structs, and assign them size zero. C++, in contrast, gives empty structs a size of 1, unless they are inherited from or they are fields that have the[[no_unique_address]]
attribute, in which case they do not increase the overall size of the struct.
#[repr(C)]
Unions
A union declared with #[repr(C)]
will have the same size and alignment as an
equivalent C union declaration in the C language for the target platform.
The union will have a size of the maximum size of all of its fields rounded to
its alignment, and an alignment of the maximum alignment of all of its fields.
These maximums may come from different fields.
#![allow(unused)] fn main() { #[repr(C)] union Union { f1: u16, f2: [u8; 4], } assert_eq!(std::mem::size_of::<Union>(), 4); // From f2 assert_eq!(std::mem::align_of::<Union>(), 2); // From f1 #[repr(C)] union SizeRoundedUp { a: u32, b: [u16; 3], } assert_eq!(std::mem::size_of::<SizeRoundedUp>(), 8); // Size of 6 from b, // rounded up to 8 from // alignment of a. assert_eq!(std::mem::align_of::<SizeRoundedUp>(), 4); // From a }
#[repr(C)]
Field-less Enums
For field-less enums, the C
representation has the size and alignment of
the default enum
size and alignment for the target platform’s C ABI.
Note: The enum representation in C is implementation defined, so this is really a “best guess”. In particular, this may be incorrect when the C code of interest is compiled with certain flags.
Warning: There are crucial differences between an enum
in the C language and
Rust’s field-less enums with this representation. An enum
in C is
mostly a typedef
plus some named constants; in other words, an object of an
enum
type can hold any integer value. For example, this is often used for
bitflags in C
. In contrast, Rust’s field-less enums can only legally hold
the discriminant values, everything else is undefined behavior. Therefore,
using a field-less enum in FFI to model a C enum
is often wrong.
#[repr(C)]
Enums With Fields
The representation of a repr(C)
enum with fields is a repr(C)
struct with
two fields, also called a “tagged union” in C:
- a
repr(C)
version of the enum with all fields removed (“the tag”) - a
repr(C)
union ofrepr(C)
structs for the fields of each variant that had them (“the payload”)
Note: Due to the representation of
repr(C)
structs and unions, if a variant has a single field there is no difference between putting that field directly in the union or wrapping it in a struct; any system which wishes to manipulate such anenum
’s representation may therefore use whichever form is more convenient or consistent for them.
#![allow(unused)] fn main() { // This Enum has the same representation as ... #[repr(C)] enum MyEnum { A(u32), B(f32, u64), C { x: u32, y: u8 }, D, } // ... this struct. #[repr(C)] struct MyEnumRepr { tag: MyEnumDiscriminant, payload: MyEnumFields, } // This is the discriminant enum. #[repr(C)] enum MyEnumDiscriminant { A, B, C, D } // This is the variant union. #[repr(C)] union MyEnumFields { A: MyAFields, B: MyBFields, C: MyCFields, D: MyDFields, } #[repr(C)] #[derive(Copy, Clone)] struct MyAFields(u32); #[repr(C)] #[derive(Copy, Clone)] struct MyBFields(f32, u64); #[repr(C)] #[derive(Copy, Clone)] struct MyCFields { x: u32, y: u8 } // This struct could be omitted (it is a zero-sized type), and it must be in // C/C++ headers. #[repr(C)] #[derive(Copy, Clone)] struct MyDFields; }
Note:
union
s with non-Copy
fields are unstable, see 55149.
Primitive representations
The primitive representations are the representations with the same names as
the primitive integer types. That is: u8
, u16
, u32
, u64
, u128
,
usize
, i8
, i16
, i32
, i64
, i128
, and isize
.
Primitive representations can only be applied to enumerations and have different behavior whether the enum has fields or no fields. It is an error for zero-variant enums to have a primitive representation. Combining two primitive representations together is an error.
Primitive Representation of Field-less Enums
For field-less enums, primitive representations set the size and alignment to
be the same as the primitive type of the same name. For example, a field-less
enum with a u8
representation can only have discriminants between 0 and 255
inclusive.
Primitive Representation of Enums With Fields
The representation of a primitive representation enum is a repr(C)
union of
repr(C)
structs for each variant with a field. The first field of each struct
in the union is the primitive representation version of the enum with all fields
removed (“the tag”) and the remaining fields are the fields of that variant.
Note: This representation is unchanged if the tag is given its own member in the union, should that make manipulation more clear for you (although to follow the C++ standard the tag member should be wrapped in a
struct
).
#![allow(unused)] fn main() { // This enum has the same representation as ... #[repr(u8)] enum MyEnum { A(u32), B(f32, u64), C { x: u32, y: u8 }, D, } // ... this union. #[repr(C)] union MyEnumRepr { A: MyVariantA, B: MyVariantB, C: MyVariantC, D: MyVariantD, } // This is the discriminant enum. #[repr(u8)] #[derive(Copy, Clone)] enum MyEnumDiscriminant { A, B, C, D } #[repr(C)] #[derive(Clone, Copy)] struct MyVariantA(MyEnumDiscriminant, u32); #[repr(C)] #[derive(Clone, Copy)] struct MyVariantB(MyEnumDiscriminant, f32, u64); #[repr(C)] #[derive(Clone, Copy)] struct MyVariantC { tag: MyEnumDiscriminant, x: u32, y: u8 } #[repr(C)] #[derive(Clone, Copy)] struct MyVariantD(MyEnumDiscriminant); }
Note:
union
s with non-Copy
fields are unstable, see 55149.
Combining primitive representations of enums with fields and #[repr(C)]
For enums with fields, it is also possible to combine repr(C)
and a
primitive representation (e.g., repr(C, u8)
). This modifies the repr(C)
by
changing the representation of the discriminant enum to the chosen primitive
instead. So, if you chose the u8
representation, then the discriminant enum
would have a size and alignment of 1 byte.
The discriminant enum from the example earlier then becomes:
#![allow(unused)] fn main() { #[repr(C, u8)] // `u8` was added enum MyEnum { A(u32), B(f32, u64), C { x: u32, y: u8 }, D, } // ... #[repr(u8)] // So `u8` is used here instead of `C` enum MyEnumDiscriminant { A, B, C, D } // ... }
For example, with a repr(C, u8)
enum it is not possible to have 257 unique
discriminants (“tags”) whereas the same enum with only a repr(C)
attribute
will compile without any problems.
Using a primitive representation in addition to repr(C)
can change the size of
an enum from the repr(C)
form:
#![allow(unused)] fn main() { #[repr(C)] enum EnumC { Variant0(u8), Variant1, } #[repr(C, u8)] enum Enum8 { Variant0(u8), Variant1, } #[repr(C, u16)] enum Enum16 { Variant0(u8), Variant1, } // The size of the C representation is platform dependant assert_eq!(std::mem::size_of::<EnumC>(), 8); // One byte for the discriminant and one byte for the value in Enum8::Variant0 assert_eq!(std::mem::size_of::<Enum8>(), 2); // Two bytes for the discriminant and one byte for the value in Enum16::Variant0 // plus one byte of padding. assert_eq!(std::mem::size_of::<Enum16>(), 4); }
The alignment modifiers
The align
and packed
modifiers can be used to respectively raise or lower
the alignment of struct
s and union
s. packed
may also alter the padding
between fields (although it will not alter the padding inside of any field).
On their own, align
and packed
do not provide guarantees about the order
of fields in the layout of a struct or the layout of an enum variant, although
they may be combined with representations (such as C
) which do provide such
guarantees.
The alignment is specified as an integer parameter in the form of
#[repr(align(x))]
or #[repr(packed(x))]
. The alignment value must be a
power of two from 1 up to 229. For packed
, if no value is given,
as in #[repr(packed)]
, then the value is 1.
For align
, if the specified alignment is less than the alignment of the type
without the align
modifier, then the alignment is unaffected.
For packed
, if the specified alignment is greater than the type’s alignment
without the packed
modifier, then the alignment and layout is unaffected.
The alignments of each field, for the purpose of positioning fields, is the
smaller of the specified alignment and the alignment of the field’s type.
Inter-field padding is guaranteed to be the minimum required in order to
satisfy each field’s (possibly altered) alignment (although note that, on its
own, packed
does not provide any guarantee about field ordering). An
important consequence of these rules is that a type with #[repr(packed(1))]
(or #[repr(packed)]
) will have no inter-field padding.
The align
and packed
modifiers cannot be applied on the same type and a
packed
type cannot transitively contain another align
ed type. align
and
packed
may only be applied to the Rust
and C
representations.
The align
modifier can also be applied on an enum
.
When it is, the effect on the enum
’s alignment is the same as if the enum
was wrapped in a newtype struct
with the same align
modifier.
Note: References to unaligned fields are not allowed because it is undefined behavior. When fields are unaligned due to an alignment modifier, consider the following options for using references and dereferences:
#![allow(unused)] fn main() { #[repr(packed)] struct Packed { f1: u8, f2: u16, } let mut e = Packed { f1: 1, f2: 2 }; // Instead of creating a reference to a field, copy the value to a local variable. let x = e.f2; // Or in situations like `println!` which creates a reference, use braces // to change it to a copy of the value. println!("{}", {e.f2}); // Or if you need a pointer, use the unaligned methods for reading and writing // instead of dereferencing the pointer directly. let ptr: *const u16 = &raw const e.f2; let value = unsafe { ptr.read_unaligned() }; let mut_ptr: *mut u16 = &raw mut e.f2; unsafe { mut_ptr.write_unaligned(3) } }
The transparent
Representation
The transparent
representation can only be used on a struct
or an enum
with a single variant that has:
- any number of fields with size 0 and alignment 1 (e.g.
PhantomData<T>
), and - at most one other field.
Structs and enums with this representation have the same layout and ABI as the only non-size 0 non-alignment 1 field, if present, or unit otherwise.
This is different than the C
representation because
a struct with the C
representation will always have the ABI of a C
struct
while, for example, a struct with the transparent
representation with a
primitive field will have the ABI of the primitive field.
Because this representation delegates type layout to another type, it cannot be used with any other representation.
Interior Mutability
Sometimes a type needs to be mutated while having multiple aliases. In Rust this is achieved using a pattern called interior mutability. A type has interior mutability if its internal state can be changed through a shared reference to it. This goes against the usual requirement that the value pointed to by a shared reference is not mutated.
std::cell::UnsafeCell<T>
type is the only allowed way to disable
this requirement. When UnsafeCell<T>
is immutably aliased, it is still safe to
mutate, or obtain a mutable reference to, the T
it contains. As with all
other types, it is undefined behavior to have multiple &mut UnsafeCell<T>
aliases.
Other types with interior mutability can be created by using UnsafeCell<T>
as
a field. The standard library provides a variety of types that provide safe
interior mutability APIs. For example, std::cell::RefCell<T>
uses run-time
borrow checks to ensure the usual rules around multiple references. The
std::sync::atomic
module contains types that wrap a value that is only
accessed with atomic operations, allowing the value to be shared and mutated
across threads.
Subtyping and Variance
Subtyping is implicit and can occur at any stage in type checking or inference. Subtyping is restricted to two cases: variance with respect to lifetimes and between types with higher ranked lifetimes. If we were to erase lifetimes from types, then the only subtyping would be due to type equality.
Consider the following example: string literals always have 'static
lifetime. Nevertheless, we can assign s
to t
:
#![allow(unused)] fn main() { fn bar<'a>() { let s: &'static str = "hi"; let t: &'a str = s; } }
Since 'static
outlives the lifetime parameter 'a
, &'static str
is a
subtype of &'a str
.
Higher-ranked function pointers and trait objects have another subtype relation. They are subtypes of types that are given by substitutions of the higher-ranked lifetimes. Some examples:
#![allow(unused)] fn main() { // Here 'a is substituted for 'static let subtype: &(for<'a> fn(&'a i32) -> &'a i32) = &((|x| x) as fn(&_) -> &_); let supertype: &(fn(&'static i32) -> &'static i32) = subtype; // This works similarly for trait objects let subtype: &(dyn for<'a> Fn(&'a i32) -> &'a i32) = &|x| x; let supertype: &(dyn Fn(&'static i32) -> &'static i32) = subtype; // We can also substitute one higher-ranked lifetime for another let subtype: &(for<'a, 'b> fn(&'a i32, &'b i32))= &((|x, y| {}) as fn(&_, &_)); let supertype: &for<'c> fn(&'c i32, &'c i32) = subtype; }
Variance
Variance is a property that generic types have with respect to their arguments. A generic type’s variance in a parameter is how the subtyping of the parameter affects the subtyping of the type.
F<T>
is covariant overT
ifT
being a subtype ofU
implies thatF<T>
is a subtype ofF<U>
(subtyping “passes through”)F<T>
is contravariant overT
ifT
being a subtype ofU
implies thatF<U>
is a subtype ofF<T>
F<T>
is invariant overT
otherwise (no subtyping relation can be derived)
Variance of types is automatically determined as follows
Type | Variance in 'a | Variance in T |
---|---|---|
&'a T | covariant | covariant |
&'a mut T | covariant | invariant |
*const T | covariant | |
*mut T | invariant | |
[T] and [T; n] | covariant | |
fn() -> T | covariant | |
fn(T) -> () | contravariant | |
std::cell::UnsafeCell<T> | invariant | |
std::marker::PhantomData<T> | covariant | |
dyn Trait<T> + 'a | covariant | invariant |
The variance of other struct
, enum
, and union
types is decided by
looking at the variance of the types of their fields. If the parameter is used
in positions with different variances then the parameter is invariant. For
example the following struct is covariant in 'a
and T
and invariant in 'b
, 'c
,
and U
.
#![allow(unused)] fn main() { use std::cell::UnsafeCell; struct Variance<'a, 'b, 'c, T, U: 'a> { x: &'a U, // This makes `Variance` covariant in 'a, and would // make it covariant in U, but U is used later y: *const T, // Covariant in T z: UnsafeCell<&'b f64>, // Invariant in 'b w: *mut U, // Invariant in U, makes the whole struct invariant f: fn(&'c ()) -> &'c () // Both co- and contravariant, makes 'c invariant // in the struct. } }
When used outside of an struct
, enum
, or union
, the variance for parameters is checked at each location separately.
#![allow(unused)] fn main() { use std::cell::UnsafeCell; fn generic_tuple<'short, 'long: 'short>( // 'long is used inside of a tuple in both a co- and invariant position. x: (&'long u32, UnsafeCell<&'long u32>), ) { // As the variance at these positions is computed separately, // we can freely shrink 'long in the covariant position. let _: (&'short u32, UnsafeCell<&'long u32>) = x; } fn takes_fn_ptr<'short, 'middle: 'short>( // 'middle is used in both a co- and contravariant position. f: fn(&'middle ()) -> &'middle (), ) { // As the variance at these positions is computed separately, // we can freely shrink 'middle in the covariant position // and extend it in the contravariant position. let _: fn(&'static ()) -> &'short () = f; } }
Trait and lifetime bounds
Syntax
TypeParamBounds :
TypeParamBound (+
TypeParamBound )*+
?TypeParamBound :
Lifetime | TraitBoundTraitBound :
?
? ForLifetimes? TypePath
|(
?
