The Unstable Book

Welcome to the Unstable Book! This book consists of a number of chapters, each one organized by a "feature flag." That is, when using an unstable feature of Rust, you must use a flag, like this:

#![feature(box_syntax)]

fn main() {
    let five = box 5;
}

The box_syntax feature has a chapter describing how to use it.

Because this documentation relates to unstable features, we make no guarantees that what is contained here is accurate or up to date. It's developed on a best-effort basis. Each page will have a link to its tracking issue with the latest developments; you might want to check those as well.

Compiler flags

codegen-backend

The tracking issue for this feature is: #77933.


This feature allows you to specify a path to a dynamic library to use as rustc's code generation backend at runtime.

Set the -Zcodegen-backend=<path> compiler flag to specify the location of the backend. The library must be of crate type dylib and must contain a function named __rustc_codegen_backend with a signature of fn() -> Box<dyn rustc_codegen_ssa::traits::CodegenBackend>.

Example

See also the hotplug_codegen_backend test for a full example.

use rustc_codegen_ssa::traits::CodegenBackend;

struct MyBackend;

impl CodegenBackend for MyBackend {
   // Implement codegen methods
}

#[no_mangle]
pub fn __rustc_codegen_backend() -> Box<dyn CodegenBackend> {
    Box::new(MyBackend)
}

control-flow-guard

The tracking issue for this feature is: #68793.


The rustc flag -Z control-flow-guard enables the Windows Control Flow Guard (CFG) platform security feature.

CFG is an exploit mitigation designed to enforce control-flow integrity for software running on supported Windows platforms (Windows 8.1 onwards). Specifically, CFG uses runtime checks to validate the target address of every indirect call/jump before allowing the call to complete.

During compilation, the compiler identifies all indirect calls/jumps and adds CFG checks. It also emits metadata containing the relative addresses of all address-taken functions. At runtime, if the binary is run on a CFG-aware operating system, the loader uses the CFG metadata to generate a bitmap of the address space and marks those addresses that contain valid targets. On each indirect call, the inserted check determines whether the target address is marked in this bitmap. If the target is not valid, the process is terminated.

In terms of interoperability:

  • Code compiled with CFG enabled can be linked with libraries and object files that are not compiled with CFG. In this case, a CFG-aware linker can identify address-taken functions in the non-CFG libraries.
  • Libraries compiled with CFG can linked into non-CFG programs. In this case, the CFG runtime checks in the libraries are not used (i.e. the mitigation is completely disabled).

CFG functionality is completely implemented in the LLVM backend and is supported for X86 (32-bit and 64-bit), ARM, and Aarch64 targets. The rustc flag adds the relevant LLVM module flags to enable the feature. This flag will be ignored for all non-Windows targets.

When to use Control Flow Guard

The primary motivation for enabling CFG in Rust is to enhance security when linking against non-Rust code, especially C/C++ code. To achieve full CFG protection, all indirect calls (including any from Rust code) must have the appropriate CFG checks, as added by this flag. CFG can also improve security for Rust code that uses the unsafe keyword.

Another motivation behind CFG is to harden programs against return-oriented programming (ROP) attacks. CFG disallows an attacker from taking advantage of the program's own instructions while redirecting control flow in unexpected ways.

Overhead of Control Flow Guard

The CFG checks and metadata can potentially increase binary size and runtime overhead. The magnitude of any increase depends on the number and frequency of indirect calls. For example, enabling CFG for the Rust standard library increases binary size by approximately 0.14%. Enabling CFG in the SPEC CPU 2017 Integer Speed benchmark suite (compiled with Clang/LLVM) incurs approximate runtime overheads of between 0% and 8%, with a geometric mean of 2.9%.

Testing Control Flow Guard

The rustc flag -Z control-flow-guard=nochecks instructs LLVM to emit the list of valid call targets without inserting runtime checks. This flag should only be used for testing purposes as it does not provide security enforcement.

Control Flow Guard in libraries

It is strongly recommended to also enable CFG checks for all linked libraries, including the standard library.

To enable CFG in the standard library, use the cargo -Z build-std functionality to recompile the standard library with the same configuration options as the main program.

For example:

rustup toolchain install --force nightly
rustup component add rust-src
SET RUSTFLAGS=-Z control-flow-guard
cargo +nightly build -Z build-std --target x86_64-pc-windows-msvc
rustup toolchain install --force nightly
rustup component add rust-src
$Env:RUSTFLAGS = "-Z control-flow-guard"
cargo +nightly build -Z build-std --target x86_64-pc-windows-msvc

Alternatively, if you are building the standard library from source, you can set control-flow-guard = true in the config.toml file.

emit-stack-sizes

The tracking issue for this feature is: #54192


The rustc flag -Z emit-stack-sizes makes LLVM emit stack size metadata.

NOTE: This LLVM feature only supports the ELF object format as of LLVM 8.0. Using this flag with targets that use other object formats (e.g. macOS and Windows) will result in it being ignored.

Consider this crate:

#![crate_type = "lib"]

use std::ptr;

pub fn foo() {
    // this function doesn't use the stack
}

pub fn bar() {
    let xs = [0u32; 2];

    // force LLVM to allocate `xs` on the stack
    unsafe { ptr::read_volatile(&xs.as_ptr()); }
}

Using the -Z emit-stack-sizes flag produces extra linker sections in the output object file.

$ rustc -C opt-level=3 --emit=obj foo.rs

$ size -A foo.o
foo.o  :
section                                 size   addr
.text                                      0      0
.text._ZN3foo3foo17he211d7b4a3a0c16eE      1      0
.text._ZN3foo3bar17h1acb594305f70c2eE     22      0
.note.GNU-stack                            0      0
.eh_frame                                 72      0
Total                                     95

$ rustc -C opt-level=3 --emit=obj -Z emit-stack-sizes foo.rs

$ size -A foo.o
foo.o  :
section                                 size   addr
.text                                      0      0
.text._ZN3foo3foo17he211d7b4a3a0c16eE      1      0
.stack_sizes                               9      0
.text._ZN3foo3bar17h1acb594305f70c2eE     22      0
.stack_sizes                               9      0
.note.GNU-stack                            0      0
.eh_frame                                 72      0
Total                                    113

As of LLVM 7.0 the data will be written into a section named .stack_sizes and the format is "an array of pairs of function symbol values (pointer size) and stack sizes (unsigned LEB128)".

$ objdump -d foo.o

foo.o:     file format elf64-x86-64

Disassembly of section .text._ZN3foo3foo17he211d7b4a3a0c16eE:

0000000000000000 <_ZN3foo3foo17he211d7b4a3a0c16eE>:
   0:   c3                      retq

Disassembly of section .text._ZN3foo3bar17h1acb594305f70c2eE:

0000000000000000 <_ZN3foo3bar17h1acb594305f70c2eE>:
   0:   48 83 ec 10             sub    $0x10,%rsp
   4:   48 8d 44 24 08          lea    0x8(%rsp),%rax
   9:   48 89 04 24             mov    %rax,(%rsp)
   d:   48 8b 04 24             mov    (%rsp),%rax
  11:   48 83 c4 10             add    $0x10,%rsp
  15:   c3                      retq

$ objdump -s -j .stack_sizes foo.o

foo.o:     file format elf64-x86-64

Contents of section .stack_sizes:
 0000 00000000 00000000 00                 .........
Contents of section .stack_sizes:
 0000 00000000 00000000 10                 .........