? ForLifetimes? TypePath)
LifetimeBounds :
( Lifetime+
)* Lifetime?Lifetime :
LIFETIME_OR_LABEL
|'static
Trait and lifetime bounds provide a way for generic items to restrict which types and lifetimes are used as their parameters. Bounds can be provided on any type in a where clause. There are also shorter forms for certain common cases:
- Bounds written after declaring a generic parameter:
fn f<A: Copy>() {}
is the same asfn f<A>() where A: Copy {}
. - In trait declarations as supertraits:
trait Circle : Shape {}
is equivalent totrait Circle where Self : Shape {}
. - In trait declarations as bounds on associated types:
trait A { type B: Copy; }
is equivalent totrait A where Self::B: Copy { type B; }
.
Bounds on an item must be satisfied when using the item. When type checking and
borrow checking a generic item, the bounds can be used to determine that a
trait is implemented for a type. For example, given Ty: Trait
- In the body of a generic function, methods from
Trait
can be called onTy
values. Likewise associated constants on theTrait
can be used. - Associated types from
Trait
can be used. - Generic functions and types with a
T: Trait
bounds can be used withTy
being used forT
.
#![allow(unused)] fn main() { type Surface = i32; trait Shape { fn draw(&self, surface: Surface); fn name() -> &'static str; } fn draw_twice<T: Shape>(surface: Surface, sh: T) { sh.draw(surface); // Can call method because T: Shape sh.draw(surface); } fn copy_and_draw_twice<T: Copy>(surface: Surface, sh: T) where T: Shape { let shape_copy = sh; // doesn't move sh because T: Copy draw_twice(surface, sh); // Can use generic function because T: Shape } struct Figure<S: Shape>(S, S); fn name_figure<U: Shape>( figure: Figure<U>, // Type Figure<U> is well-formed because U: Shape ) { println!( "Figure of two {}", U::name(), // Can use associated function ); } }
Bounds that don’t use the item’s parameters or higher-ranked lifetimes are checked when the item is defined. It is an error for such a bound to be false.
Copy
, Clone
, and Sized
bounds are also checked for certain generic types when using the item, even if the use does not provide a concrete type.
It is an error to have Copy
or Clone
as a bound on a mutable reference, trait object, or slice.
It is an error to have Sized
as a bound on a trait object or slice.
#![allow(unused)] fn main() { struct A<'a, T> where i32: Default, // Allowed, but not useful i32: Iterator, // Error: `i32` is not an iterator &'a mut T: Copy, // (at use) Error: the trait bound is not satisfied [T]: Sized, // (at use) Error: size cannot be known at compilation { f: &'a T, } struct UsesA<'a, T>(A<'a, T>); }
Trait and lifetime bounds are also used to name trait objects.
?Sized
?
is only used to relax the implicit Sized
trait bound for type parameters or associated types.
?Sized
may not be used as a bound for other types.
Lifetime bounds
Lifetime bounds can be applied to types or to other lifetimes.
The bound 'a: 'b
is usually read as 'a
outlives 'b
.
'a: 'b
means that 'a
lasts at least as long as 'b
, so a reference &'a ()
is valid whenever &'b ()
is valid.
#![allow(unused)] fn main() { fn f<'a, 'b>(x: &'a i32, mut y: &'b i32) where 'a: 'b { y = x; // &'a i32 is a subtype of &'b i32 because 'a: 'b let r: &'b &'a i32 = &&0; // &'b &'a i32 is well formed because 'a: 'b } }
T: 'a
means that all lifetime parameters of T
outlive 'a
.
For example, if 'a
is an unconstrained lifetime parameter, then i32: 'static
and &'static str: 'a
are satisfied, but Vec<&'a ()>: 'static
is not.
Higher-ranked trait bounds
ForLifetimes :
for
GenericParams
Trait bounds may be higher ranked over lifetimes. These bounds specify a bound
that is true for all lifetimes. For example, a bound such as for<'a> &'a T: PartialEq<i32>
would require an implementation like
#![allow(unused)] fn main() { struct T; impl<'a> PartialEq<i32> for &'a T { // ... fn eq(&self, other: &i32) -> bool {true} } }
and could then be used to compare a &'a T
with any lifetime to an i32
.
Only a higher-ranked bound can be used here, because the lifetime of the reference is shorter than any possible lifetime parameter on the function:
#![allow(unused)] fn main() { fn call_on_ref_zero<F>(f: F) where for<'a> F: Fn(&'a i32) { let zero = 0; f(&zero); } }
Higher-ranked lifetimes may also be specified just before the trait: the only difference is the scope of the lifetime parameter, which extends only to the end of the following trait instead of the whole bound. This function is equivalent to the last one.
#![allow(unused)] fn main() { fn call_on_ref_zero<F>(f: F) where F: for<'a> Fn(&'a i32) { let zero = 0; f(&zero); } }
Implied bounds
Lifetime bounds required for types to be well-formed are sometimes inferred.
#![allow(unused)] fn main() { fn requires_t_outlives_a<'a, T>(x: &'a T) {} }
The type parameter T
is required to outlive 'a
for the type &'a T
to be well-formed.
This is inferred because the function signature contains the type &'a T
which is
only valid if T: 'a
holds.
Implied bounds are added for all parameters and outputs of functions. Inside of requires_t_outlives_a
you can assume T: 'a
to hold even if you don’t explicitly specify this:
#![allow(unused)] fn main() { fn requires_t_outlives_a_not_implied<'a, T: 'a>() {} fn requires_t_outlives_a<'a, T>(x: &'a T) { // This compiles, because `T: 'a` is implied by // the reference type `&'a T`. requires_t_outlives_a_not_implied::<'a, T>(); } }
#![allow(unused)] fn main() { fn requires_t_outlives_a_not_implied<'a, T: 'a>() {} fn not_implied<'a, T>() { // This errors, because `T: 'a` is not implied by // the function signature. requires_t_outlives_a_not_implied::<'a, T>(); } }
Only lifetime bounds are implied, trait bounds still have to be explicitly added. The following example therefore causes an error:
#![allow(unused)] fn main() { use std::fmt::Debug; struct IsDebug<T: Debug>(T); // error[E0277]: `T` doesn't implement `Debug` fn doesnt_specify_t_debug<T>(x: IsDebug<T>) {} }
Lifetime bounds are also inferred for type definitions and impl blocks for any type:
#![allow(unused)] fn main() { struct Struct<'a, T> { // This requires `T: 'a` to be well-formed // which is inferred by the compiler. field: &'a T, } enum Enum<'a, T> { // This requires `T: 'a` to be well-formed, // which is inferred by the compiler. // // Note that `T: 'a` is required even when only // using `Enum::OtherVariant`. SomeVariant(&'a T), OtherVariant, } trait Trait<'a, T: 'a> {} // This would error because `T: 'a` is not implied by any type // in the impl header. // impl<'a, T> Trait<'a, T> for () {} // This compiles as `T: 'a` is implied by the self type `&'a T`. impl<'a, T> Trait<'a, T> for &'a T {} }
Type coercions
Type coercions are implicit operations that change the type of a value. They happen automatically at specific locations and are highly restricted in what types actually coerce.
Any conversions allowed by coercion can also be explicitly performed by the
type cast operator, as
.
Coercions are originally defined in RFC 401 and expanded upon in RFC 1558.
Coercion sites
A coercion can only occur at certain coercion sites in a program; these are typically places where the desired type is explicit or can be derived by propagation from explicit types (without type inference). Possible coercion sites are:
-
let
statements where an explicit type is given.For example,
&mut 42
is coerced to have type&i8
in the following:#![allow(unused)] fn main() { let _: &i8 = &mut 42; }
-
static
andconst
item declarations (similar tolet
statements). -
Arguments for function calls
The value being coerced is the actual parameter, and it is coerced to the type of the formal parameter.
For example,
&mut 42
is coerced to have type&i8
in the following:fn bar(_: &i8) { } fn main() { bar(&mut 42); }
For method calls, the receiver (
self
parameter) type is coerced differently, see the documentation on method-call expressions for details. -
Instantiations of struct, union, or enum variant fields
For example,
&mut 42
is coerced to have type&i8
in the following:struct Foo<'a> { x: &'a i8 } fn main() { Foo { x: &mut 42 }; }
-
Function results—either the final line of a block if it is not semicolon-terminated or any expression in a
return
statementFor example,
x
is coerced to have type&dyn Display
in the following:#![allow(unused)] fn main() { use std::fmt::Display; fn foo(x: &u32) -> &dyn Display { x } }
If the expression in one of these coercion sites is a coercion-propagating expression, then the relevant sub-expressions in that expression are also coercion sites. Propagation recurses from these new coercion sites. Propagating expressions and their relevant sub-expressions are:
-
Array literals, where the array has type
[U; n]
. Each sub-expression in the array literal is a coercion site for coercion to typeU
. -
Array literals with repeating syntax, where the array has type
[U; n]
. The repeated sub-expression is a coercion site for coercion to typeU
. -
Tuples, where a tuple is a coercion site to type
(U_0, U_1, ..., U_n)
. Each sub-expression is a coercion site to the respective type, e.g. the zeroth sub-expression is a coercion site to typeU_0
. -
Parenthesized sub-expressions (
(e)
): if the expression has typeU
, then the sub-expression is a coercion site toU
. -
Blocks: if a block has type
U
, then the last expression in the block (if it is not semicolon-terminated) is a coercion site toU
. This includes blocks which are part of control flow statements, such asif
/else
, if the block has a known type.
Coercion types
Coercion is allowed between the following types:
-
T
toU
ifT
is a subtype ofU
(reflexive case) -
T_1
toT_3
whereT_1
coerces toT_2
andT_2
coerces toT_3
(transitive case)Note that this is not fully supported yet.
-
&mut T
to&T
-
*mut T
to*const T
-
&T
to*const T
-
&mut T
to*mut T
-
&T
or&mut T
to&U
ifT
implementsDeref<Target = U>
. For example:use std::ops::Deref; struct CharContainer { value: char, } impl Deref for CharContainer { type Target = char; fn deref<'a>(&'a self) -> &'a char { &self.value } } fn foo(arg: &char) {} fn main() { let x = &mut CharContainer { value: 'y' }; foo(x); //&mut CharContainer is coerced to &char. }
-
&mut T
to&mut U
ifT
implementsDerefMut<Target = U>
. -
TyCtor(
T
) to TyCtor(U
), where TyCtor(T
) is one of&T
&mut T
*const T
*mut T
Box<T>
and where
U
can be obtained fromT
by unsized coercion. -
Function item types to
fn
pointers -
Non capturing closures to
fn
pointers -
!
to anyT
Unsized Coercions
The following coercions are called unsized coercions
, since they
relate to converting sized types to unsized types, and are permitted in a few
cases where other coercions are not, as described above. They can still happen
anywhere else a coercion can occur.
Two traits, Unsize
and CoerceUnsized
, are used
to assist in this process and expose it for library use. The following
coercions are built-ins and, if T
can be coerced to U
with one of them, then
an implementation of Unsize<U>
for T
will be provided:
-
[T; n]
to[T]
. -
T
todyn U
, whenT
implementsU + Sized
, andU
is object safe. -
Foo<..., T, ...>
toFoo<..., U, ...>
, when:Foo
is a struct.T
implementsUnsize<U>
.- The last field of
Foo
has a type involvingT
. - If that field has type
Bar<T>
, thenBar<T>
implementsUnsized<Bar<U>>
. - T is not part of the type of any other fields.
Additionally, a type Foo<T>
can implement CoerceUnsized<Foo<U>>
when T
implements Unsize<U>
or CoerceUnsized<Foo<U>>
. This allows it to provide an
unsized coercion to Foo<U>
.
Note: While the definition of the unsized coercions and their implementation has been stabilized, the traits themselves are not yet stable and therefore can’t be used directly in stable Rust.
Least upper bound coercions
In some contexts, the compiler must coerce together multiple types to try and find the most general type. This is called a “Least Upper Bound” coercion. LUB coercion is used and only used in the following situations:
- To find the common type for a series of if branches.
- To find the common type for a series of match arms.
- To find the common type for array elements.
- To find the type for the return type of a closure with multiple return statements.
- To check the type for the return type of a function with multiple return statements.
In each such case, there are a set of types T0..Tn
to be mutually coerced
to some target type T_t
, which is unknown to start. Computing the LUB
coercion is done iteratively. The target type T_t
begins as the type T0
.
For each new type Ti
, we consider whether
- If
Ti
can be coerced to the current target typeT_t
, then no change is made. - Otherwise, check whether
T_t
can be coerced toTi
; if so, theT_t
is changed toTi
. (This check is also conditioned on whether all of the source expressions considered thus far have implicit coercions.) - If not, try to compute a mutual supertype of
T_t
andTi
, which will become the new target type.
Examples:
#![allow(unused)] fn main() { let (a, b, c) = (0, 1, 2); // For if branches let bar = if true { a } else if false { b } else { c }; // For match arms let baw = match 42 { 0 => a, 1 => b, _ => c, }; // For array elements let bax = [a, b, c]; // For closure with multiple return statements let clo = || { if true { a } else if false { b } else { c } }; let baz = clo(); // For type checking of function with multiple return statements fn foo() -> i32 { let (a, b, c) = (0, 1, 2); match 42 { 0 => a, 1 => b, _ => c, } } }
In these examples, types of the ba*
are found by LUB coercion. And the
compiler checks whether LUB coercion result of a
, b
, c
is i32
in the
processing of the function foo
.
Caveat
This description is obviously informal. Making it more precise is expected to proceed as part of a general effort to specify the Rust type checker more precisely.
Destructors
When an initialized variable or temporary goes out of scope, its destructor is run, or it is dropped. Assignment also runs the destructor of its left-hand operand, if it’s initialized. If a variable has been partially initialized, only its initialized fields are dropped.
The destructor of a type T
consists of:
- If
T: Drop
, calling<T as std::ops::Drop>::drop
- Recursively running the destructor of all of its fields.
- The fields of a struct are dropped in declaration order.
- The fields of the active enum variant are dropped in declaration order.
- The fields of a tuple are dropped in order.
- The elements of an array or owned slice are dropped from the first element to the last.
- The variables that a closure captures by move are dropped in an unspecified order.
- Trait objects run the destructor of the underlying type.
- Other types don’t result in any further drops.
If a destructor must be run manually, such as when implementing your own smart
pointer, std::ptr::drop_in_place
can be used.
Some examples:
#![allow(unused)] fn main() { struct PrintOnDrop(&'static str); impl Drop for PrintOnDrop { fn drop(&mut self) { println!("{}", self.0); } } let mut overwritten = PrintOnDrop("drops when overwritten"); overwritten = PrintOnDrop("drops when scope ends"); let tuple = (PrintOnDrop("Tuple first"), PrintOnDrop("Tuple second")); let moved; // No destructor run on assignment. moved = PrintOnDrop("Drops when moved"); // Drops now, but is then uninitialized. moved; // Uninitialized does not drop. let uninitialized: PrintOnDrop; // After a partial move, only the remaining fields are dropped. let mut partial_move = (PrintOnDrop("first"), PrintOnDrop("forgotten")); // Perform a partial move, leaving only `partial_move.0` initialized. core::mem::forget(partial_move.1); // When partial_move's scope ends, only the first field is dropped. }
Drop scopes
Each variable or temporary is associated to a drop scope. When control flow leaves a drop scope all variables associated to that scope are dropped in reverse order of declaration (for variables) or creation (for temporaries).