It's important to note that linkers will discard this linker section by default. To preserve the section you can use a linker script like the one shown below.

/* file: keep-stack-sizes.x */
SECTIONS
{
  /* `INFO` makes the section not allocatable so it won't be loaded into memory */
  .stack_sizes (INFO) :
  {
    KEEP(*(.stack_sizes));
  }
}

The linker script must be passed to the linker using a rustc flag like -C link-arg.

// file: src/main.rs
use std::ptr;

#[inline(never)]
fn main() {
    let xs = [0u32; 2];

    // force LLVM to allocate `xs` on the stack
    unsafe { ptr::read_volatile(&xs.as_ptr()); }
}
$ RUSTFLAGS="-Z emit-stack-sizes" cargo build --release

$ size -A target/release/hello | grep stack_sizes || echo section was not found
section was not found

$ RUSTFLAGS="-Z emit-stack-sizes" cargo rustc --release -- \
    -C link-arg=-Wl,-Tkeep-stack-sizes.x \
    -C link-arg=-N

$ size -A target/release/hello | grep stack_sizes
.stack_sizes                               90   176272

$ # non-allocatable section (flags don't contain the "A" (alloc) flag)
$ readelf -S target/release/hello
Section Headers:
  [Nr]   Name              Type             Address           Offset
       Size              EntSize            Flags  Link  Info  Align
(..)
  [1031] .stack_sizes      PROGBITS         000000000002b090  0002b0f0
       000000000000005a  0000000000000000   L       5     0     1

$ objdump -s -j .stack_sizes target/release/hello

target/release/hello:     file format elf64-x86-64

Contents of section .stack_sizes:
 2b090 c0040000 00000000 08f00400 00000000  ................
 2b0a0 00080005 00000000 00000810 05000000  ................
 2b0b0 00000000 20050000 00000000 10400500  .... ........@..
 2b0c0 00000000 00087005 00000000 00000080  ......p.........
 2b0d0 05000000 00000000 90050000 00000000  ................
 2b0e0 00a00500 00000000 0000               ..........

Author note: I'm not entirely sure why, in this case, -N is required in addition to -Tkeep-stack-sizes.x. For example, it's not required when producing statically linked files for the ARM Cortex-M architecture.

profile

The tracking issue for this feature is: #42524.


This feature allows the generation of code coverage reports.

Set the -Zprofile compiler flag in order to enable gcov profiling.

For example:

cargo new testgcov --bin
cd testgcov
export RUSTFLAGS="-Zprofile -Ccodegen-units=1 -Copt-level=0 -Clink-dead-code -Coverflow-checks=off -Zpanic_abort_tests -Cpanic=abort"
export CARGO_INCREMENTAL=0
cargo build
cargo run

Once you've built and run your program, files with the gcno (after build) and gcda (after execution) extensions will be created. You can parse them with llvm-cov gcov or grcov.

Please note that RUSTFLAGS by default applies to everything that cargo builds and runs during a build! When the --target flag is explicitly passed to cargo, the RUSTFLAGS no longer apply to build scripts and procedural macros. For more fine-grained control consider passing a RUSTC_WRAPPER program to cargo that only adds the profiling flags to rustc for the specific crates you want to profile.

report-time

The tracking issue for this feature is: #64888


The report-time feature adds a possibility to report execution time of the tests generated via libtest.

This is unstable feature, so you have to provide -Zunstable-options to get this feature working.

Sample usage command:

./test_executable -Zunstable-options --report-time

Available options:

--report-time [plain|colored]
                Show execution time of each test. Available values:
                plain = do not colorize the execution time (default);
                colored = colorize output according to the `color`
                parameter value;
                Threshold values for colorized output can be
                configured via
                `RUST_TEST_TIME_UNIT`, `RUST_TEST_TIME_INTEGRATION`
                and
                `RUST_TEST_TIME_DOCTEST` environment variables.
                Expected format of environment variable is
                `VARIABLE=WARN_TIME,CRITICAL_TIME`.
                Not available for --format=terse
--ensure-time 
                Treat excess of the test execution time limit as
                error.
                Threshold values for this option can be configured via
                `RUST_TEST_TIME_UNIT`, `RUST_TEST_TIME_INTEGRATION`
                and
                `RUST_TEST_TIME_DOCTEST` environment variables.
                Expected format of environment variable is
                `VARIABLE=WARN_TIME,CRITICAL_TIME`.
                `CRITICAL_TIME` here means the limit that should not be
                exceeded by test.

Example of the environment variable format:

RUST_TEST_TIME_UNIT=100,200

where 100 stands for warn time, and 200 stands for critical time.

Examples

cargo test --tests -- -Zunstable-options --report-time
    Finished dev [unoptimized + debuginfo] target(s) in 0.02s
     Running target/debug/deps/example-27fb188025bec02c

running 3 tests
test tests::unit_test_quick ... ok <0.000s>
test tests::unit_test_warn ... ok <0.055s>
test tests::unit_test_critical ... ok <0.110s>

test result: ok. 3 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

     Running target/debug/deps/tests-cedb06f6526d15d9

running 3 tests
test unit_test_quick ... ok <0.000s>
test unit_test_warn ... ok <0.550s>
test unit_test_critical ... ok <1.100s>

test result: ok. 3 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

sanitizer

The tracking issue for this feature is: #39699.


This feature allows for use of one of following sanitizers:

To enable a sanitizer compile with -Zsanitizer=address, -Zsanitizer=leak, -Zsanitizer=memory or -Zsanitizer=thread.

AddressSanitizer

AddressSanitizer is a memory error detector. It can detect the following types of bugs:

  • Out of bound accesses to heap, stack and globals
  • Use after free
  • Use after return (runtime flag ASAN_OPTIONS=detect_stack_use_after_return=1)
  • Use after scope
  • Double-free, invalid free
  • Memory leaks

The memory leak detection is enabled by default on Linux, and can be enabled with runtime flag ASAN_OPTIONS=detect_leaks=1 on macOS.

AddressSanitizer is supported on the following targets:

  • x86_64-apple-darwin
  • x86_64-unknown-linux-gnu

AddressSanitizer works with non-instrumented code although it will impede its ability to detect some bugs. It is not expected to produce false positive reports.

Examples

Stack buffer overflow:

fn main() {
    let xs = [0, 1, 2, 3];
    let _y = unsafe { *xs.as_ptr().offset(4) };
}
$ export RUSTFLAGS=-Zsanitizer=address RUSTDOCFLAGS=-Zsanitizer=address
$ cargo run -Zbuild-std --target x86_64-unknown-linux-gnu
==37882==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x7ffe400e6250 at pc 0x5609a841fb20 bp 0x7ffe400e6210 sp 0x7ffe400e6208
READ of size 4 at 0x7ffe400e6250 thread T0
    #0 0x5609a841fb1f in example::main::h628ffc6626ed85b2 /.../src/main.rs:3:23
    ...

Address 0x7ffe400e6250 is located in stack of thread T0 at offset 48 in frame
    #0 0x5609a841f8af in example::main::h628ffc6626ed85b2 /.../src/main.rs:1

  This frame has 1 object(s):
    [32, 48) 'xs' (line 2) <== Memory access at offset 48 overflows this variable
HINT: this may be a false positive if your program uses some custom stack unwind mechanism, swapcontext or vfork
      (longjmp and C++ exceptions *are* supported)
SUMMARY: AddressSanitizer: stack-buffer-overflow /.../src/main.rs:3:23 in example::main::h628ffc6626ed85b2
Shadow bytes around the buggy address:
  0x100048014bf0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x100048014c00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x100048014c10: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x100048014c20: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x100048014c30: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
=>0x100048014c40: 00 00 00 00 f1 f1 f1 f1 00 00[f3]f3 00 00 00 00
  0x100048014c50: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x100048014c60: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x100048014c70: f1 f1 f1 f1 00 00 f3 f3 00 00 00 00 00 00 00 00
  0x100048014c80: 00 00 00 00 00 00 00 00 00 00 00 00 f1 f1 f1 f1
  0x100048014c90: 00 00 f3 f3 00 00 00 00 00 00 00 00 00 00 00 00
Shadow byte legend (one shadow byte represents 8 application bytes):
  Addressable:           00
  Partially addressable: 01 02 03 04 05 06 07
  Heap left redzone:       fa
  Freed heap region:       fd
  Stack left redzone:      f1
  Stack mid redzone:       f2
  Stack right redzone:     f3
  Stack after return:      f5
  Stack use after scope:   f8
  Global redzone:          f9
  Global init order:       f6
  Poisoned by user:        f7
  Container overflow:      fc
  Array cookie:            ac
  Intra object redzone:    bb
  ASan internal:           fe
  Left alloca redzone:     ca
  Right alloca redzone:    cb
  Shadow gap:              cc
==37882==ABORTING

Use of a stack object after its scope has already ended:

static mut P: *mut usize = std::ptr::null_mut();

fn main() {
    unsafe {
        {
            let mut x = 0;
            P = &mut x;
        }
        std::ptr::write_volatile(P, 123);
    }
}
$ export RUSTFLAGS=-Zsanitizer=address RUSTDOCFLAGS=-Zsanitizer=address
$ cargo run -Zbuild-std --target x86_64-unknown-linux-gnu
=================================================================
==39249==ERROR: AddressSanitizer: stack-use-after-scope on address 0x7ffc7ed3e1a0 at pc 0x55c98b262a8e bp 0x7ffc7ed3e050 sp 0x7ffc7ed3e048
WRITE of size 8 at 0x7ffc7ed3e1a0 thread T0
    #0 0x55c98b262a8d in core::ptr::write_volatile::he21f1df5a82f329a /.../src/rust/src/libcore/ptr/mod.rs:1048:5
    #1 0x55c98b262cd2 in example::main::h628ffc6626ed85b2 /.../src/main.rs:9:9
    ...

Address 0x7ffc7ed3e1a0 is located in stack of thread T0 at offset 32 in frame
    #0 0x55c98b262bdf in example::main::h628ffc6626ed85b2 /.../src/main.rs:3

  This frame has 1 object(s):
    [32, 40) 'x' (line 6) <== Memory access at offset 32 is inside this variable
HINT: this may be a false positive if your program uses some custom stack unwind mechanism, swapcontext or vfork
      (longjmp and C++ exceptions *are* supported)
SUMMARY: AddressSanitizer: stack-use-after-scope /.../src/rust/src/libcore/ptr/mod.rs:1048:5 in core::ptr::write_volatile::he21f1df5a82f329a
Shadow bytes around the buggy address:
  0x10000fd9fbe0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x10000fd9fbf0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x10000fd9fc00: 00 00 00 00 00 00 00 00 00 00 00 00 f1 f1 f1 f1
  0x10000fd9fc10: f8 f8 f3 f3 00 00 00 00 00 00 00 00 00 00 00 00
  0x10000fd9fc20: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
=>0x10000fd9fc30: f1 f1 f1 f1[f8]f3 f3 f3 00 00 00 00 00 00 00 00
  0x10000fd9fc40: 00 00 00 00 00 00 00 00 00 00 00 00 f1 f1 f1 f1
  0x10000fd9fc50: 00 00 f3 f3 00 00 00 00 00 00 00 00 00 00 00 00
  0x10000fd9fc60: 00 00 00 00 00 00 00 00 f1 f1 f1 f1 00 00 f3 f3
  0x10000fd9fc70: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x10000fd9fc80: 00 00 00 00 f1 f1 f1 f1 00 00 f3 f3 00 00 00 00
Shadow byte legend (one shadow byte represents 8 application bytes):
  Addressable:           00
  Partially addressable: 01 02 03 04 05 06 07
  Heap left redzone:       fa
  Freed heap region:       fd
  Stack left redzone:      f1
  Stack mid redzone:       f2
  Stack right redzone:     f3
  Stack after return:      f5
  Stack use after scope:   f8
  Global redzone:          f9
  Global init order:       f6
  Poisoned by user:        f7
  Container overflow:      fc
  Array cookie:            ac
  Intra object redzone:    bb
  ASan internal:           fe
  Left alloca redzone:     ca
  Right alloca redzone:    cb
  Shadow gap:              cc
==39249==ABORTING

MemorySanitizer

MemorySanitizer is detector of uninitialized reads. It is only supported on the x86_64-unknown-linux-gnu target.

MemorySanitizer requires all program code to be instrumented. C/C++ dependencies need to be recompiled using Clang with -fsanitize=memory option. Failing to achieve that will result in false positive reports.

Example

Detecting the use of uninitialized memory. The -Zbuild-std flag rebuilds and instruments the standard library, and is strictly necessary for the correct operation of the tool. The -Zsanitizer-track-origins enables tracking of the origins of uninitialized memory:

use std::mem::MaybeUninit;

fn main() {
    unsafe {
        let a = MaybeUninit::<[usize; 4]>::uninit();
        let a = a.assume_init();
        println!("{}", a[2]);
    }
}
$ export \
  RUSTFLAGS='-Zsanitizer=memory -Zsanitizer-memory-track-origins' \
  RUSTDOCFLAGS='-Zsanitizer=memory -Zsanitizer-memory-track-origins'
$ cargo clean
$ cargo run -Zbuild-std --target x86_64-unknown-linux-gnu
==9416==WARNING: MemorySanitizer: use-of-uninitialized-value
    #0 0x560c04f7488a in core::fmt::num::imp::fmt_u64::haa293b0b098501ca $RUST/build/x86_64-unknown-linux-gnu/stage1/lib/rustlib/src/rust/src/libcore/fmt/num.rs:202:16
...
  Uninitialized value was stored to memory at
    #0 0x560c04ae898a in __msan_memcpy.part.0 $RUST/src/llvm-project/compiler-rt/lib/msan/msan_interceptors.cc:1558:3
    #1 0x560c04b2bf88 in memory::main::hd2333c1899d997f5 $CWD/src/main.rs:6:16

  Uninitialized value was created by an allocation of 'a' in the stack frame of function '_ZN6memory4main17hd2333c1899d997f5E'
    #0 0x560c04b2bc50 in memory::main::hd2333c1899d997f5 $CWD/src/main.rs:3

ThreadSanitizer

ThreadSanitizer is a data race detection tool. It is supported on the following targets:

  • x86_64-apple-darwin
  • x86_64-unknown-linux-gnu

To work correctly ThreadSanitizer needs to be "aware" of all synchronization operations in a program. It generally achieves that through combination of library interception (for example synchronization performed through pthread_mutex_lock / pthread_mutex_unlock) and compile time instrumentation (e.g. atomic operations). Using it without instrumenting all the program code can lead to false positive reports.