Drop scopes are determined after replacing for
, if let
, and
while let
expressions with the equivalent expressions using match
.
Overloaded operators are not distinguished from built-in operators and binding modes are not considered.
Given a function, or closure, there are drop scopes for:
- The entire function
- Each statement
- Each expression
- Each block, including the function body
- In the case of a block expression, the scope for the block and the expression are the same scope.
- Each arm of a
match
expression
Drop scopes are nested within one another as follows. When multiple scopes are left at once, such as when returning from a function, variables are dropped from the inside outwards.
- The entire function scope is the outer most scope.
- The function body block is contained within the scope of the entire function.
- The parent of the expression in an expression statement is the scope of the statement.
- The parent of the initializer of a
let
statement is thelet
statement’s scope.
- The parent of a statement scope is the scope of the block that contains the statement.
- The parent of the expression for a
match
guard is the scope of the arm that the guard is for.
- The parent of the expression after the
=>
in amatch
expression is the scope of the arm that it’s in.
- The parent of the arm scope is the scope of the
match
expression that it belongs to.
- The parent of all other scopes is the scope of the immediately enclosing expression.
Scopes of function parameters
All function parameters are in the scope of the entire function body, so are dropped last when evaluating the function. Each actual function parameter is dropped after any bindings introduced in that parameter’s pattern.
#![allow(unused)] fn main() { struct PrintOnDrop(&'static str); impl Drop for PrintOnDrop { fn drop(&mut self) { println!("drop({})", self.0); } } // Drops `y`, then the second parameter, then `x`, then the first parameter fn patterns_in_parameters( (x, _): (PrintOnDrop, PrintOnDrop), (_, y): (PrintOnDrop, PrintOnDrop), ) {} // drop order is 3 2 0 1 patterns_in_parameters( (PrintOnDrop("0"), PrintOnDrop("1")), (PrintOnDrop("2"), PrintOnDrop("3")), ); }
Scopes of local variables
Local variables declared in a let
statement are associated to the scope of
the block that contains the let
statement. Local variables declared in a
match
expression are associated to the arm scope of the match
arm that they
are declared in.
#![allow(unused)] fn main() { struct PrintOnDrop(&'static str); impl Drop for PrintOnDrop { fn drop(&mut self) { println!("drop({})", self.0); } } let declared_first = PrintOnDrop("Dropped last in outer scope"); { let declared_in_block = PrintOnDrop("Dropped in inner scope"); } let declared_last = PrintOnDrop("Dropped first in outer scope"); }
If multiple patterns are used in the same arm for a match
expression, then an
unspecified pattern will be used to determine the drop order.
Temporary scopes
The temporary scope of an expression is the scope that is used for the temporary variable that holds the result of that expression when used in a place context, unless it is promoted.
Apart from lifetime extension, the temporary scope of an expression is the smallest scope that contains the expression and is one of the following:
- The entire function.
- A statement.
- The body of an
if
,while
orloop
expression. - The
else
block of anif
expression. - The condition expression of an
if
orwhile
expression, or amatch
guard. - The body expression for a match arm.
- The second operand of a lazy boolean expression.
Notes:
Temporaries that are created in the final expression of a function body are dropped after any named variables bound in the function body. Their drop scope is the entire function, as there is no smaller enclosing temporary scope.
The scrutinee of a
match
expression is not a temporary scope, so temporaries in the scrutinee can be dropped after thematch
expression. For example, the temporary for1
inmatch 1 { ref mut z => z };
lives until the end of the statement.
Some examples:
#![allow(unused)] fn main() { struct PrintOnDrop(&'static str); impl Drop for PrintOnDrop { fn drop(&mut self) { println!("drop({})", self.0); } } let local_var = PrintOnDrop("local var"); // Dropped once the condition has been evaluated if PrintOnDrop("If condition").0 == "If condition" { // Dropped at the end of the block PrintOnDrop("If body").0 } else { unreachable!() }; // Dropped at the end of the statement (PrintOnDrop("first operand").0 == "" // Dropped at the ) || PrintOnDrop("second operand").0 == "") // Dropped at the end of the expression || PrintOnDrop("third operand").0 == ""; // Dropped at the end of the function, after local variables. // Changing this to a statement containing a return expression would make the // temporary be dropped before the local variables. Binding to a variable // which is then returned would also make the temporary be dropped first. match PrintOnDrop("Matched value in final expression") { // Dropped once the condition has been evaluated _ if PrintOnDrop("guard condition").0 == "" => (), _ => (), } }
Operands
Temporaries are also created to hold the result of operands to an expression while the other operands are evaluated. The temporaries are associated to the scope of the expression with that operand. Since the temporaries are moved from once the expression is evaluated, dropping them has no effect unless one of the operands to an expression breaks out of the expression, returns, or panics.
#![allow(unused)] fn main() { struct PrintOnDrop(&'static str); impl Drop for PrintOnDrop { fn drop(&mut self) { println!("drop({})", self.0); } } loop { // Tuple expression doesn't finish evaluating so operands drop in reverse order ( PrintOnDrop("Outer tuple first"), PrintOnDrop("Outer tuple second"), ( PrintOnDrop("Inner tuple first"), PrintOnDrop("Inner tuple second"), break, ), PrintOnDrop("Never created"), ); } }
Constant promotion
Promotion of a value expression to a 'static
slot occurs when the expression
could be written in a constant and borrowed, and that borrow could be dereferenced
where
the expression was originally written, without changing the runtime behavior.
That is, the promoted expression can be evaluated at compile-time and the
resulting value does not contain interior mutability or destructors (these
properties are determined based on the value where possible, e.g. &None
always has the type &'static Option<_>
, as it contains nothing disallowed).
Temporary lifetime extension
Note: The exact rules for temporary lifetime extension are subject to change. This is describing the current behavior only.
The temporary scopes for expressions in let
statements are sometimes
extended to the scope of the block containing the let
statement. This is
done when the usual temporary scope would be too small, based on certain
syntactic rules. For example:
#![allow(unused)] fn main() { let x = &mut 0; // Usually a temporary would be dropped by now, but the temporary for `0` lives // to the end of the block. println!("{}", x); }
If a borrow, dereference, field, or tuple indexing expression has an extended temporary scope then so does its operand. If an indexing expression has an extended temporary scope then the indexed expression also has an extended temporary scope.
Extending based on patterns
An extending pattern is either
- An identifier pattern that binds by reference or mutable reference.
- A struct, tuple, tuple struct, or slice pattern where at least one of the direct subpatterns is an extending pattern.
So ref x
, V(ref x)
and [ref x, y]
are all extending patterns, but x
,
&ref x
and &(ref x,)
are not.
If the pattern in a let
statement is an extending pattern then the temporary
scope of the initializer expression is extended.
Extending based on expressions
For a let statement with an initializer, an extending expression is an expression which is one of the following:
- The initializer expression.
- The operand of an extending borrow expression.
- The operand(s) of an extending array, cast, braced struct, or tuple expression.
- The final expression of any extending block expression.
So the borrow expressions in &mut 0
, (&1, &mut 2)
, and Some { 0: &mut 3 }
are all extending expressions. The borrows in &0 + &1
and Some(&mut 0)
are
not: the latter is syntactically a function call expression.
The operand of any extending borrow expression has its temporary scope extended.
Examples
Here are some examples where expressions have extended temporary scopes:
#![allow(unused)] fn main() { fn temp() {} trait Use { fn use_temp(&self) -> &Self { self } } impl Use for () {} // The temporary that stores the result of `temp()` lives in the same scope // as x in these cases. let x = &temp(); let x = &temp() as &dyn Send; let x = (&*&temp(),); let x = { [Some { 0: &temp(), }] }; let ref x = temp(); let ref x = *&temp(); x; }
Here are some examples where expressions don’t have extended temporary scopes:
#![allow(unused)] fn main() { fn temp() {} trait Use { fn use_temp(&self) -> &Self { self } } impl Use for () {} // The temporary that stores the result of `temp()` only lives until the // end of the let statement in these cases. let x = Some(&temp()); // ERROR let x = (&temp()).use_temp(); // ERROR x; }
Not running destructors
std::mem::forget
can be used to prevent the destructor of a variable from being run,
and std::mem::ManuallyDrop
provides a wrapper to prevent a
variable or field from being dropped automatically.
Note: Preventing a destructor from being run via
std::mem::forget
or other means is safe even if it has a type that isn’t'static
. Besides the places where destructors are guaranteed to run as defined by this document, types may not safely rely on a destructor being run for soundness.
Lifetime elision
Rust has rules that allow lifetimes to be elided in various places where the compiler can infer a sensible default choice.
Lifetime elision in functions
In order to make common patterns more ergonomic, lifetime arguments can be
elided in function item, function pointer, and closure trait signatures.
The following rules are used to infer lifetime parameters for elided lifetimes.
It is an error to elide lifetime parameters that cannot be inferred. The
placeholder lifetime, '_
, can also be used to have a lifetime inferred in the
same way. For lifetimes in paths, using '_
is preferred. Trait object
lifetimes follow different rules discussed
below.
- Each elided lifetime in the parameters becomes a distinct lifetime parameter.
- If there is exactly one lifetime used in the parameters (elided or not), that lifetime is assigned to all elided output lifetimes.
In method signatures there is another rule
- If the receiver has type
&Self
or&mut Self
, then the lifetime of that reference toSelf
is assigned to all elided output lifetime parameters.
Examples:
#![allow(unused)] fn main() { trait T {} trait ToCStr {} struct Thing<'a> {f: &'a i32} struct Command; trait Example { fn print1(s: &str); // elided fn print2(s: &'_ str); // also elided fn print3<'a>(s: &'a str); // expanded fn debug1(lvl: usize, s: &str); // elided fn debug2<'a>(lvl: usize, s: &'a str); // expanded fn substr1(s: &str, until: usize) -> &str; // elided fn substr2<'a>(s: &'a str, until: usize) -> &'a str; // expanded fn get_mut1(&mut self) -> &mut dyn T; // elided fn get_mut2<'a>(&'a mut self) -> &'a mut dyn T; // expanded fn args1<T: ToCStr>(&mut self, args: &[T]) -> &mut Command; // elided fn args2<'a, 'b, T: ToCStr>(&'a mut self, args: &'b [T]) -> &'a mut Command; // expanded fn new1(buf: &mut [u8]) -> Thing<'_>; // elided - preferred fn new2(buf: &mut [u8]) -> Thing; // elided fn new3<'a>(buf: &'a mut [u8]) -> Thing<'a>; // expanded } type FunPtr1 = fn(&str) -> &str; // elided type FunPtr2 = for<'a> fn(&'a str) -> &'a str; // expanded type FunTrait1 = dyn Fn(&str) -> &str; // elided type FunTrait2 = dyn for<'a> Fn(&'a str) -> &'a str; // expanded }
#![allow(unused)] fn main() { // The following examples show situations where it is not allowed to elide the // lifetime parameter. trait Example { // Cannot infer, because there are no parameters to infer from. fn get_str() -> &str; // ILLEGAL // Cannot infer, ambiguous if it is borrowed from the first or second parameter. fn frob(s: &str, t: &str) -> &str; // ILLEGAL } }
Default trait object lifetimes
The assumed lifetime of references held by a trait object is called its default object lifetime bound. These were defined in RFC 599 and amended in RFC 1156.
These default object lifetime bounds are used instead of the lifetime parameter
elision rules defined above when the lifetime bound is omitted entirely. If
'_
is used as the lifetime bound then the bound follows the usual elision
rules.
If the trait object is used as a type argument of a generic type then the containing type is first used to try to infer a bound.
- If there is a unique bound from the containing type then that is the default
- If there is more than one bound from the containing type then an explicit bound must be specified
If neither of those rules apply, then the bounds on the trait are used:
- If the trait is defined with a single lifetime bound then that bound is used.
- If
'static
is used for any lifetime bound then'static
is used. - If the trait has no lifetime bounds, then the lifetime is inferred in
expressions and is
'static
outside of expressions.
#![allow(unused)] fn main() { // For the following trait... trait Foo { } // These two are the same because Box<T> has no lifetime bound on T type T1 = Box<dyn Foo>; type T2 = Box<dyn Foo + 'static>; // ...and so are these: impl dyn Foo {} impl dyn Foo + 'static {} // ...so are these, because &'a T requires T: 'a type T3<'a> = &'a dyn Foo; type T4<'a> = &'a (dyn Foo + 'a); // std::cell::Ref<'a, T> also requires T: 'a, so these are the same type T5<'a> = std::cell::Ref<'a, dyn Foo>; type T6<'a> = std::cell::Ref<'a, dyn Foo + 'a>; }
#![allow(unused)] fn main() { // This is an example of an error. trait Foo { } struct TwoBounds<'a, 'b, T: ?Sized + 'a + 'b> { f1: &'a i32, f2: &'b i32, f3: T, } type T7<'a, 'b> = TwoBounds<'a, 'b, dyn Foo>; // ^^^^^^^ // Error: the lifetime bound for this object type cannot be deduced from context }
Note that the innermost object sets the bound, so &'a Box<dyn Foo>
is still
&'a Box<dyn Foo + 'static>
.
#![allow(unused)] fn main() { // For the following trait... trait Bar<'a>: 'a { } // ...these two are the same: type T1<'a> = Box<dyn Bar<'a>>; type T2<'a> = Box<dyn Bar<'a> + 'a>; // ...and so are these: impl<'a> dyn Bar<'a> {} impl<'a> dyn Bar<'a> + 'a {} }
'static
lifetime elision
Both constant and static declarations of reference types have implicit
'static
lifetimes unless an explicit lifetime is specified. As such, the
constant declarations involving 'static
above may be written without the
lifetimes.
#![allow(unused)] fn main() { // STRING: &'static str const STRING: &str = "bitstring"; struct BitsNStrings<'a> { mybits: [u32; 2], mystring: &'a str, } // BITS_N_STRINGS: BitsNStrings<'static> const BITS_N_STRINGS: BitsNStrings<'_> = BitsNStrings { mybits: [1, 2], mystring: STRING, }; }
Note that if the static
or const
items include function or closure
references, which themselves include references, the compiler will first try
the standard elision rules. If it is unable to resolve the lifetimes by its
usual rules, then it will error. By way of example:
#![allow(unused)] fn main() { struct Foo; struct Bar; struct Baz; fn somefunc(a: &Foo, b: &Bar, c: &Baz) -> usize {42} // Resolved as `for<'a> fn(&'a str) -> &'a str`. const RESOLVED_SINGLE: fn(&str) -> &str = |x| x; // Resolved as `for<'a, 'b, 'c> Fn(&'a Foo, &'b Bar, &'c Baz) -> usize`. const RESOLVED_MULTIPLE: &dyn Fn(&Foo, &Bar, &Baz) -> usize = &somefunc; }
#![allow(unused)] fn main() { struct Foo; struct Bar; struct Baz; fn somefunc<'a,'b>(a: &'a Foo, b: &'b Bar) -> &'a Baz {unimplemented!()} // There is insufficient information to bound the return reference lifetime // relative to the argument lifetimes, so this is an error. const RESOLVED_STATIC: &dyn Fn(&Foo, &Bar) -> &Baz = &somefunc; // ^ // this function's return type contains a borrowed value, but the signature // does not say whether it is borrowed from argument 1 or argument 2 }
Special types and traits
Certain types and traits that exist in the standard library are known to the Rust compiler. This chapter documents the special features of these types and traits.