ThreadSanitizer does not support atomic fences std::sync::atomic::fence, nor synchronization performed using inline assembly code.

Example

static mut A: usize = 0;

fn main() {
    let t = std::thread::spawn(|| {
        unsafe { A += 1 };
    });
    unsafe { A += 1 };

    t.join().unwrap();
}
$ export RUSTFLAGS=-Zsanitizer=thread RUSTDOCFLAGS=-Zsanitizer=thread
$ cargo run -Zbuild-std --target x86_64-unknown-linux-gnu
==================
WARNING: ThreadSanitizer: data race (pid=10574)
  Read of size 8 at 0x5632dfe3d030 by thread T1:
    #0 example::main::_$u7b$$u7b$closure$u7d$$u7d$::h23f64b0b2f8c9484 ../src/main.rs:5:18 (example+0x86cec)
    ...

  Previous write of size 8 at 0x5632dfe3d030 by main thread:
    #0 example::main::h628ffc6626ed85b2 /.../src/main.rs:7:14 (example+0x868c8)
    ...
    #11 main <null> (example+0x86a1a)

  Location is global 'example::A::h43ac149ddf992709' of size 8 at 0x5632dfe3d030 (example+0x000000bd9030)

Instrumentation of external dependencies and std

The sanitizers to varying degrees work correctly with partially instrumented code. On the one extreme is LeakSanitizer that doesn't use any compile time instrumentation, on the other is MemorySanitizer that requires that all program code to be instrumented (failing to achieve that will inevitably result in false positives).

It is strongly recommended to combine sanitizers with recompiled and instrumented standard library, for example using cargo -Zbuild-std functionality.

Build scripts and procedural macros

Use of sanitizers together with build scripts and procedural macros is technically possible, but in almost all cases it would be best avoided. This is especially true for procedural macros which would require an instrumented version of rustc.

In more practical terms when using cargo always remember to pass --target flag, so that rustflags will not be applied to build scripts and procedural macros.

Symbolizing the Reports

Sanitizers produce symbolized stacktraces when llvm-symbolizer binary is in PATH.

Additional Information

self-profile


The -Zself-profile compiler flag enables rustc's internal profiler. When enabled, the compiler will output three binary files in the specified directory (or the current working directory if no directory is specified). These files can be analyzed by using the tools in the measureme repository.

To control the data recorded in the trace files, use the -Zself-profile-events flag.

For example:

First, run a compilation session and provide the -Zself-profile flag:

$ rustc --crate-name foo -Zself-profile

This will generate three files in the working directory such as:

  • foo-1234.events
  • foo-1234.string_data
  • foo-1234.string_index

Where foo is the name of the crate and 1234 is the process id of the rustc process.

To get a summary of where the compiler is spending its time:

$ ../measureme/target/release/summarize summarize foo-1234

To generate a flamegraph of the same data:

$ ../measureme/target/release/inferno foo-1234

To dump the event data in a Chromium-profiler compatible format:

$ ../measureme/target/release/crox foo-1234

For more information, consult the measureme documentation.

self-profile-events


The -Zself-profile-events compiler flag controls what events are recorded by the self-profiler when it is enabled via the -Zself-profile flag.

This flag takes a comma delimited list of event types to record.

For example:

$ rustc -Zself-profile -Zself-profile-events=default,args

Event types

  • query-provider

    • Traces each query used internally by the compiler.
  • generic-activity

    • Traces other parts of the compiler not covered by the query system.
  • query-cache-hit

    • Adds tracing information that records when the in-memory query cache is "hit" and does not need to re-execute a query which has been cached.
    • Disabled by default because this significantly increases the trace file size.
  • query-blocked

    • Tracks time that a query tries to run but is blocked waiting on another thread executing the same query to finish executing.
    • Query blocking only occurs when the compiler is built with parallel mode support.
  • incr-cache-load

    • Tracks time that is spent loading and deserializing query results from the incremental compilation on-disk cache.
  • query-keys

    • Adds a serialized representation of each query's query key to the tracing data.
    • Disabled by default because this significantly increases the trace file size.
  • function-args

    • Adds additional tracing data to some generic-activity events.
    • Disabled by default for parity with query-keys.
  • llvm

    • Adds tracing information about LLVM passes and codegeneration.
    • Disabled by default because this only works when -Znew-llvm-pass-manager is enabled.

Event synonyms

  • none

    • Disables all events. Equivalent to the self-profiler being disabled.
  • default

    • The default set of events which stikes a balance between providing detailed tracing data and adding additional overhead to the compilation.
  • args

    • Equivalent to query-keys and function-args.
  • all

    • Enables all events.

Examples

Enable the profiler and capture the default set of events (both invocations are equivalent):

$ rustc -Zself-profile
$ rustc -Zself-profile -Zself-profile-events=default

Enable the profiler and capture the default events and their arguments:

$ rustc -Zself-profile -Zself-profile-events=default,args

src-hash-algorithm

The tracking issue for this feature is: #70401.


The -Z src-hash-algorithm compiler flag controls which algorithm is used when hashing each source file. The hash is stored in the debug info and can be used by a debugger to verify the source code matches the executable.

Supported hash algorithms are: md5, and sha1. Note that not all hash algorithms are supported by all debug info formats.

By default, the compiler chooses the hash algorithm based on the target specification.

strip

The tracking issue for this feature is: #72110.


Option -Z strip=val controls stripping of debuginfo and similar auxiliary data from binaries during linking.

Supported values for this option are:

  • none - debuginfo and symbols (if they exist) are copied to the produced binary or separate files depending on the target (e.g. .pdb files in case of MSVC).
  • debuginfo - debuginfo sections and debuginfo symbols from the symbol table section are stripped at link time and are not copied to the produced binary or separate files.
  • symbols - same as debuginfo, but the rest of the symbol table section is stripped as well if the linker supports it.

tls_model

The tracking issue for this feature is: None.


Option -Z tls-model controls TLS model used to generate code for accessing #[thread_local] static items.

Supported values for this option are:

  • global-dynamic - General Dynamic TLS Model (alternatively called Global Dynamic) is the most general option usable in all circumstances, even if the TLS data is defined in a shared library loaded at runtime and is accessed from code outside of that library.
    This is the default for most targets.
  • local-dynamic - model usable if the TLS data is only accessed from the shared library or executable it is defined in. The TLS data may be in a library loaded after startup (via dlopen).
  • initial-exec - model usable if the TLS data is defined in the executable or in a shared library loaded at program startup. The TLS data must not be in a library loaded after startup (via dlopen).
  • local-exec - model usable only if the TLS data is defined directly in the executable, but not in a shared library, and is accessed only from that executable.

rustc and LLVM may use a more optimized model than specified if they know that we are producing an executable rather than a library, or that the static item is private enough.

unsound-mir-opts


The -Zunsound-mir-opts compiler flag enables MIR optimization passes which can cause unsound behavior. This flag should only be used by MIR optimization tests in the rustc test suite.

Language features

aarch64_target_feature

The tracking issue for this feature is: #44839


abi_amdgpu_kernel

The tracking issue for this feature is: #51575


abi_avr_interrupt

The tracking issue for this feature is: #69664


abi_efiapi

The tracking issue for this feature is: #65815


abi_msp430_interrupt

The tracking issue for this feature is: #38487


In the MSP430 architecture, interrupt handlers have a special calling convention. You can use the "msp430-interrupt" ABI to make the compiler apply the right calling convention to the interrupt handlers you define.

#![feature(abi_msp430_interrupt)]
#![no_std]

// Place the interrupt handler at the appropriate memory address
// (Alternatively, you can use `#[used]` and remove `pub` and `#[no_mangle]`)
#[link_section = "__interrupt_vector_10"]
#[no_mangle]
pub static TIM0_VECTOR: extern "msp430-interrupt" fn() = tim0;

// The interrupt handler
extern "msp430-interrupt" fn tim0() {
    // ..
}
$ msp430-elf-objdump -CD ./target/msp430/release/app
Disassembly of section __interrupt_vector_10:

0000fff2 <TIM0_VECTOR>:
    fff2:       00 c0           interrupt service routine at 0xc000

Disassembly of section .text:

0000c000 <int::tim0>:
    c000:       00 13           reti

abi_ptx

The tracking issue for this feature is: #38788


When emitting PTX code, all vanilla Rust functions (fn) get translated to "device" functions. These functions are not callable from the host via the CUDA API so a crate with only device functions is not too useful!

OTOH, "global" functions can be called by the host; you can think of them as the real public API of your crate. To produce a global function use the "ptx-kernel" ABI.

#![feature(abi_ptx)]
#![no_std]

pub unsafe extern "ptx-kernel" fn global_function() {
    device_function();
}

pub fn device_function() {
    // ..
}
$ xargo rustc --target nvptx64-nvidia-cuda --release -- --emit=asm

$ cat $(find -name '*.s')
//
// Generated by LLVM NVPTX Back-End
//

.version 3.2
.target sm_20
.address_size 64

        // .globl       _ZN6kernel15global_function17h46111ebe6516b382E

.visible .entry _ZN6kernel15global_function17h46111ebe6516b382E()
{


        ret;
}

        // .globl       _ZN6kernel15device_function17hd6a0e4993bbf3f78E
.visible .func _ZN6kernel15device_function17hd6a0e4993bbf3f78E()
{


        ret;
}

abi_thiscall

The tracking issue for this feature is: #42202


The MSVC ABI on x86 Windows uses the thiscall calling convention for C++ instance methods by default; it is identical to the usual (C) calling convention on x86 Windows except that the first parameter of the method, the this pointer, is passed in the ECX register.

abi_unadjusted

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


abi_vectorcall

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


abi_x86_interrupt

The tracking issue for this feature is: #40180


adx_target_feature

The tracking issue for this feature is: #44839


alloc_error_handler

The tracking issue for this feature is: #51540


allocator_internals

This feature does not have a tracking issue, it is an unstable implementation detail of the global_allocator feature not intended for use outside the compiler.


allow_fail

The tracking issue for this feature is: #46488


allow_internal_unsafe

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


allow_internal_unstable

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


arbitrary_enum_discriminant

The tracking issue for this feature is: #60553


The arbitrary_enum_discriminant feature permits tuple-like and struct-like enum variants with #[repr(<int-type>)] to have explicit discriminants.

Examples


#![allow(unused)]
#![feature(arbitrary_enum_discriminant)]

fn main() {
#[allow(dead_code)]
#[repr(u8)]
enum Enum {
    Unit = 3,
    Tuple(u16) = 2,
    Struct {
        a: u8,
        b: u16,
    } = 1,
}

impl Enum {
    fn tag(&self) -> u8 {
        unsafe { *(self as *const Self as *const u8) }
    }
}

assert_eq!(3, Enum::Unit.tag());
assert_eq!(2, Enum::Tuple(5).tag());
assert_eq!(1, Enum::Struct{a: 7, b: 11}.tag());
}

arbitrary_self_types

The tracking issue for this feature is: #44874


arm_target_feature

The tracking issue for this feature is: #44839


associated_type_bounds

The tracking issue for this feature is: #52662


associated_type_defaults

The tracking issue for this feature is: #29661


async_closure

The tracking issue for this feature is: #62290


avx512_target_feature

The tracking issue for this feature is: #44839


bindings_after_at

The tracking issue for this feature is: #65490


box_patterns

The tracking issue for this feature is: #29641

See also box_syntax


Box patterns let you match on Box<T>s:

#![feature(box_patterns)]

fn main() {
    let b = Some(Box::new(5));
    match b {
        Some(box n) if n < 0 => {
            println!("Box contains negative number {}", n);
        },
        Some(box n) if n >= 0 => {
            println!("Box contains non-negative number {}", n);
        },
        None => {
            println!("No box");
        },
        _ => unreachable!()
    }
}

box_syntax

The tracking issue for this feature is: #49733

See also box_patterns


Currently the only stable way to create a Box is via the Box::new method. Also it is not possible in stable Rust to destructure a Box in a match pattern. The unstable box keyword can be used to create a Box. An example usage would be:

#![feature(box_syntax)]

fn main() {
    let b = box 5;
}

c_variadic

The tracking issue for this feature is: #44930


The c_variadic language feature enables C-variadic functions to be defined in Rust. The may be called both from within Rust and via FFI.

Examples


#![allow(unused)]
#![feature(c_variadic)]

fn main() {
pub unsafe extern "C" fn add(n: usize, mut args: ...) -> usize {
    let mut sum = 0;
    for _ in 0..n {
        sum += args.arg::<usize>();
    }
    sum
}
}

cfg_sanitize

The tracking issue for this feature is: #39699


The cfg_sanitize feature makes it possible to execute different code depending on whether a particular sanitizer is enabled or not.

Examples


#![allow(unused)]
#![feature(cfg_sanitize)]

fn main() {
#[cfg(sanitize = "thread")]
fn a() {
    // ...
}

#[cfg(not(sanitize = "thread"))]
fn a() {
    // ...
}

fn b() {
    if cfg!(sanitize = "leak") {
        // ...
    } else {
        // ...
    }
}
}

cfg_target_has_atomic

The tracking issue for this feature is: #32976


cfg_target_thread_local

The tracking issue for this feature is: #29594


cfg_version

The tracking issue for this feature is: #64796


The cfg_version feature makes it possible to execute different code depending on the compiler version.

Examples


#![allow(unused)]
#![feature(cfg_version)]

fn main() {
#[cfg(version("1.42"))]
fn a() {
    // ...
}

#[cfg(not(version("1.42")))]
fn a() {
    // ...
}

fn b() {
    if cfg!(version("1.42")) {
        // ...
    } else {
        // ...
    }
}
}

cmpxchg16b_target_feature

The tracking issue for this feature is: #44839


cmse_nonsecure_entry

The tracking issue for this feature is: #75835


The TrustZone-M feature is available for targets with the Armv8-M architecture profile (thumbv8m in their target name). LLVM, the Rust compiler and the linker are providing support for the TrustZone-M feature.