Box<T>
Box<T>
has a few special features that Rust doesn’t currently allow for user
defined types.
- The dereference operator for
Box<T>
produces a place which can be moved from. This means that the*
operator and the destructor ofBox<T>
are built-in to the language. - Methods can take
Box<Self>
as a receiver. - A trait may be implemented for
Box<T>
in the same crate asT
, which the orphan rules prevent for other generic types.
Rc<T>
Methods can take Rc<Self>
as a receiver.
Arc<T>
Methods can take Arc<Self>
as a receiver.
Pin<P>
Methods can take Pin<P>
as a receiver.
UnsafeCell<T>
std::cell::UnsafeCell<T>
is used for interior mutability. It ensures that
the compiler doesn’t perform optimisations that are incorrect for such types.
It also ensures that static
items which have a type with interior
mutability aren’t placed in memory marked as read only.
PhantomData<T>
std::marker::PhantomData<T>
is a zero-sized, minimum alignment, type that
is considered to own a T
for the purposes of variance, drop check, and
auto traits.
Operator Traits
The traits in std::ops
and std::cmp
are used to overload operators,
indexing expressions, and call expressions.
Deref
and DerefMut
As well as overloading the unary *
operator, Deref
and DerefMut
are
also used in method resolution and deref coercions.
Drop
The Drop
trait provides a destructor, to be run whenever a value of this
type is to be destroyed.
Copy
The Copy
trait changes the semantics of a type implementing it. Values
whose type implements Copy
are copied rather than moved upon assignment.
Copy
can only be implemented for types which do not implement Drop
, and whose fields are all Copy
.
For enums, this means all fields of all variants have to be Copy
.
For unions, this means all variants have to be Copy
.
Copy
is implemented by the compiler for
- Tuples of
Copy
types - Function pointers
- Function items
- Closures that capture no values or that only capture values of
Copy
types
Clone
The Clone
trait is a supertrait of Copy
, so it also needs compiler
generated implementations. It is implemented by the compiler for the following
types:
- Types with a built-in
Copy
implementation (see above) - Tuples of
Clone
types - Closures that only capture values of
Clone
types or capture no values from the environment
Send
The Send
trait indicates that a value of this type is safe to send from one
thread to another.
Sync
The Sync
trait indicates that a value of this type is safe to share between
multiple threads. This trait must be implemented for all types used in
immutable static
items.
Termination
The Termination
trait indicates the acceptable return types for the main function and test functions.
Auto traits
The Send
, Sync
, Unpin
, UnwindSafe
, and RefUnwindSafe
traits are auto
traits. Auto traits have special properties.
If no explicit implementation or negative implementation is written out for an auto trait for a given type, then the compiler implements it automatically according to the following rules:
&T
,&mut T
,*const T
,*mut T
,[T; n]
, and[T]
implement the trait ifT
does.- Function item types and function pointers automatically implement the trait.
- Structs, enums, unions, and tuples implement the trait if all of their fields do.
- Closures implement the trait if the types of all of their captures do. A
closure that captures a
T
by shared reference and aU
by value implements any auto traits that both&T
andU
do.
For generic types (counting the built-in types above as generic over T
), if a
generic implementation is available, then the compiler does not automatically
implement it for types that could use the implementation except that they do not
meet the requisite trait bounds. For instance, the standard library implements
Send
for all &T
where T
is Sync
; this means that the compiler will not
implement Send
for &T
if T
is Send
but not Sync
.
Auto traits can also have negative implementations, shown as impl !AutoTrait for T
in the standard library documentation, that override the automatic
implementations. For example *mut T
has a negative implementation of Send
,
and so *mut T
is not Send
, even if T
is. There is currently no stable way
to specify additional negative implementations; they exist only in the standard
library.
Auto traits may be added as an additional bound to any trait object, even
though normally only one trait is allowed. For instance, Box<dyn Debug + Send + UnwindSafe>
is a valid type.
Sized
The Sized
trait indicates that the size of this type is known at compile-time; that is, it’s not a dynamically sized type.
Type parameters (except Self
in traits) are Sized
by default, as are associated types.
Sized
is always implemented automatically by the compiler, not by implementation items.
These implicit Sized
bounds may be relaxed by using the special ?Sized
bound.
Names
An entity is a language construct that can be referred to in some way within the source program, usually via a path. Entities include types, items, generic parameters, variable bindings, loop labels, lifetimes, fields, attributes, and lints.
A declaration is a syntactical construct that can introduce a name to refer to an entity. Entity names are valid within a scope — a region of source text where that name may be referenced.
Some entities are explicitly declared in the source code, and some are implicitly declared as part of the language or compiler extensions.
Paths are used to refer to an entity, possibly in another module or type. Lifetimes and loop labels use a dedicated syntax using a leading quote.
Names are segregated into different namespaces, allowing entities in different namespaces to share the same name without conflict.
Name resolution is the compile-time process of tying paths, identifiers, and labels to entity declarations.
Access to certain names may be restricted based on their visibility.
Explicitly declared entities
Entities that explicitly introduce a name in the source code are:
- Items:
- Module declarations
- External crate declarations
- Use declarations
- Function declarations and function parameters
- Type aliases
- struct, union, enum, enum variant declarations, and their named fields
- Constant item declarations
- Static item declarations
- Trait item declarations and their associated items
- External block items
macro_rules
declarations and matcher metavariables- Implementation associated items
- Expressions:
- Generic parameters
- Higher ranked trait bounds
let
statement pattern bindings- The
macro_use
attribute can introduce macro names from another crate - The
macro_export
attribute can introduce an alias for the macro into the crate root
Additionally, macro invocations and attributes can introduce names by expanding to one of the above items.
Implicitly declared entities
The following entities are implicitly defined by the language, or are introduced by compiler options and extensions:
- Language prelude:
- Boolean type —
bool
- Textual types —
char
andstr
- Integer types —
i8
,i16
,i32
,i64
,i128
,u8
,u16
,u32
,u64
,u128
- Machine-dependent integer types —
usize
andisize
- floating-point types —
f32
andf64
- Boolean type —
- Built-in attributes
- Standard library prelude items, attributes, and macros
- Standard library crates in the root module
- External crates linked by the compiler
- Tool attributes
- Lints and tool lint attributes
- Derive helper attributes are valid within an item without being explicitly imported
- The
'static
lifetime
Additionally, the crate root module does not have a name, but can be referred to with certain path qualifiers or aliases.
Namespaces
A namespace is a logical grouping of declared names. Names are segregated into separate namespaces based on the kind of entity the name refers to. Namespaces allow the occurrence of a name in one namespace to not conflict with the same name in another namespace.
There are several different namespaces that each contain different kinds of entities. The usage of a name will look for the declaration of that name in different namespaces, based on the context, as described in the name resolution chapter.
The following is a list of namespaces, with their corresponding entities:
- Type Namespace
- Module declarations
- External crate declarations
- External crate prelude items
- Struct, union, enum, enum variant declarations
- Trait item declarations
- Type aliases
- Associated type declarations
- Built-in types: boolean, numeric, and textual
- Generic type parameters
Self
type- Tool attribute modules
- Value Namespace
- Function declarations
- Constant item declarations
- Static item declarations
- Struct constructors
- Enum variant constructors
Self
constructors- Generic const parameters
- Associated const declarations
- Associated function declarations
- Local bindings —
let
,if let
,while let
,for
,match
arms, function parameters, closure parameters - Captured closure variables
- Macro Namespace
- Lifetime Namespace
- Label Namespace
An example of how overlapping names in different namespaces can be used unambiguously:
#![allow(unused)] fn main() { // Foo introduces a type in the type namespace and a constructor in the value // namespace. struct Foo(u32); // The `Foo` macro is declared in the macro namespace. macro_rules! Foo { () => {}; } // `Foo` in the `f` parameter type refers to `Foo` in the type namespace. // `'Foo` introduces a new lifetime in the lifetime namespace. fn example<'Foo>(f: Foo) { // `Foo` refers to the `Foo` constructor in the value namespace. let ctor = Foo; // `Foo` refers to the `Foo` macro in the macro namespace. Foo!{} // `'Foo` introduces a label in the label namespace. 'Foo: loop { // `'Foo` refers to the `'Foo` lifetime parameter, and `Foo` // refers to the type namespace. let x: &'Foo Foo; // `'Foo` refers to the label. break 'Foo; } } }
Named entities without a namespace
The following entities have explicit names, but the names are not a part of any specific namespace.
Fields
Even though struct, enum, and union fields are named, the named fields do not live in an explicit namespace. They can only be accessed via a field expression, which only inspects the field names of the specific type being accessed.
Use declarations
A use declaration has named aliases that it imports into scope, but the
use
item itself does not belong to a specific namespace. Instead, it can
introduce aliases into multiple namespaces, depending on the item kind being
imported.
Sub-namespaces
The macro namespace is split into two sub-namespaces: one for bang-style macros and one for attributes. When an attribute is resolved, any bang-style macros in scope will be ignored. And conversely resolving a bang-style macro will ignore attribute macros in scope. This prevents one style from shadowing another.
For example, the cfg
attribute and the cfg
macro are two different entities with the same name in the macro namespace, but they can still be used in their respective context.
It is still an error for a use
import to shadow another macro, regardless of their sub-namespaces.
Scopes
A scope is the region of source text where a named entity may be referenced with that name. The following sections provide details on the scoping rules and behavior, which depend on the kind of entity and where it is declared. The process of how names are resolved to entities is described in the name resolution chapter. More information on “drop scopes” used for the purpose of running destructors may be found in the destructors chapter.
Item scopes
The name of an item declared directly in a module has a scope that extends from the start of the module to the end of the module. These items are also members of the module and can be referred to with a path leading from their module.
The name of an item declared as a statement has a scope that extends from the start of the block the item statement is in until the end of the block.
It is an error to introduce an item with a duplicate name of another item in the same namespace within the same module or block. Asterisk glob imports have special behavior for dealing with duplicate names and shadowing, see the linked chapter for more details. Items in a module may shadow items in a prelude.
Item names from outer modules are not in scope within a nested module. A path may be used to refer to an item in another module.
Associated item scopes
Associated items are not scoped and can only be referred to by using a path leading from the type or trait they are associated with. Methods can also be referred to via call expressions.
Similar to items within a module or block, it is an error to introduce an item within a trait or implementation that is a duplicate of another item in the trait or impl in the same namespace.
Pattern binding scopes
The scope of a local variable pattern binding depends on where it is used:
let
statement bindings range from just after thelet
statement until the end of the block where it is declared.- Function parameter bindings are within the body of the function.
- Closure parameter bindings are within the closure body.
for
andwhile let
bindings are within the loop body.if let
bindings are within the consequent block.match
arms bindings are within the match guard and the match arm expression.
Local variable scopes do not extend into item declarations.
Pattern binding shadowing
Pattern bindings are allowed to shadow any name in scope with the following exceptions which are an error:
- Const generic parameters
- Static items
- Const items
- Constructors for structs and enums
The following example illustrates how local bindings can shadow item declarations:
#![allow(unused)] fn main() { fn shadow_example() { // Since there are no local variables in scope yet, this resolves to the function. foo(); // prints `function` let foo = || println!("closure"); fn foo() { println!("function"); } // This resolves to the local closure since it shadows the item. foo(); // prints `closure` } }
Generic parameter scopes
Generic parameters are declared in a GenericParams list. The scope of a generic parameter is within the item it is declared on.
All parameters are in scope within the generic parameter list regardless of the order they are declared. The following shows some examples where a parameter may be referenced before it is declared:
#![allow(unused)] fn main() { // The 'b bound is referenced before it is declared. fn params_scope<'a: 'b, 'b>() {} trait SomeTrait<const Z: usize> {} // The const N is referenced in the trait bound before it is declared. fn f<T: SomeTrait<N>, const N: usize>() {} }
Generic parameters are also in scope for type bounds and where clauses, for example:
#![allow(unused)] fn main() { trait SomeTrait<'a, T> {} // The <'a, U> for `SomeTrait` refer to the 'a and U parameters of `bounds_scope`. fn bounds_scope<'a, T: SomeTrait<'a, U>, U>() {} fn where_scope<'a, T, U>() where T: SomeTrait<'a, U> {} }
It is an error for items declared inside a function to refer to a generic parameter from their outer scope.
#![allow(unused)] fn main() { fn example<T>() { fn inner(x: T) {} // ERROR: can't use generic parameters from outer function } }
Generic parameter shadowing
It is an error to shadow a generic parameter with the exception that items declared within functions are allowed to shadow generic parameter names from the function.
#![allow(unused)] fn main() { fn example<'a, T, const N: usize>() { // Items within functions are allowed to shadow generic parameter in scope. fn inner_lifetime<'a>() {} // OK fn inner_type<T>() {} // OK fn inner_const<const N: usize>() {} // OK } }
#![allow(unused)] fn main() { trait SomeTrait<'a, T, const N: usize> { fn example_lifetime<'a>() {} // ERROR: 'a is already in use fn example_type<T>() {} // ERROR: T is already in use fn example_const<const N: usize>() {} // ERROR: N is already in use fn example_mixed<const T: usize>() {} // ERROR: T is already in use } }
Lifetime scopes
Lifetime parameters are declared in a GenericParams list and higher-ranked trait bounds.
The 'static
lifetime and placeholder lifetime '_
have a special meaning and cannot be declared as a parameter.
Lifetime generic parameter scopes
Constant and static items and const contexts only ever allow 'static
lifetime references, so no other lifetime may be in scope within them.
Associated consts do allow referring to lifetimes declared in their trait or implementation.
Higher-ranked trait bound scopes
The scope of a lifetime parameter declared as a higher-ranked trait bound depends on the scenario where it is used.
- As a TypeBoundWhereClauseItem the declared lifetimes are in scope in the type and the type bounds.
- As a TraitBound the declared lifetimes are in scope within the bound type path.
- As a BareFunctionType the declared lifetimes are in scope within the function parameters and return type.
#![allow(unused)] fn main() { trait Trait<'a>{} fn where_clause<T>() // 'a is in scope in both the type and the type bounds. where for <'a> &'a T: Trait<'a> {} fn bound<T>() // 'a is in scope within the bound. where T: for <'a> Trait<'a> {} struct Example<'a> { field: &'a u32 } // 'a is in scope in both the parameters and return type. type FnExample = for<'a> fn(x: Example<'a>) -> Example<'a>; }
Impl trait restrictions
Impl trait types can only reference lifetimes declared on a function or implementation.
#![allow(unused)] fn main() { trait Trait1 { type Item; } trait Trait2<'a> {} struct Example; impl Trait1 for Example { type Item = Element; } struct Element; impl<'a> Trait2<'a> for Element {} // The `impl Trait2` here is not allowed to refer to 'b but it is allowed to // refer to 'a. fn foo<'a>() -> impl for<'b> Trait1<Item = impl Trait2<'a>> { // ... Example } }
Loop label scopes
Loop labels may be declared by a loop expression.