One of the things provided, with this unstable feature, is the cmse_nonsecure_entry attribute. This attribute marks a Secure function as an entry function (see section 5.4 for details). With this attribute, the compiler will do the following:

  • add a special symbol on the function which is the __acle_se_ prefix and the standard function name
  • constrain the number of parameters to avoid using the Non-Secure stack
  • before returning from the function, clear registers that might contain Secure information
  • use the BXNS instruction to return

Because the stack can not be used to pass parameters, there will be compilation errors if:

  • the total size of all parameters is too big (for example more than four 32 bits integers)
  • the entry function is not using a C ABI

The special symbol __acle_se_ will be used by the linker to generate a secure gateway veneer.

#![feature(cmse_nonsecure_entry)]

#[no_mangle]
#[cmse_nonsecure_entry]
pub extern "C" fn entry_function(input: u32) -> u32 {
    input + 6
}
$ rustc --emit obj --crate-type lib --target thumbv8m.main-none-eabi function.rs
$ arm-none-eabi-objdump -D function.o

00000000 <entry_function>:
   0:   b580            push    {r7, lr}
   2:   466f            mov     r7, sp
   4:   b082            sub     sp, #8
   6:   9001            str     r0, [sp, #4]
   8:   1d81            adds    r1, r0, #6
   a:   460a            mov     r2, r1
   c:   4281            cmp     r1, r0
   e:   9200            str     r2, [sp, #0]
  10:   d30b            bcc.n   2a <entry_function+0x2a>
  12:   e7ff            b.n     14 <entry_function+0x14>
  14:   9800            ldr     r0, [sp, #0]
  16:   b002            add     sp, #8
  18:   e8bd 4080       ldmia.w sp!, {r7, lr}
  1c:   4671            mov     r1, lr
  1e:   4672            mov     r2, lr
  20:   4673            mov     r3, lr
  22:   46f4            mov     ip, lr
  24:   f38e 8800       msr     CPSR_f, lr
  28:   4774            bxns    lr
  2a:   f240 0000       movw    r0, #0
  2e:   f2c0 0000       movt    r0, #0
  32:   f240 0200       movw    r2, #0
  36:   f2c0 0200       movt    r2, #0
  3a:   211c            movs    r1, #28
  3c:   f7ff fffe       bl      0 <_ZN4core9panicking5panic17h5c028258ca2fb3f5E>
  40:   defe            udf     #254    ; 0xfe

compiler_builtins

This feature is internal to the Rust compiler and is not intended for general use.


const_eval_limit

The tracking issue for this feature is: #67217

The const_eval_limit allows someone to limit the evaluation steps the CTFE undertakes to evaluate a const fn.

const_evaluatable_checked

The tracking issue for this feature is: #76560


const_extern_fn

The tracking issue for this feature is: #64926


const_fn

The tracking issue for this feature is: #57563


The const_fn feature allows marking free functions and inherent methods as const, enabling them to be called in constants contexts, with constant arguments.

Examples

#![feature(const_fn)]

const fn double(x: i32) -> i32 {
    x * 2
}

const FIVE: i32 = 5;
const TEN: i32 = double(FIVE);

fn main() {
    assert_eq!(5, FIVE);
    assert_eq!(10, TEN);
}

const_fn_floating_point_arithmetic

The tracking issue for this feature is: #57241


const_fn_fn_ptr_basics

The tracking issue for this feature is: #57563


const_fn_transmute

The tracking issue for this feature is: #53605


const_fn_union

The tracking issue for this feature is: #51909


const_generics

The tracking issue for this feature is: #44580


const_impl_trait

The tracking issue for this feature is: #77463


const_in_array_repeat_expressions

The tracking issue for this feature is: #49147


Relaxes the rules for repeat expressions, [x; N] such that x may also be const (strictly speaking rvalue promotable), in addition to typeof(x): Copy. The result of [x; N] where x is const is itself also const.

const_mut_refs

The tracking issue for this feature is: #57349


const_panic

The tracking issue for this feature is: #51999


const_precise_live_drops

The tracking issue for this feature is: #73255


const_raw_ptr_deref

The tracking issue for this feature is: #51911


const_raw_ptr_to_usize_cast

The tracking issue for this feature is: #51910


const_trait_bound_opt_out

The tracking issue for this feature is: #67794


const_trait_impl

The tracking issue for this feature is: #67792


crate_visibility_modifier

The tracking issue for this feature is: #53120


The crate_visibility_modifier feature allows the crate keyword to be used as a visibility modifier synonymous to pub(crate), indicating that a type (function, &c.) is to be visible to the entire enclosing crate, but not to other crates.


#![allow(unused)]
#![feature(crate_visibility_modifier)]

fn main() {
crate struct Foo {
    bar: usize,
}
}

custom_inner_attributes

The tracking issue for this feature is: #54726


custom_test_frameworks

The tracking issue for this feature is: #50297


The custom_test_frameworks feature allows the use of #[test_case] and #![test_runner]. Any function, const, or static can be annotated with #[test_case] causing it to be aggregated (like #[test]) and be passed to the test runner determined by the #![test_runner] crate attribute.


#![allow(unused)]
#![feature(custom_test_frameworks)]
#![test_runner(my_runner)]

fn main() {
fn my_runner(tests: &[&i32]) {
    for t in tests {
        if **t == 0 {
            println!("PASSED");
        } else {
            println!("FAILED");
        }
    }
}

#[test_case]
const WILL_PASS: i32 = 0;

#[test_case]
const WILL_FAIL: i32 = 4;
}

decl_macro

The tracking issue for this feature is: #39412


default_alloc_error_handler

The tracking issue for this feature is: #66741


default_type_parameter_fallback

The tracking issue for this feature is: #27336


doc_cfg

The tracking issue for this feature is: #43781


The doc_cfg feature allows an API be documented as only available in some specific platforms. This attribute has two effects:

  1. In the annotated item's documentation, there will be a message saying "This is supported on (platform) only".

  2. The item's doc-tests will only run on the specific platform.

In addition to allowing the use of the #[doc(cfg)] attribute, this feature enables the use of a special conditional compilation flag, #[cfg(doc)], set whenever building documentation on your crate.

This feature was introduced as part of PR #43348 to allow the platform-specific parts of the standard library be documented.


#![allow(unused)]
#![feature(doc_cfg)]

fn main() {
#[cfg(any(windows, doc))]
#[doc(cfg(windows))]
/// The application's icon in the notification area (a.k.a. system tray).
///
/// # Examples
///
/// ```no_run
/// extern crate my_awesome_ui_library;
/// use my_awesome_ui_library::current_app;
/// use my_awesome_ui_library::windows::notification;
///
/// let icon = current_app().get::<notification::Icon>();
/// icon.show();
/// icon.show_message("Hello");
/// ```
pub struct Icon {
    // ...
}
}

doc_keyword

The tracking issue for this feature is: #51315


doc_masked

The tracking issue for this feature is: #44027


The doc_masked feature allows a crate to exclude types from a given crate from appearing in lists of trait implementations. The specifics of the feature are as follows:

  1. When rustdoc encounters an extern crate statement annotated with a #[doc(masked)] attribute, it marks the crate as being masked.

  2. When listing traits a given type implements, rustdoc ensures that traits from masked crates are not emitted into the documentation.

  3. When listing types that implement a given trait, rustdoc ensures that types from masked crates are not emitted into the documentation.

This feature was introduced in PR #44026 to ensure that compiler-internal and implementation-specific types and traits were not included in the standard library's documentation. Such types would introduce broken links into the documentation.

doc_spotlight

The tracking issue for this feature is: #45040

The doc_spotlight feature allows the use of the spotlight parameter to the #[doc] attribute, to "spotlight" a specific trait on the return values of functions. Adding a #[doc(spotlight)] attribute to a trait definition will make rustdoc print extra information for functions which return a type that implements that trait. This attribute is applied to the Iterator, io::Read, and io::Write traits in the standard library.

You can do this on your own traits, like this:

#![feature(doc_spotlight)]

#[doc(spotlight)]
pub trait MyTrait {}

pub struct MyStruct;
impl MyTrait for MyStruct {}

/// The docs for this function will have an extra line about `MyStruct` implementing `MyTrait`,
/// without having to write that yourself!
pub fn my_fn() -> MyStruct { MyStruct }

This feature was originally implemented in PR #45039.

dropck_eyepatch

The tracking issue for this feature is: #34761


exclusive_range_pattern

The tracking issue for this feature is: #37854


exhaustive_patterns

The tracking issue for this feature is: #51085


extern_types

The tracking issue for this feature is: #43467


external_doc

The tracking issue for this feature is: #44732

The external_doc feature allows the use of the include parameter to the #[doc] attribute, to include external files in documentation. Use the attribute in place of, or in addition to, regular doc comments and #[doc] attributes, and rustdoc will load the given file when it renders documentation for your crate.

With the following files in the same directory:

external-doc.md:

# My Awesome Type

This is the documentation for this spectacular type.

lib.rs:

#![feature(external_doc)]

#[doc(include = "external-doc.md")]
pub struct MyAwesomeType;

rustdoc will load the file external-doc.md and use it as the documentation for the MyAwesomeType struct.

When locating files, rustdoc will base paths in the src/ directory, as if they were alongside the lib.rs for your crate. So if you want a docs/ folder to live alongside the src/ directory, start your paths with ../docs/ for rustdoc to properly find the file.

This feature was proposed in RFC #1990 and initially implemented in PR #44781.

f16c_target_feature

The tracking issue for this feature is: #44839


ffi_const

The tracking issue for this feature is: #58328


The #[ffi_const] attribute applies clang's const attribute to foreign functions declarations.

That is, #[ffi_const] functions shall have no effects except for its return value, which can only depend on the values of the function parameters, and is not affected by changes to the observable state of the program.

Applying the #[ffi_const] attribute to a function that violates these requirements is undefined behaviour.

This attribute enables Rust to perform common optimizations, like sub-expression elimination, and it can avoid emitting some calls in repeated invocations of the function with the same argument values regardless of other operations being performed in between these functions calls (as opposed to #[ffi_pure] functions).

Pitfalls

A #[ffi_const] function can only read global memory that would not affect its return value for the whole execution of the program (e.g. immutable global memory). #[ffi_const] functions are referentially-transparent and therefore more strict than #[ffi_pure] functions.

A common pitfall involves applying the #[ffi_const] attribute to a function that reads memory through pointer arguments which do not necessarily point to immutable global memory.

A #[ffi_const] function that returns unit has no effect on the abstract machine's state, and a #[ffi_const] function cannot be #[ffi_pure].

A #[ffi_const] function must not diverge, neither via a side effect (e.g. a call to abort) nor by infinite loops.

When translating C headers to Rust FFI, it is worth verifying for which targets the const attribute is enabled in those headers, and using the appropriate cfg macros in the Rust side to match those definitions. While the semantics of const are implemented identically by many C and C++ compilers, e.g., clang, GCC, ARM C/C++ compiler, IBM ILE C/C++, etc. they are not necessarily implemented in this way on all of them. It is therefore also worth verifying that the semantics of the C toolchain used to compile the binary being linked against are compatible with those of the #[ffi_const].

ffi_pure

The tracking issue for this feature is: #58329


The #[ffi_pure] attribute applies clang's pure attribute to foreign functions declarations.

That is, #[ffi_pure] functions shall have no effects except for its return value, which shall not change across two consecutive function calls with the same parameters.

Applying the #[ffi_pure] attribute to a function that violates these requirements is undefined behavior.

This attribute enables Rust to perform common optimizations, like sub-expression elimination and loop optimizations. Some common examples of pure functions are strlen or memcmp.

These optimizations are only applicable when the compiler can prove that no program state observable by the #[ffi_pure] function has changed between calls of the function, which could alter the result. See also the #[ffi_const] attribute, which provides stronger guarantees regarding the allowable behavior of a function, enabling further optimization.

Pitfalls

A #[ffi_pure] function can read global memory through the function parameters (e.g. pointers), globals, etc. #[ffi_pure] functions are not referentially-transparent, and are therefore more relaxed than #[ffi_const] functions.

However, accesing global memory through volatile or atomic reads can violate the requirement that two consecutive function calls shall return the same value.

A pure function that returns unit has no effect on the abstract machine's state.

A #[ffi_pure] function must not diverge, neither via a side effect (e.g. a call to abort) nor by infinite loops.

When translating C headers to Rust FFI, it is worth verifying for which targets the pure attribute is enabled in those headers, and using the appropriate cfg macros in the Rust side to match those definitions. While the semantics of pure are implemented identically by many C and C++ compilers, e.g., clang, GCC, ARM C/C++ compiler, IBM ILE C/C++, etc. they are not necessarily implemented in this way on all of them. It is therefore also worth verifying that the semantics of the C toolchain used to compile the binary being linked against are compatible with those of the #[ffi_pure].

ffi_returns_twice

The tracking issue for this feature is: #58314


format_args_capture

The tracking issue for this feature is: #67984


fundamental

The tracking issue for this feature is: #29635


generators

The tracking issue for this feature is: #43122


The generators feature gate in Rust allows you to define generator or coroutine literals. A generator is a "resumable function" that syntactically resembles a closure but compiles to much different semantics in the compiler itself. The primary feature of a generator is that it can be suspended during execution to be resumed at a later date. Generators use the yield keyword to "return", and then the caller can resume a generator to resume execution just after the yield keyword.

Generators are an extra-unstable feature in the compiler right now. Added in RFC 2033 they're mostly intended right now as a information/constraint gathering phase. The intent is that experimentation can happen on the nightly compiler before actual stabilization. A further RFC will be required to stabilize generators/coroutines and will likely contain at least a few small tweaks to the overall design.

A syntactical example of a generator is:

#![feature(generators, generator_trait)]

use std::ops::{Generator, GeneratorState};
use std::pin::Pin;

fn main() {
    let mut generator = || {
        yield 1;
        return "foo"
    };

    match Pin::new(&mut generator).resume(()) {
        GeneratorState::Yielded(1) => {}
        _ => panic!("unexpected value from resume"),
    }
    match Pin::new(&mut generator).resume(()) {
        GeneratorState::Complete("foo") => {}
        _ => panic!("unexpected value from resume"),
    }
}

Generators are closure-like literals which can contain a yield statement. The yield statement takes an optional expression of a value to yield out of the generator. All generator literals implement the Generator trait in the std::ops module. The Generator trait has one main method, resume, which resumes execution of the generator at the previous suspension point.