The scope of a loop label is from the point it is declared till the end of the loop expression.
The scope does not extend into items, closures, async blocks, const arguments, const contexts, and the iterator expression of the defining for
loop.
#![allow(unused)] fn main() { 'a: for n in 0..3 { if n % 2 == 0 { break 'a; } fn inner() { // Using 'a here would be an error. // break 'a; } } // The label is in scope for the expression of `while` loops. 'a: while break 'a {} // Loop does not run. 'a: while let _ = break 'a {} // Loop does not run. // The label is not in scope in the defining `for` loop: 'a: for outer in 0..5 { // This will break the outer loop, skipping the inner loop and stopping // the outer loop. 'a: for inner in { break 'a; 0..1 } { println!("{}", inner); // This does not run. } println!("{}", outer); // This does not run, either. } }
Loop labels may shadow labels of the same name in outer scopes. References to a label refer to the closest definition.
#![allow(unused)] fn main() { // Loop label shadowing example. 'a: for outer in 0..5 { 'a: for inner in 0..5 { // This terminates the inner loop, but the outer loop continues to run. break 'a; } } }
Prelude scopes
Preludes bring entities into scope of every module. The entities are not members of the module, but are implicitly queried during name resolution. The prelude names may be shadowed by declarations in a module.
The preludes are layered such that one shadows another if they contain entities of the same name. The order that preludes may shadow other preludes is the following where earlier entries may shadow later ones:
macro_rules
scopes
The scope of macro_rules
macros is described in the Macros By Example chapter.
The behavior depends on the use of the macro_use
and macro_export
attributes.
Derive macro helper attributes
Derive macro helper attributes are in scope in the item where their corresponding derive
attribute is specified.
The scope extends from just after the derive
attribute to the end of the item.
Helper attributes shadow other attributes of the same name in scope.
Self
scope
Although Self
is a keyword with special meaning, it interacts with name resolution in a way similar to normal names.
The implicit Self
type in the definition of a struct, enum, union, trait, or implementation is treated similarly to a generic parameter, and is in scope in the same way as a generic type parameter.
The implicit Self
constructor in the value namespace of an implementation is in scope within the body of the implementation (the implementation’s associated items).
#![allow(unused)] fn main() { // Self type within struct definition. struct Recursive { f1: Option<Box<Self>> } // Self type within generic parameters. struct SelfGeneric<T: Into<Self>>(T); // Self value constructor within an implementation. struct ImplExample(); impl ImplExample { fn example() -> Self { // Self type Self() // Self value constructor } } }
Preludes
A prelude is a collection of names that are automatically brought into scope of every module in a crate.
These prelude names are not part of the module itself: they are implicitly
queried during name resolution. For example, even though something like
Box
is in scope in every module, you cannot refer to it as self::Box
because it is not a member of the current module.
There are several different preludes:
Standard library prelude
Each crate has a standard library prelude, which consists of the names from a single standard library module.
The module used depends on the crate’s edition, and on whether the no_std
attribute is applied to the crate:
Edition | no_std not applied | no_std applied |
---|---|---|
2015 | std::prelude::rust_2015 | core::prelude::rust_2015 |
2018 | std::prelude::rust_2018 | core::prelude::rust_2018 |
2021 | std::prelude::rust_2021 | core::prelude::rust_2021 |
Note:
std::prelude::rust_2015
andstd::prelude::rust_2018
have the same contents asstd::prelude::v1
.
core::prelude::rust_2015
andcore::prelude::rust_2018
have the same contents ascore::prelude::v1
.
Extern prelude
External crates imported with extern crate
in the root module or provided
to the compiler (as with the --extern
flag with rustc
) are added to the
extern prelude. If imported with an alias such as extern crate orig_name as new_name
, then the symbol new_name
is instead added to the prelude.
The core
crate is always added to the extern prelude. The std
crate is
added as long as the no_std
attribute is not specified in the crate root.
Edition differences: In the 2015 edition, crates in the extern prelude cannot be referenced via use declarations, so it is generally standard practice to include
extern crate
declarations to bring them into scope.Beginning in the 2018 edition, use declarations can reference crates in the extern prelude, so it is considered unidiomatic to use
extern crate
.
Note: Additional crates that ship with
rustc
, such asalloc
, andtest
, are not automatically included with the--extern
flag when using Cargo. They must be brought into scope with anextern crate
declaration, even in the 2018 edition.#![allow(unused)] fn main() { extern crate alloc; use alloc::rc::Rc; }
Cargo does bring in
proc_macro
to the extern prelude for proc-macro crates only.
The no_std
attribute
By default, the standard library is automatically included in the crate root
module. The std
crate is added to the root, along with an implicit
macro_use
attribute pulling in all macros exported from std
into the
macro_use
prelude. Both core
and std
are added to the extern
prelude.
The no_std
attribute may be applied at the crate level to prevent the
std
crate from being automatically added into scope. It does three things:
- Prevents
std
from being added to the extern prelude. - Affects which module is used to make up the standard library prelude (as described above).
- Injects the
core
crate into the crate root instead ofstd
, and pulls in all macros exported fromcore
in themacro_use
prelude.
Note: Using the core prelude over the standard prelude is useful when either the crate is targeting a platform that does not support the standard library or is purposefully not using the capabilities of the standard library. Those capabilities are mainly dynamic memory allocation (e.g.
Box
andVec
) and file and network capabilities (e.g.std::fs
andstd::io
).
Warning: Using no_std
does not prevent the standard library from being
linked in. It is still valid to put extern crate std;
into the crate and
dependencies can also link it in.
Language prelude
The language prelude includes names of types and attributes that are built-in to the language. The language prelude is always in scope. It includes the following:
- Type namespace
- Boolean type —
bool
- Textual types —
char
andstr
- Integer types —
i8
,i16
,i32
,i64
,i128
,u8
,u16
,u32
,u64
,u128
- Machine-dependent integer types —
usize
andisize
- floating-point types —
f32
andf64
- Boolean type —
- Macro namespace
macro_use
prelude
The macro_use
prelude includes macros from external crates that were
imported by the macro_use
attribute applied to an extern crate
.
Tool prelude
The tool prelude includes tool names for external tools in the type namespace. See the tool attributes section for more details.
The no_implicit_prelude
attribute
The no_implicit_prelude
attribute may be applied at the crate level or
on a module to indicate that it should not automatically bring the standard
library prelude, extern prelude, or tool prelude into scope for that
module or any of its descendants.
This attribute does not affect the language prelude.
Edition differences: In the 2015 edition, the
no_implicit_prelude
attribute does not affect themacro_use
prelude, and all macros exported from the standard library are still included in themacro_use
prelude. Starting in the 2018 edition, it will remove themacro_use
prelude.
Paths
A path is a sequence of one or more path segments separated by ::
tokens.
Paths are used to refer to items, values, types, macros, and attributes.
Two examples of simple paths consisting of only identifier segments:
x;
x::y::z;
Types of paths
Simple Paths
Syntax
SimplePath :
::
? SimplePathSegment (::
SimplePathSegment)*SimplePathSegment :
IDENTIFIER |super
|self
|crate
|$crate
Simple paths are used in visibility markers, attributes, macros, and use
items.
For example:
#![allow(unused)] fn main() { use std::io::{self, Write}; mod m { #[clippy::cyclomatic_complexity = "0"] pub (in super) fn f1() {} } }
Paths in expressions
Syntax
PathInExpression :
::
? PathExprSegment (::
PathExprSegment)*PathExprSegment :
PathIdentSegment (::
GenericArgs)?PathIdentSegment :
IDENTIFIER |super
|self
|Self
|crate
|$crate
GenericArgs :
<
>
|<
( GenericArg,
)* GenericArg,
?>
GenericArg :
Lifetime | Type | GenericArgsConst | GenericArgsBinding | GenericArgsBoundsGenericArgsConst :
BlockExpression
| LiteralExpression
|-
LiteralExpression
| SimplePathSegmentGenericArgsBinding :
IDENTIFIER GenericArgs?=
TypeGenericArgsBounds :
IDENTIFIER GenericArgs?:
TypeParamBounds
Paths in expressions allow for paths with generic arguments to be specified. They are used in various places in expressions and patterns.
The ::
token is required before the opening <
for generic arguments to avoid
ambiguity with the less-than operator. This is colloquially known as “turbofish” syntax.
#![allow(unused)] fn main() { (0..10).collect::<Vec<_>>(); Vec::<u8>::with_capacity(1024); }
The order of generic arguments is restricted to lifetime arguments, then type arguments, then const arguments, then equality constraints.
Const arguments must be surrounded by braces unless they are a literal or a single segment path.
The synthetic type parameters corresponding to impl Trait
types are implicit,
and these cannot be explicitly specified.
Qualified paths
Syntax
QualifiedPathInExpression :
QualifiedPathType (::
PathExprSegment)+QualifiedPathType :
<
Type (as
TypePath)?>
QualifiedPathInType :
QualifiedPathType (::
TypePathSegment)+
Fully qualified paths allow for disambiguating the path for trait implementations and for specifying canonical paths. When used in a type specification, it supports using the type syntax specified below.
#![allow(unused)] fn main() { struct S; impl S { fn f() { println!("S"); } } trait T1 { fn f() { println!("T1 f"); } } impl T1 for S {} trait T2 { fn f() { println!("T2 f"); } } impl T2 for S {} S::f(); // Calls the inherent impl. <S as T1>::f(); // Calls the T1 trait function. <S as T2>::f(); // Calls the T2 trait function. }
Paths in types
Syntax
TypePath :
::
? TypePathSegment (::
TypePathSegment)*TypePathSegment :
PathIdentSegment (::
? (GenericArgs | TypePathFn))?TypePathFn :
(
TypePathFnInputs?)
(->
TypeNoBounds)?
Type paths are used within type definitions, trait bounds, type parameter bounds, and qualified paths.
Although the ::
token is allowed before the generics arguments, it is not required
because there is no ambiguity like there is in PathInExpression.
#![allow(unused)] fn main() { mod ops { pub struct Range<T> {f1: T} pub trait Index<T> {} pub struct Example<'a> {f1: &'a i32} } struct S; impl ops::Index<ops::Range<usize>> for S { /*...*/ } fn i<'a>() -> impl Iterator<Item = ops::Example<'a>> { // ... const EXAMPLE: Vec<ops::Example<'static>> = Vec::new(); EXAMPLE.into_iter() } type G = std::boxed::Box<dyn std::ops::FnOnce(isize) -> isize>; }
Path qualifiers
Paths can be denoted with various leading qualifiers to change the meaning of how it is resolved.
::
Paths starting with ::
are considered to be global paths where the segments of the path
start being resolved from a place which differs based on edition. Each identifier in
the path must resolve to an item.
Edition differences: In the 2015 Edition, identifiers resolve from the “crate root” (
crate::
in the 2018 edition), which contains a variety of different items, including external crates, default crates such asstd
orcore
, and items in the top level of the crate (includinguse
imports).Beginning with the 2018 Edition, paths starting with
::
resolve from crates in the extern prelude. That is, they must be followed by the name of a crate.
#![allow(unused)] fn main() { pub fn foo() { // In the 2018 edition, this accesses `std` via the extern prelude. // In the 2015 edition, this accesses `std` via the crate root. let now = ::std::time::Instant::now(); println!("{:?}", now); } }
// 2015 Edition mod a { pub fn foo() {} } mod b { pub fn foo() { ::a::foo(); // call `a`'s foo function // In Rust 2018, `::a` would be interpreted as the crate `a`. } } fn main() {}
self
self
resolves the path relative to the current module. self
can only be used as the
first segment, without a preceding ::
.
In a method body, a path which consists of a single self
segment resolves to the method’s self parameter.
fn foo() {} fn bar() { self::foo(); } struct S(bool); impl S { fn baz(self) { self.0; } } fn main() {}
Self
Self
, with a capital “S”, is used to refer to the current type being implemented or defined. It may be used in the following situations:
- In a trait definition, it refers to the type implementing the trait.
- In an implementation, it refers to the type being implemented. When implementing a tuple or unit struct, it also refers to the constructor in the value namespace.
- In the definition of a struct, enumeration, or union, it refers to the type being defined. The definition is not allowed to be infinitely recursive (there must be an indirection).
The scope of Self
behaves similarly to a generic parameter; see the Self
scope section for more details.
Self
can only be used as the first segment, without a preceding ::
.
The Self
path cannot include generic arguments (as in Self::<i32>
).
#![allow(unused)] fn main() { trait T { type Item; const C: i32; // `Self` will be whatever type that implements `T`. fn new() -> Self; // `Self::Item` will be the type alias in the implementation. fn f(&self) -> Self::Item; } struct S; impl T for S { type Item = i32; const C: i32 = 9; fn new() -> Self { // `Self` is the type `S`. S } fn f(&self) -> Self::Item { // `Self::Item` is the type `i32`. Self::C // `Self::C` is the constant value `9`. } } // `Self` is in scope within the generics of a trait definition, // to refer to the type being defined. trait Add<Rhs = Self> { type Output; // `Self` can also reference associated items of the // type being implemented. fn add(self, rhs: Rhs) -> Self::Output; } struct NonEmptyList<T> { head: T, // A struct can reference itself (as long as it is not // infinitely recursive). tail: Option<Box<Self>>, } }
super
super
in a path resolves to the parent module. It may only be used in leading
segments of the path, possibly after an initial self
segment.
mod a { pub fn foo() {} } mod b { pub fn foo() { super::a::foo(); // call a's foo function } } fn main() {}
super
may be repeated several times after the first super
or self
to refer to
ancestor modules.
mod a { fn foo() {} mod b { mod c { fn foo() { super::super::foo(); // call a's foo function self::super::super::foo(); // call a's foo function } } } } fn main() {}
crate
crate
resolves the path relative to the current crate. crate
can only be used as the
first segment, without a preceding ::
.
fn foo() {} mod a { fn bar() { crate::foo(); } } fn main() {}
$crate
$crate
is only used within macro transcribers, and can only be used as the first
segment, without a preceding ::
. $crate
will expand to a path to access items from the
top level of the crate where the macro is defined, regardless of which crate the macro is
invoked.
pub fn increment(x: u32) -> u32 { x + 1 } #[macro_export] macro_rules! inc { ($x:expr) => ( $crate::increment($x) ) } fn main() { }
Canonical paths
Items defined in a module or implementation have a canonical path that corresponds to where within its crate it is defined. All other paths to these items are aliases. The canonical path is defined as a path prefix appended by the path segment the item itself defines.
Implementations and use declarations do not have canonical paths, although the items that implementations define do have them. Items defined in block expressions do not have canonical paths. Items defined in a module that does not have a canonical path do not have a canonical path. Associated items defined in an implementation that refers to an item without a canonical path, e.g. as the implementing type, the trait being implemented, a type parameter or bound on a type parameter, do not have canonical paths.
The path prefix for modules is the canonical path to that module. For bare
implementations, it is the canonical path of the item being implemented
surrounded by angle (<>
) brackets. For
trait implementations, it is the canonical path of the item being implemented
followed by as
followed by the canonical path to the trait all surrounded in
angle (<>
) brackets.
The canonical path is only meaningful within a given crate. There is no global namespace across crates; an item’s canonical path merely identifies it within the crate.
// Comments show the canonical path of the item. mod a { // crate::a pub struct Struct; // crate::a::Struct pub trait Trait { // crate::a::Trait fn f(&self); // crate::a::Trait::f } impl Trait for Struct { fn f(&self) {} // <crate::a::Struct as crate::a::Trait>::f } impl Struct { fn g(&self) {} // <crate::a::Struct>::g } } mod without { // crate::without fn canonicals() { // crate::without::canonicals struct OtherStruct; // None trait OtherTrait { // None fn g(&self); // None } impl OtherTrait for OtherStruct { fn g(&self) {} // None } impl OtherTrait for crate::a::Struct { fn g(&self) {} // None } impl crate::a::Trait for OtherStruct { fn f(&self) {} // None } } } fn main() {}
Name resolution
Note: This is a placeholder for future expansion.
Visibility and Privacy
Syntax
Visibility :
pub
|pub
(
crate
)
|pub
(
self
)
|pub
(
super
)
|pub
(
in
SimplePath)
These two terms are often used interchangeably, and what they are attempting to convey is the answer to the question “Can this item be used at this location?”
Rust’s name resolution operates on a global hierarchy of namespaces. Each level in the hierarchy can be thought of as some item. The items are one of those mentioned above, but also include external crates. Declaring or defining a new module can be thought of as inserting a new tree into the hierarchy at the location of the definition.
To control whether interfaces can be used across modules, Rust checks each use of an item to see whether it should be allowed or not. This is where privacy warnings are generated, or otherwise “you used a private item of another module and weren’t allowed to.”
By default, everything is private, with two exceptions: Associated
items in a pub
Trait are public by default; Enum variants
in a pub
enum are also public by default. When an item is declared as pub
,
it can be thought of as being accessible to the outside world. For example:
fn main() {} // Declare a private struct struct Foo; // Declare a public struct with a private field pub struct Bar { field: i32, } // Declare a public enum with two public variants pub enum State { PubliclyAccessibleState, PubliclyAccessibleState2, }
With the notion of an item being either public or private, Rust allows item accesses in two cases:
- If an item is public, then it can be accessed externally from some module
m
if you can access all the item’s ancestor modules fromm
. You can also potentially be able to name the item through re-exports. See below. - If an item is private, it may be accessed by the current module and its descendants.
These two cases are surprisingly powerful for creating module hierarchies exposing public APIs while hiding internal implementation details. To help explain, here’s a few use cases and what they would entail:
-
A library developer needs to expose functionality to crates which link against their library. As a consequence of the first case, this means that anything which is usable externally must be
pub
from the root down to the destination item. Any private item in the chain will disallow external accesses. -
A crate needs a global available “helper module” to itself, but it doesn’t want to expose the helper module as a public API. To accomplish this, the root of the crate’s hierarchy would have a private module which then internally has a “public API”. Because the entire crate is a descendant of the root, then the entire local crate can access this private module through the second case.
-
When writing unit tests for a module, it’s often a common idiom to have an immediate child of the module to-be-tested named
mod test
. This module could access any items of the parent module through the second case, meaning that internal implementation details could also be seamlessly tested from the child module.
In the second case, it mentions that a private item “can be accessed” by the current module and its descendants, but the exact meaning of accessing an item depends on what the item is. Accessing a module, for example, would mean looking inside of it (to import more items). On the other hand, accessing a function would mean that it is invoked. Additionally, path expressions and import statements are considered to access an item in the sense that the import/expression is only valid if the destination is in the current visibility scope.
Here’s an example of a program which exemplifies the three cases outlined above:
// This module is private, meaning that no external crate can access this // module. Because it is private at the root of this current crate, however, any // module in the crate may access any publicly visible item in this module. mod crate_helper_module { // This function can be used by anything in the current crate pub fn crate_helper() {} // This function *cannot* be used by anything else in the crate. It is not // publicly visible outside of the `crate_helper_module`, so only this // current module and its descendants may access it. fn implementation_detail() {} } // This function is "public to the root" meaning that it's available to external // crates linking against this one. pub fn public_api() {} // Similarly to 'public_api', this module is public so external crates may look // inside of it. pub mod submodule { use crate::crate_helper_module; pub fn my_method() { // Any item in the local crate may invoke the helper module's public // interface through a combination of the two rules above. crate_helper_module::crate_helper(); } // This function is hidden to any module which is not a descendant of // `submodule` fn my_implementation() {} #[cfg(test)] mod test { #[test] fn test_my_implementation() { // Because this module is a descendant of `submodule`, it's allowed // to access private items inside of `submodule` without a privacy // violation. super::my_implementation(); } } } fn main() {}
For a Rust program to pass the privacy checking pass, all paths must be valid accesses given the two rules above. This includes all use statements, expressions, types, etc.
pub(in path)
, pub(crate)
, pub(super)
, and pub(self)
In addition to public and private, Rust allows users to declare an item as
visible only within a given scope. The rules for pub
restrictions are as
follows:
pub(in path)
makes an item visible within the providedpath
.path
must be a simple path which resolves to an ancestor module of the item whose visibility is being declared. Each identifier inpath
must refer directly to a module (not to a name introduced by ause
statement).pub(crate)
makes an item visible within the current crate.pub(super)
makes an item visible to the parent module. This is equivalent topub(in super)
.pub(self)
makes an item visible to the current module. This is equivalent topub(in self)
or not usingpub
at all.
Edition differences: Starting with the 2018 edition, paths for
pub(in path)
must start withcrate
,self
, orsuper
. The 2015 edition may also use paths starting with::
or modules from the crate root.
Here’s an example:
pub mod outer_mod { pub mod inner_mod { // This function is visible within `outer_mod` pub(in crate::outer_mod) fn outer_mod_visible_fn() {} // Same as above, this is only valid in the 2015 edition. pub(in outer_mod) fn outer_mod_visible_fn_2015() {} // This function is visible to the entire crate pub(crate) fn crate_visible_fn() {} // This function is visible within `outer_mod` pub(super) fn super_mod_visible_fn() { // This function is visible since we're in the same `mod` inner_mod_visible_fn(); } // This function is visible only within `inner_mod`, // which is the same as leaving it private. pub(self) fn inner_mod_visible_fn() {} } pub fn foo() { inner_mod::outer_mod_visible_fn(); inner_mod::crate_visible_fn(); inner_mod::super_mod_visible_fn(); // This function is no longer visible since we're outside of `inner_mod` // Error! `inner_mod_visible_fn` is private //inner_mod::inner_mod_visible_fn(); } } fn bar() { // This function is still visible since we're in the same crate outer_mod::inner_mod::crate_visible_fn(); // This function is no longer visible since we're outside of `outer_mod` // Error! `super_mod_visible_fn` is private //outer_mod::inner_mod::super_mod_visible_fn(); // This function is no longer visible since we're outside of `outer_mod` // Error! `outer_mod_visible_fn` is private //outer_mod::inner_mod::outer_mod_visible_fn(); outer_mod::foo(); } fn main() { bar() }
Note: This syntax only adds another restriction to the visibility of an item. It does not guarantee that the item is visible within all parts of the specified scope. To access an item, all of its parent items up to the current scope must still be visible as well.
Re-exporting and Visibility
Rust allows publicly re-exporting items through a pub use
directive. Because
this is a public directive, this allows the item to be used in the current
module through the rules above. It essentially allows public access into the
re-exported item. For example, this program is valid:
pub use self::implementation::api; mod implementation { pub mod api { pub fn f() {} } } fn main() {}
This means that any external crate referencing implementation::api::f
would
receive a privacy violation, while the path api::f
would be allowed.
When re-exporting a private item, it can be thought of as allowing the “privacy chain” being short-circuited through the reexport instead of passing through the namespace hierarchy as it normally would.
Memory model
Rust does not yet have a defined memory model. Various academics and industry professionals are working on various proposals, but for now, this is an under-defined place in the language.
Memory allocation and lifetime
The items of a program are those functions, modules, and types that have their value calculated at compile-time and stored uniquely in the memory image of the rust process. Items are neither dynamically allocated nor freed.
The heap is a general term that describes boxes. The lifetime of an allocation in the heap depends on the lifetime of the box values pointing to it. Since box values may themselves be passed in and out of frames, or stored in the heap, heap allocations may outlive the frame they are allocated within. An allocation in the heap is guaranteed to reside at a single location in the heap for the whole lifetime of the allocation - it will never be relocated as a result of moving a box value.
Variables
A variable is a component of a stack frame, either a named function parameter, an anonymous temporary, or a named local variable.
A local variable (or stack-local allocation) holds a value directly, allocated within the stack’s memory. The value is a part of the stack frame.
Local variables are immutable unless declared otherwise. For example:
let mut x = ...
.
Function parameters are immutable unless declared with mut
. The mut
keyword
applies only to the following parameter. For example: |mut x, y|
and
fn f(mut x: Box<i32>, y: Box<i32>)
declare one mutable variable x
and one
immutable variable y
.
Local variables are not initialized when allocated. Instead, the entire frame worth of local variables are allocated, on frame-entry, in an uninitialized state. Subsequent statements within a function may or may not initialize the local variables. Local variables can be used only after they have been initialized through all reachable control flow paths.
In this next example, init_after_if
is initialized after the if
expression
while uninit_after_if
is not because it is not initialized in the else
case.
#![allow(unused)] fn main() { fn random_bool() -> bool { true } fn initialization_example() { let init_after_if: (); let uninit_after_if: (); if random_bool() { init_after_if = (); uninit_after_if = (); } else { init_after_if = (); } init_after_if; // ok // uninit_after_if; // err: use of possibly uninitialized `uninit_after_if` } }
Linkage
Note: This section is described more in terms of the compiler than of the language.
The compiler supports various methods to link crates together both statically and dynamically. This section will explore the various methods to link crates together, and more information about native libraries can be found in the FFI section of the book.
In one session of compilation, the compiler can generate multiple artifacts
through the usage of either command line flags or the crate_type
attribute.
If one or more command line flags are specified, all crate_type
attributes will
be ignored in favor of only building the artifacts specified by command line.
-
--crate-type=bin
,#![crate_type = "bin"]
- A runnable executable will be produced. This requires that there is amain
function in the crate which will be run when the program begins executing. This will link in all Rust and native dependencies, producing a single distributable binary. This is the default crate type. -
--crate-type=lib
,#![crate_type = "lib"]
- A Rust library will be produced. This is an ambiguous concept as to what exactly is produced because a library can manifest itself in several forms. The purpose of this genericlib
option is to generate the “compiler recommended” style of library. The output library will always be usable by rustc, but the actual type of library may change from time-to-time. The remaining output types are all different flavors of libraries, and thelib
type can be seen as an alias for one of them (but the actual one is compiler-defined). -
--crate-type=dylib
,#![crate_type = "dylib"]
- A dynamic Rust library will be produced. This is different from thelib
output type in that this forces dynamic library generation. The resulting dynamic library can be used as a dependency for other libraries and/or executables. This output type will create*.so
files on Linux,*.dylib
files on macOS, and*.dll
files on Windows. -
--crate-type=staticlib
,#![crate_type = "staticlib"]
- A static system library will be produced. This is different from other library outputs in that the compiler will never attempt to link tostaticlib
outputs. The purpose of this output type is to create a static library containing all of the local crate’s code along with all upstream dependencies. This output type will create*.a
files on Linux, macOS and Windows (MinGW), and*.lib
files on Windows (MSVC). This format is recommended for use in situations such as linking Rust code into an existing non-Rust application because it will not have dynamic dependencies on other Rust code.Note that any dynamic dependencies that the static library may have (such as dependencies on system libraries, or dependencies on Rust libraries that are compiled as dynamic libraries) will have to be specified manually when linking that static library from somewhere. The
--print=native-static-libs
flag may help with this.Note that, because the resulting static library contains the code of all the dependencies, including the standard library, and also exports all public symbols of them, linking the static library into an executable or shared library may need special care. In case of a shared library the list of exported symbols will have to be limited via e.g. a linker or symbol version script, exported symbols list (macOS), or module definition file (Windows). Additionally, unused sections can be removed to remove all code of dependencies that is not actually used (e.g.
--gc-sections
or-dead_strip
for macOS). -
--crate-type=cdylib
,#![crate_type = "cdylib"]
- A dynamic system library will be produced. This is used when compiling a dynamic library to be loaded from another language. This output type will create*.so
files on Linux,*.dylib
files on macOS, and*.dll
files on Windows. -
--crate-type=rlib
,#![crate_type = "rlib"]
- A “Rust library” file will be produced. This is used as an intermediate artifact and can be thought of as a “static Rust library”. Theserlib
files, unlikestaticlib
files, are interpreted by the compiler in future linkage. This essentially means thatrustc
will look for metadata inrlib
files like it looks for metadata in dynamic libraries. This form of output is used to produce statically linked executables as well asstaticlib
outputs. -
--crate-type=proc-macro
,#![crate_type = "proc-macro"]
- The output produced is not specified, but if a-L
path is provided to it then the compiler will recognize the output artifacts as a macro and it can be loaded for a program. Crates compiled with this crate type must only export procedural macros. The compiler will automatically set theproc_macro
configuration option. The crates are always compiled with the same target that the compiler itself was built with. For example, if you are executing the compiler from Linux with anx86_64
CPU, the target will bex86_64-unknown-linux-gnu
even if the crate is a dependency of another crate being built for a different target.
Note that these outputs are stackable in the sense that if multiple are
specified, then the compiler will produce each form of output without
having to recompile. However, this only applies for outputs specified by the
same method. If only crate_type
attributes are specified, then they will all
be built, but if one or more --crate-type
command line flags are specified,
then only those outputs will be built.
With all these different kinds of outputs, if crate A depends on crate B, then
the compiler could find B in various different forms throughout the system. The
only forms looked for by the compiler, however, are the rlib
format and the
dynamic library format. With these two options for a dependent library, the
compiler must at some point make a choice between these two formats. With this
in mind, the compiler follows these rules when determining what format of
dependencies will be used:
-
If a static library is being produced, all upstream dependencies are required to be available in
rlib
formats. This requirement stems from the reason that a dynamic library cannot be converted into a static format.Note that it is impossible to link in native dynamic dependencies to a static library, and in this case warnings will be printed about all unlinked native dynamic dependencies.
-
If an
rlib
file is being produced, then there are no restrictions on what format the upstream dependencies are available in. It is simply required that all upstream dependencies be available for reading metadata from.The reason for this is that
rlib
files do not contain any of their upstream dependencies. It wouldn’t be very efficient for allrlib
files to contain a copy oflibstd.rlib
! -
If an executable is being produced and the
-C prefer-dynamic
flag is not specified, then dependencies are first attempted to be found in therlib
format. If some dependencies are not available in an rlib format, then dynamic linking is attempted (see below). -
If a dynamic library or an executable that is being dynamically linked is being produced, then the compiler will attempt to reconcile the available dependencies in either the rlib or dylib format to create a final product.
A major goal of the compiler is to ensure that a library never appears more than once in any artifact. For example, if dynamic libraries B and C were each statically linked to library A, then a crate could not link to B and C together because there would be two copies of A. The compiler allows mixing the rlib and dylib formats, but this restriction must be satisfied.
The compiler currently implements no method of hinting what format a library should be linked with. When dynamically linking, the compiler will attempt to maximize dynamic dependencies while still allowing some dependencies to be linked in via an rlib.
For most situations, having all libraries available as a dylib is recommended if dynamically linking. For other situations, the compiler will emit a warning if it is unable to determine which formats to link each library with.
In general, --crate-type=bin
or --crate-type=lib
should be sufficient for
all compilation needs, and the other options are just available if more
fine-grained control is desired over the output format of a crate.
Static and dynamic C runtimes
The standard library in general strives to support both statically linked and
dynamically linked C runtimes for targets as appropriate. For example the
x86_64-pc-windows-msvc
and x86_64-unknown-linux-musl
targets typically come
with both runtimes and the user selects which one they’d like. All targets in
the compiler have a default mode of linking to the C runtime. Typically targets
are linked dynamically by default, but there are exceptions which are static by
default such as:
arm-unknown-linux-musleabi
arm-unknown-linux-musleabihf
armv7-unknown-linux-musleabihf
i686-unknown-linux-musl
x86_64-unknown-linux-musl
The linkage of the C runtime is configured to respect the crt-static
target
feature. These target features are typically configured from the command line
via flags to the compiler itself. For example to enable a static runtime you
would execute:
rustc -C target-feature=+crt-static foo.rs
whereas to link dynamically to the C runtime you would execute:
rustc -C target-feature=-crt-static foo.rs
Targets which do not support switching between linkage of the C runtime will ignore this flag. It’s recommended to inspect the resulting binary to ensure that it’s linked as you would expect after the compiler succeeds.
Crates may also learn about how the C runtime is being linked. Code on MSVC, for
example, needs to be compiled differently (e.g. with /MT
or /MD
) depending
on the runtime being linked. This is exported currently through the
cfg
attribute target_feature
option:
#![allow(unused)] fn main() { #[cfg(target_feature = "crt-static")] fn foo() { println!("the C runtime should be statically linked"); } #[cfg(not(target_feature = "crt-static"))] fn foo() { println!("the C runtime should be dynamically linked"); } }
Also note that Cargo build scripts can learn about this feature through environment variables. In a build script you can detect the linkage via:
use std::env; fn main() { let linkage = env::var("CARGO_CFG_TARGET_FEATURE").unwrap_or(String::new()); if linkage.contains("crt-static") { println!("the C runtime will be statically linked"); } else { println!("the C runtime will be dynamically linked"); } }
To use this feature locally, you typically will use the RUSTFLAGS
environment
variable to specify flags to the compiler through Cargo. For example to compile
a statically linked binary on MSVC you would execute:
RUSTFLAGS='-C target-feature=+crt-static' cargo build --target x86_64-pc-windows-msvc
Inline assembly
Support for inline assembly is provided via the asm!
and global_asm!
macros.
It can be used to embed handwritten assembly in the assembly output generated by the compiler.
Support for inline assembly is stable on the following architectures:
- x86 and x86-64
- ARM
- AArch64
- RISC-V
- LoongArch
The compiler will emit an error if asm!
is used on an unsupported target.
Example
#![allow(unused)] fn main() { #[cfg(target_arch = "x86_64")] { use std::arch::asm; // Multiply x by 6 using shifts and adds let mut x: u64 = 4; unsafe { asm!( "mov {tmp}, {x}", "shl {tmp}, 1", "shl {x}, 2", "add {x}, {tmp}", x = inout(reg) x, tmp = out(reg) _, ); } assert_eq!(x, 4 * 6); } }
Syntax
The following ABNF specifies the general syntax:
format_string := STRING_LITERAL / RAW_STRING_LITERAL
dir_spec := "in" / "out" / "lateout" / "inout" / "inlateout"
reg_spec := <register class> / "\"" <explicit register> "\""
operand_expr := expr / "_" / expr "=>" expr / expr "=>" "_"
reg_operand := [ident "="] dir_spec "(" reg_spec ")" operand_expr / sym <path> / const <expr>
clobber_abi := "clobber_abi(" <abi> *("," <abi>) [","] ")"
option := "pure" / "nomem" / "readonly" / "preserves_flags" / "noreturn" / "nostack" / "att_syntax" / "raw"
options := "options(" option *("," option) [","] ")"
operand := reg_operand / clobber_abi / options
asm := "asm!(" format_string *("," format_string) *("," operand) [","] ")"
global_asm := "global_asm!(" format_string *("," format_string) *("," operand) [","] ")"
Scope
Inline assembly can be used in one of two ways.
With the asm!
macro, the assembly code is emitted in a function scope and integrated into the compiler-generated assembly code of a function.
This assembly code must obey strict rules to avoid undefined behavior.
Note that in some cases the compiler may choose to emit the assembly code as a separate function and generate a call to it.
With the global_asm!
macro, the assembly code is emitted in a global scope, outside a function.
This can be used to hand-write entire functions using assembly code, and generally provides much more freedom to use arbitrary registers and assembler directives.
Template string arguments
The assembler template uses the same syntax as format strings (i.e. placeholders are specified by curly braces).
The corresponding arguments are accessed in order, by index, or by name.
However, implicit named arguments (introduced by RFC #2795) are not supported.
An asm!
invocation may have one or more template string arguments; an asm!
with multiple template string arguments is treated as if all the strings were concatenated with a \n
between them.
The expected usage is for each template string argument to correspond to a line of assembly code.
All template string arguments must appear before any other arguments.
As with format strings, positional arguments must appear before named arguments and explicit register operands.
Explicit register operands cannot be used by placeholders in the template string.
All other named and positional operands must appear at least once in the template string, otherwise a compiler error is generated.
The exact assembly code syntax is target-specific and opaque to the compiler except for the way operands are substituted into the template string to form the code passed to the assembler.
Currently, all supported targets follow the assembly code syntax used by LLVM’s internal assembler which usually corresponds to that of the GNU assembler (GAS).
On x86, the .intel_syntax noprefix
mode of GAS is used by default.
On ARM, the .syntax unified
mode is used.
These targets impose an additional restriction on the assembly code: any assembler state (e.g. the current section which can be changed with .section
) must be restored to its original value at the end of the asm string.
Assembly code that does not conform to the GAS syntax will result in assembler-specific behavior.
Further constraints on the directives used by inline assembly are indicated by Directives Support.
Operand type
Several types of operands are supported:
in(<reg>) <expr>
<reg>
can refer to a register class or an explicit register. The allocated register name is substituted into the asm template string.- The allocated register will contain the value of
<expr>
at the start of the asm code. - The allocated register must contain the same value at the end of the asm code (except if a
lateout
is allocated to the same register).
out(<reg>) <expr>
<reg>
can refer to a register class or an explicit register. The allocated register name is substituted into the asm template string.- The allocated register will contain an undefined value at the start of the asm code.
<expr>
must be a (possibly uninitialized) place expression, to which the contents of the allocated register are written at the end of the asm code.- An underscore (
_
) may be specified instead of an expression, which will cause the contents of the register to be discarded at the end of the asm code (effectively acting as a clobber).
lateout(<reg>) <expr>
- Identical to
out
except that the register allocator can reuse a register allocated to anin
. - You should only write to the register after all inputs are read, otherwise you may clobber an input.
- Identical to
inout(<reg>) <expr>
<reg>
can refer to a register class or an explicit register. The allocated register name is substituted into the asm template string.- The allocated register will contain the value of
<expr>
at the start of the asm code. <expr>
must be a mutable initialized place expression, to which the contents of the allocated register are written at the end of the asm code.
inout(<reg>) <in expr> => <out expr>
- Same as
inout
except that the initial value of the register is taken from the value of<in expr>
. <out expr>
must be a (possibly uninitialized) place expression, to which the contents of the allocated register are written at the end of the asm code.- An underscore (
_
) may be specified instead of an expression for<out expr>
, which will cause the contents of the register to be discarded at the end of the asm code (effectively acting as a clobber). <in expr>
and<out expr>
may have different types.
- Same as
inlateout(<reg>) <expr>
/inlateout(<reg>) <in expr> => <out expr>
- Identical to
inout
except that the register allocator can reuse a register allocated to anin
(this can happen if the compiler knows thein
has the same initial value as theinlateout
). - You should only write to the register after all inputs are read, otherwise you may clobber an input.
- Identical to
sym <path>
<path>
must refer to afn
orstatic
.- A mangled symbol name referring to the item is substituted into the asm template string.
- The substituted string does not include any modifiers (e.g. GOT, PLT, relocations, etc).
<path>
is allowed to point to a#[thread_local]
static, in which case the asm code can combine the symbol with relocations (e.g.@plt
,@TPOFF
) to read from thread-local data.
const <expr>
<expr>
must be an integer constant expression. This expression follows the same rules as inlineconst
blocks.- The type of the expression may be any integer type, but defaults to
i32
just like integer literals. - The value of the expression is formatted as a string and substituted directly into the asm template string.
Operand expressions are evaluated from left to right, just like function call arguments.
After the asm!
has executed, outputs are written to in left to right order.
This is significant if two outputs point to the same place: that place will contain the value of the rightmost output.
Since global_asm!
exists outside a function, it can only use sym
and const
operands.
Register operands
Input and output operands can be specified either as an explicit register or as a register class from which the register allocator can select a register.
Explicit registers are specified as string literals (e.g. "eax"
) while register classes are specified as identifiers (e.g. reg
).
Note that explicit registers treat register aliases (e.g. r14
vs lr
on ARM) and smaller views of a register (e.g. eax
vs rax
) as equivalent to the base register.
It is a compile-time error to use the same explicit register for two input operands or two output operands.
Additionally, it is also a compile-time error to use overlapping registers (e.g. ARM VFP) in input operands or in output operands.
Only the following types are allowed as operands for inline assembly:
- Integers (signed and unsigned)
- Floating-point numbers
- Pointers (thin only)
- Function pointers
- SIMD vectors (structs defined with
#[repr(simd)]
and which implementCopy
). This includes architecture-specific vector types defined instd::arch
such as__m128
(x86) orint8x16_t
(ARM).
Here is the list of currently supported register classes:
Architecture | Register class | Registers | LLVM constraint code |
---|---|---|---|
x86 | reg | ax , bx , cx , dx , si , di , bp , r[8-15] (x86-64 only) | r |
x86 | reg_abcd | ax , bx , cx , dx | Q |
x86-32 | reg_byte | al , bl , cl , dl , ah , bh , ch , dh | q |
x86-64 | reg_byte * | al , bl , cl , dl , sil , dil , bpl , r[8-15]b | q |
x86 | xmm_reg | xmm[0-7] (x86) xmm[0-15] (x86-64) | x |
x86 | ymm_reg | ymm[0-7] (x86) ymm[0-15] (x86-64) | x |
x86 | zmm_reg | zmm[0-7] (x86) zmm[0-31] (x86-64) | v |
x86 | kreg | k[1-7] | Yk |
x86 | kreg0 | k0 | Only clobbers |
x86 | x87_reg | st([0-7]) | Only clobbers |
x86 | mmx_reg | mm[0-7] | Only clobbers |
x86-64 | tmm_reg | tmm[0-7] | Only clobbers |
AArch64 | reg | x[0-30] | r |
AArch64 | vreg | v[0-31] | w |
AArch64 | vreg_low16 | v[0-15] | x |
AArch64 | preg | p[0-15] , ffr | Only clobbers |
ARM (ARM/Thumb2) | reg | r[0-12] , r14 | r |
ARM (Thumb1) | reg | r[0-7] | r |
ARM | sreg | s[0-31] | t |
ARM | sreg_low16 | s[0-15] | x |
ARM | dreg | d[0-31] | w |
ARM | dreg_low16 | d[0-15] | t |
ARM | dreg_low8 | d[0-8] | x |
ARM | qreg | q[0-15] | w |
ARM | qreg_low8 | q[0-7] | t |
ARM | qreg_low4 | q[0-3] | x |
RISC-V | reg | x1 , x[5-7] , x[9-15] , x[16-31] (non-RV32E) | r |
RISC-V | freg | f[0-31] | f |
RISC-V | vreg | v[0-31] | Only clobbers |
LoongArch | reg | $r1 , $r[4-20] , $r[23,30] | r |
LoongArch | freg | $f[0-31] | f |
Notes:
- On x86 we treat
reg_byte
differently fromreg
because the compiler can allocateal
andah
separately whereasreg
reserves the whole register.- On x86-64 the high byte registers (e.g.
ah
) are not available in thereg_byte
register class.- Some register classes are marked as “Only clobbers” which means that registers in these classes cannot be used for inputs or outputs, only clobbers of the form
out(<explicit register>) _
orlateout(<explicit register>) _
.
Each register class has constraints on which value types they can be used with.
This is necessary because the way a value is loaded into a register depends on its type.
For example, on big-endian systems, loading a i32x4
and a i8x16
into a SIMD register may result in different register contents even if the byte-wise memory representation of both values is identical.
The availability of supported types for a particular register class may depend on what target features are currently enabled.
Architecture | Register class | Target feature | Allowed types |
---|---|---|---|
x86-32 | reg | None | i16 , i32 , f32 |
x86-64 | reg | None | i16 , i32 , f32 , i64 , f64 |
x86 | reg_byte | None | i8 |
x86 | xmm_reg | sse | i32 , f32 , i64 , f64 , i8x16 , i16x8 , i32x4 , i64x2 , f32x4 , f64x2 |
x86 | ymm_reg | avx | i32 , f32 , i64 , f64 , i8x16 , i16x8 , i32x4 , i64x2 , f32x4 , f64x2 i8x32 , i16x16 , i32x8 , i64x4 , f32x8 , f64x4 |
x86 | zmm_reg | avx512f | i32 , f32 , i64 , f64 , i8x16 , i16x8 , i32x4 , i64x2 , f32x4 , f64x2 i8x32 , i16x16 , i32x8 , i64x4 , f32x8 , f64x4 i8x64 , i16x32 , i32x16 , i64x8 , f32x16 , f64x8 |
x86 | kreg | avx512f | i8 , i16 |
x86 | kreg | avx512bw | i32 , i64 |
x86 | mmx_reg | N/A | Only clobbers |
x86 | x87_reg | N/A | Only clobbers |
x86 | tmm_reg | N/A | Only clobbers |
AArch64 | reg | None | i8 , i16 , i32 , f32 , i64 , f64 |
AArch64 | vreg | neon | i8 , i16 , i32 , f32 , i64 , f64 , i8x8 , i16x4 , i32x2 , i64x1 , f32x2 , f64x1 , i8x16 , i16x8 , i32x4 , i64x2 , f32x4 , f64x2 |
AArch64 | preg | N/A | Only clobbers |
ARM | reg | None | i8 , i16 , i32 , f32 |
ARM | sreg | vfp2 | i32 , f32 |
ARM | dreg | vfp2 | i64 , f64 , i8x8 , i16x4 , i32x2 , i64x1 , f32x2 |
ARM | qreg | neon | i8x16 , i16x8 , i32x4 , i64x2 , f32x4 |
RISC-V32 | reg | None | i8 , i16 , i32 , f32 |
RISC-V64 | reg | None | i8 , i16 , i32 , f32 , i64 , f64 |
RISC-V | freg | f | f32 |
RISC-V | freg | d | f64 |
RISC-V | vreg | N/A | Only clobbers |
LoongArch64 | reg | None | i8 , i16 , i32 , i64 , f32 , f64 |
LoongArch64 | freg | None | f32 , f64 |
Note: For the purposes of the above table pointers, function pointers and
isize
/usize
are treated as the equivalent integer type (i16
/i32
/i64
depending on the target).
If a value is of a smaller size than the register it is allocated in then the upper bits of that register will have an undefined value for inputs and will be ignored for outputs.
The only exception is the freg
register class on RISC-V where f32
values are NaN-boxed in a f64
as required by the RISC-V architecture.
When separate input and output expressions are specified for an inout
operand, both expressions must have the same type.
The only exception is if both operands are pointers or integers, in which case they are only required to have the same size.
This restriction exists because the register allocators in LLVM and GCC sometimes cannot handle tied operands with different types.
Register names
Some registers have multiple names. These are all treated by the compiler as identical to the base register name. Here is the list of all supported register aliases:
Architecture | Base register | Aliases |
---|---|---|
x86 | ax | eax , rax |
x86 | bx | ebx , rbx |
x86 | cx | ecx , rcx |
x86 | dx | edx , rdx |
x86 | si | esi , rsi |
x86 | di | edi , rdi |
x86 | bp | bpl , ebp , rbp |
x86 | sp | spl , esp , rsp |
x86 | ip | eip , rip |
x86 | st(0) | st |
x86 | r[8-15] | r[8-15]b , r[8-15]w , r[8-15]d |
x86 | xmm[0-31] | ymm[0-31] , zmm[0-31] |
AArch64 | x[0-30] | w[0-30] |
AArch64 | x29 | fp |
AArch64 | x30 | lr |
AArch64 | sp | wsp |
AArch64 | xzr | wzr |
AArch64 | v[0-31] | b[0-31] , h[0-31] , s[0-31] , d[0-31] , q[0-31] |
ARM | r[0-3] | a[1-4] |
ARM | r[4-9] | v[1-6] |
ARM | r9 | rfp |
ARM | r10 | sl |
ARM | r11 | fp |
ARM | r12 | ip |
ARM | r13 | sp |
ARM | r14 | lr |
ARM | r15 | pc |
RISC-V | x0 | zero |
RISC-V | x1 | ra |
RISC-V | x2 | sp |
RISC-V | x3 | gp |
RISC-V | x4 | tp |
RISC-V | x[5-7] | t[0-2] |
RISC-V | x8 | fp , s0 |
RISC-V | x9 | s1 |
RISC-V | x[10-17] | a[0-7] |
RISC-V | x[18-27] | s[2-11] |
RISC-V | x[28-31] | t[3-6] |
RISC-V | f[0-7] | ft[0-7] |
RISC-V | f[8-9] | fs[0-1] |
RISC-V | f[10-17] | fa[0-7] |
RISC-V | f[18-27] | fs[2-11] |
RISC-V | f[28-31] | ft[8-11] |
LoongArch | $r0 | $zero |
LoongArch | $r1 | $ra |
LoongArch | $r2 | $tp |
LoongArch | $r3 | $sp |
LoongArch | $r[4-11] | $a[0-7] |
LoongArch | $r[12-20] | $t[0-8] |
LoongArch | $r21 | |
LoongArch | $r22 | $fp , $s9 |
LoongArch | $r[23-31] | $s[0-8] |
LoongArch | $f[0-7] | $fa[0-7] |
LoongArch | $f[8-23] | $ft[0-15] |
LoongArch | $f[24-31] | $fs[0-7] |
Some registers cannot be used for input or output operands:
Architecture | Unsupported register | Reason |
---|---|---|
All | sp | The stack pointer must be restored to its original value at the end of an asm code block. |
All | bp (x86), x29 (AArch64), x8 (RISC-V), $fp (LoongArch) | The frame pointer cannot be used as an input or output. |
ARM | r7 or r11 | On ARM the frame pointer can be either r7 or r11 depending on the target. The frame pointer cannot be used as an input or output. |
All | si (x86-32), bx (x86-64), r6 (ARM), x19 (AArch64), x9 (RISC-V), $s8 (LoongArch) | This is used internally by LLVM as a “base pointer” for functions with complex stack frames. |
x86 | ip | This is the program counter, not a real register. |
AArch64 | xzr | This is a constant zero register which can’t be modified. |
AArch64 | x18 | This is an OS-reserved register on some AArch64 targets. |
ARM | pc | This is the program counter, not a real register. |
ARM | r9 | This is an OS-reserved register on some ARM targets. |
RISC-V | x0 | This is a constant zero register which can’t be modified. |
RISC-V | gp , tp | These registers are reserved and cannot be used as inputs or outputs. |
LoongArch | $r0 or $zero | This is a constant zero register which can’t be modified. |
LoongArch | $r2 or $tp | This is reserved for TLS. |
LoongArch | $r21 | This is reserved by the ABI. |
The frame pointer and base pointer registers are reserved for internal use by LLVM. While asm!
statements cannot explicitly specify the use of reserved registers, in some cases LLVM will allocate one of these reserved registers for reg
operands. Assembly code making use of reserved registers should be careful since reg
operands may use the same registers.
Template modifiers
The placeholders can be augmented by modifiers which are specified after the :
in the curly braces.
These modifiers do not affect register allocation, but change the way operands are formatted when inserted into the template string.
Only one modifier is allowed per template placeholder.
The supported modifiers are a subset of LLVM’s (and GCC’s) asm template argument modifiers, but do not use the same letter codes.
Architecture | Register class | Modifier | Example output | LLVM modifier |
---|---|---|---|---|
x86-32 | reg | None | eax | k |
x86-64 | reg | None | rax | q |
x86-32 | reg_abcd | l | al | b |
x86-64 | reg | l | al | b |
x86 | reg_abcd | h | ah | h |
x86 | reg | x | ax | w |
x86 | reg | e | eax | k |
x86-64 | reg | r | rax | q |
x86 | reg_byte | None | al / ah | None |
x86 | xmm_reg | None | xmm0 | x |
x86 | ymm_reg | None | ymm0 | t |
x86 | zmm_reg | None | zmm0 | g |
x86 | *mm_reg | x | xmm0 | x |
x86 | *mm_reg | y | ymm0 | t |
x86 | *mm_reg | z | zmm0 | g |
x86 | kreg | None | k1 | None |
AArch64 | reg | None | x0 | x |
AArch64 | reg | w | w0 | w |
AArch64 | reg | x | x0 | x |
AArch64 | vreg | None | v0 | None |
AArch64 | vreg | v | v0 | None |
AArch64 | vreg | b | b0 | b |
AArch64 | vreg | h | h0 | h |
AArch64 | vreg | s | s0 | s |
AArch64 | vreg | d | d0 | d |
AArch64 | vreg | q | q0 | q |
ARM | reg | None | r0 | None |
ARM | sreg | None | s0 | None |
ARM | dreg | None | d0 | P |
ARM | qreg | None | q0 | q |
ARM | qreg | e / f | d0 / d1 | e / f |
RISC-V | reg | None | x1 | None |
RISC-V | freg | None | f0 | None |
LoongArch | reg | None | $r1 | None |
LoongArch | freg | None | $f0 | None |
Notes:
- on ARM
e
/f
: this prints the low or high doubleword register name of a NEON quad (128-bit) register.- on x86: our behavior for
reg
with no modifiers differs from what GCC does. GCC will infer the modifier based on the operand value type, while we default to the full register size.- on x86
xmm_reg
: thex
,t
andg
LLVM modifiers are not yet implemented in LLVM (they are supported by GCC only), but this should be a simple change.
As stated in the previous section, passing an input value smaller than the register width will result in the upper bits of the register containing undefined values.
This is not a problem if the inline asm only accesses the lower bits of the register, which can be done by using a template modifier to use a subregister name in the asm code (e.g. ax
instead of rax
).
Since this an easy pitfall, the compiler will suggest a template modifier to use where appropriate given the input type.
If all references to an operand already have modifiers then the warning is suppressed for that operand.
ABI clobbers
The clobber_abi
keyword can be used to apply a default set of clobbers to an asm!
block.
This will automatically insert the necessary clobber constraints as needed for calling a function with a particular calling convention: if the calling convention does not fully preserve the value of a register across a call then lateout("...") _
is implicitly added to the operands list (where the ...
is replaced by the register’s name).
clobber_abi
may be specified any number of times. It will insert a clobber for all unique registers in the union of all specified calling conventions.
Generic register class outputs are disallowed by the compiler when clobber_abi
is used: all outputs must specify an explicit register.
Explicit register outputs have precedence over the implicit clobbers inserted by clobber_abi
: a clobber will only be inserted for a register if that register is not used as an output.
The following ABIs can be used with clobber_abi
:
Architecture | ABI name | Clobbered registers |
---|---|---|
x86-32 | "C" , "system" , "efiapi" , "cdecl" , "stdcall" , "fastcall" | ax , cx , dx , xmm[0-7] , mm[0-7] , k[0-7] , st([0-7]) |
x86-64 | "C" , "system" (on Windows), "efiapi" , "win64" | ax , cx , dx , r[8-11] , xmm[0-31] , mm[0-7] , k[0-7] , st([0-7]) , tmm[0-7] |
x86-64 | "C" , "system" (on non-Windows), "sysv64" | ax , cx , dx , si , di , r[8-11] , xmm[0-31] , mm[0-7] , k[0-7] , st([0-7]) , tmm[0-7] |
AArch64 | "C" , "system" , "efiapi" | x[0-17] , x18 *, x30 , v[0-31] , p[0-15] , ffr |
ARM | "C" , "system" , "efiapi" , "aapcs" | r[0-3] , r12 , r14 , s[0-15] , d[0-7] , d[16-31] |
RISC-V | "C" , "system" , "efiapi" | x1 , x[5-7] , x[10-17] , x[28-31] , f[0-7] , f[10-17] , f[28-31] , v[0-31] |
LoongArch | "C" , "system" , "efiapi" | $r1 , $r[4-20] , $f[0-23] |
Notes:
- On AArch64
x18
only included in the clobber list if it is not considered as a reserved register on the target.
The list of clobbered registers for each ABI is updated in rustc as architectures gain new registers: this ensures that asm!
clobbers will continue to be correct when LLVM starts using these new registers in its generated code.
Options
Flags are used to further influence the behavior of the inline assembly block. Currently the following options are defined:
pure
: Theasm!
block has no side effects, must eventually return, and its outputs depend only on its direct inputs (i.e. the values themselves, not what they point to) or values read from memory (unless thenomem
options is also set). This allows the compiler to execute theasm!
block fewer times than specified in the program (e.g. by hoisting it out of a loop) or even eliminate it entirely if the outputs are not used. Thepure
option must be combined with either thenomem
orreadonly
options, otherwise a compile-time error is emitted.
nomem
: Theasm!
blocks does not read or write to any memory. This allows the compiler to cache the values of modified global variables in registers across theasm!
block since it knows that they are not read or written to by theasm!
. The compiler also assumes that thisasm!
block does not perform any kind of synchronization with other threads, e.g. via fences.
readonly
: Theasm!
block does not write to any memory. This allows the compiler to cache the values of unmodified global variables in registers across theasm!
block since it knows that they are not written to by theasm!
. The compiler also assumes that thisasm!
block does not perform any kind of synchronization with other threads, e.g. via fences.
preserves_flags
: Theasm!
block does not modify the flags register (defined in the rules below). This allows the compiler to avoid recomputing the condition flags after theasm!
block.
noreturn
: Theasm!
block never returns, and its return type is defined as!
(never). Behavior is undefined if execution falls through past the end of the asm code. Anoreturn
asm block behaves just like a function which doesn’t return; notably, local variables in scope are not dropped before it is invoked.
nostack
: Theasm!
block does not push data to the stack, or write to the stack red-zone (if supported by the target). If this option is not used then the stack pointer is guaranteed to be suitably aligned (according to the target ABI) for a function call.
att_syntax
: This option is only valid on x86, and causes the assembler to use the.att_syntax prefix
mode of the GNU assembler. Register operands are substituted in with a leading%
.
raw
: This causes the template string to be parsed as a raw assembly string, with no special handling for{
and}
. This is primarily useful when including raw assembly code from an external file usinginclude_str!
.
The compiler performs some additional checks on options:
- The
nomem
andreadonly
options are mutually exclusive: it is a compile-time error to specify both.
- It is a compile-time error to specify
pure
on an asm block with no outputs or only discarded outputs (_
).
- It is a compile-time error to specify
noreturn
on an asm block with outputs.
global_asm!
only supports the att_syntax
and raw
options.
The remaining options are not meaningful for global-scope inline assembly
Rules for inline assembly
To avoid undefined behavior, these rules must be followed when using function-scope inline assembly (asm!
):
- Any registers not specified as inputs will contain an undefined value on entry to the asm block.
- An “undefined value” in the context of inline assembly means that the register can (non-deterministically) have any one of the possible values allowed by the architecture.
Notably it is not the same as an LLVM
undef
which can have a different value every time you read it (since such a concept does not exist in assembly code).
- An “undefined value” in the context of inline assembly means that the register can (non-deterministically) have any one of the possible values allowed by the architecture.
Notably it is not the same as an LLVM
- Any registers not specified as outputs must have the same value upon exiting the asm block as they had on entry, otherwise behavior is undefined.
- This only applies to registers which can be specified as an input or output. Other registers follow target-specific rules.
- Note that a
lateout
may be allocated to the same register as anin
, in which case this rule does not apply. Code should not rely on this however since it depends on the results of register allocation.
- Behavior is undefined if execution unwinds out of an asm block.
- This also applies if the assembly code calls a function which then unwinds.
- The set of memory locations that assembly code is allowed to read and write are the same as those allowed for an FFI function.
- Refer to the unsafe code guidelines for the exact rules.
- If the
readonly
option is set, then only memory reads are allowed. - If the
nomem
option is set then no reads or writes to memory are allowed. - These rules do not apply to memory which is private to the asm code, such as stack space allocated within the asm block.
- The compiler cannot assume that the instructions in the asm are the ones that will actually end up executed.
- This effectively means that the compiler must treat the
asm!
as a black box and only take the interface specification into account, not the instructions themselves. - Runtime code patching is allowed, via target-specific mechanisms.
- However there is no guarantee that each
asm!
directly corresponds to a single instance of instructions in the object file: the compiler is free to duplicate or deduplicateasm!
blocks.
- This effectively means that the compiler must treat the
- Unless the
nostack
option is set, asm code is allowed to use stack space below the stack pointer.- On entry to the asm block the stack pointer is guaranteed to be suitably aligned (according to the target ABI) for a function call.
- You are responsible for making sure you don’t overflow the stack (e.g. use stack probing to ensure you hit a guard page).
- You should adjust the stack pointer when allocating stack memory as required by the target ABI.
- The stack pointer must be restored to its original value before leaving the asm block.
- If the
noreturn
option is set then behavior is undefined if execution falls through to the end of the asm block.
- If the
pure
option is set then behavior is undefined if theasm!
has side-effects other than its direct outputs. Behavior is also undefined if two executions of theasm!
code with the same inputs result