An example of the control flow of generators is that the following example prints all numbers in order:

#![feature(generators, generator_trait)]

use std::ops::Generator;
use std::pin::Pin;

fn main() {
    let mut generator = || {
        println!("2");
        yield;
        println!("4");
    };

    println!("1");
    Pin::new(&mut generator).resume(());
    println!("3");
    Pin::new(&mut generator).resume(());
    println!("5");
}

At this time the main intended use case of generators is an implementation primitive for async/await syntax, but generators will likely be extended to ergonomic implementations of iterators and other primitives in the future. Feedback on the design and usage is always appreciated!

The Generator trait

The Generator trait in std::ops currently looks like:


#![allow(unused)]
fn main() {
#![feature(arbitrary_self_types, generator_trait)]
use std::ops::GeneratorState;
use std::pin::Pin;

pub trait Generator<R = ()> {
    type Yield;
    type Return;
    fn resume(self: Pin<&mut Self>, resume: R) -> GeneratorState<Self::Yield, Self::Return>;
}
}

The Generator::Yield type is the type of values that can be yielded with the yield statement. The Generator::Return type is the returned type of the generator. This is typically the last expression in a generator's definition or any value passed to return in a generator. The resume function is the entry point for executing the Generator itself.

The return value of resume, GeneratorState, looks like:


#![allow(unused)]
fn main() {
pub enum GeneratorState<Y, R> {
    Yielded(Y),
    Complete(R),
}
}

The Yielded variant indicates that the generator can later be resumed. This corresponds to a yield point in a generator. The Complete variant indicates that the generator is complete and cannot be resumed again. Calling resume after a generator has returned Complete will likely result in a panic of the program.

Closure-like semantics

The closure-like syntax for generators alludes to the fact that they also have closure-like semantics. Namely:

  • When created, a generator executes no code. A closure literal does not actually execute any of the closure's code on construction, and similarly a generator literal does not execute any code inside the generator when constructed.

  • Generators can capture outer variables by reference or by move, and this can be tweaked with the move keyword at the beginning of the closure. Like closures all generators will have an implicit environment which is inferred by the compiler. Outer variables can be moved into a generator for use as the generator progresses.

  • Generator literals produce a value with a unique type which implements the std::ops::Generator trait. This allows actual execution of the generator through the Generator::resume method as well as also naming it in return types and such.

  • Traits like Send and Sync are automatically implemented for a Generator depending on the captured variables of the environment. Unlike closures, generators also depend on variables live across suspension points. This means that although the ambient environment may be Send or Sync, the generator itself may not be due to internal variables live across yield points being not-Send or not-Sync. Note that generators do not implement traits like Copy or Clone automatically.

  • Whenever a generator is dropped it will drop all captured environment variables.

Generators as state machines

In the compiler, generators are currently compiled as state machines. Each yield expression will correspond to a different state that stores all live variables over that suspension point. Resumption of a generator will dispatch on the current state and then execute internally until a yield is reached, at which point all state is saved off in the generator and a value is returned.

Let's take a look at an example to see what's going on here:

#![feature(generators, generator_trait)]

use std::ops::Generator;
use std::pin::Pin;

fn main() {
    let ret = "foo";
    let mut generator = move || {
        yield 1;
        return ret
    };

    Pin::new(&mut generator).resume(());
    Pin::new(&mut generator).resume(());
}

This generator literal will compile down to something similar to:

#![feature(arbitrary_self_types, generators, generator_trait)]

use std::ops::{Generator, GeneratorState};
use std::pin::Pin;

fn main() {
    let ret = "foo";
    let mut generator = {
        enum __Generator {
            Start(&'static str),
            Yield1(&'static str),
            Done,
        }

        impl Generator for __Generator {
            type Yield = i32;
            type Return = &'static str;

            fn resume(mut self: Pin<&mut Self>, resume: ()) -> GeneratorState<i32, &'static str> {
                use std::mem;
                match mem::replace(&mut *self, __Generator::Done) {
                    __Generator::Start(s) => {
                        *self = __Generator::Yield1(s);
                        GeneratorState::Yielded(1)
                    }

                    __Generator::Yield1(s) => {
                        *self = __Generator::Done;
                        GeneratorState::Complete(s)
                    }

                    __Generator::Done => {
                        panic!("generator resumed after completion")
                    }
                }
            }
        }

        __Generator::Start(ret)
    };

    Pin::new(&mut generator).resume(());
    Pin::new(&mut generator).resume(());
}

Notably here we can see that the compiler is generating a fresh type, __Generator in this case. This type has a number of states (represented here as an enum) corresponding to each of the conceptual states of the generator. At the beginning we're closing over our outer variable foo and then that variable is also live over the yield point, so it's stored in both states.

When the generator starts it'll immediately yield 1, but it saves off its state just before it does so indicating that it has reached the yield point. Upon resuming again we'll execute the return ret which returns the Complete state.

Here we can also note that the Done state, if resumed, panics immediately as it's invalid to resume a completed generator. It's also worth noting that this is just a rough desugaring, not a normative specification for what the compiler does.

generic_associated_types

The tracking issue for this feature is: #44265


half_open_range_patterns

The tracking issue for this feature is: #67264


hexagon_target_feature

The tracking issue for this feature is: #44839


if_let_guard

The tracking issue for this feature is: #51114


impl_trait_in_bindings

The tracking issue for this feature is: #63065


The impl_trait_in_bindings feature gate lets you use impl Trait syntax in let, static, and const bindings.

A simple example is:

#![feature(impl_trait_in_bindings)]

use std::fmt::Debug;

fn main() {
    let a: impl Debug + Clone = 42;
    let b = a.clone();
    println!("{:?}", b); // prints `42`
}

Note however that because the types of a and b are opaque in the above example, calling inherent methods or methods outside of the specified traits (e.g., a.abs() or b.abs()) is not allowed, and yields an error.

in_band_lifetimes

The tracking issue for this feature is: #44524


infer_static_outlives_requirements

The tracking issue for this feature is: #54185


The infer_static_outlives_requirements feature indicates that certain 'static outlives requirements can be inferred by the compiler rather than stating them explicitly.

Note: It is an accompanying feature to infer_outlives_requirements, which must be enabled to infer outlives requirements.

For example, currently generic struct definitions that contain references, require where-clauses of the form T: 'static. By using this feature the outlives predicates will be inferred, although they may still be written explicitly.

struct Foo<U> where U: 'static { // <-- currently required
    bar: Bar<U>
}
struct Bar<T: 'static> {
    x: T,
}

Examples:

#![feature(infer_outlives_requirements)]
#![feature(infer_static_outlives_requirements)]

#[rustc_outlives]
// Implicitly infer U: 'static
struct Foo<U> {
    bar: Bar<U>
}
struct Bar<T: 'static> {
    x: T,
}

inline_const

The tracking issue for this feature is: #76001


This feature allows you to use inline constant expressions. For example, you can turn this code:

fn add_one(x: i32) -> i32 { x + 1 }
const MY_COMPUTATION: i32 = 1 + 2 * 3 / 4;

fn main() {
    let x = add_one(MY_COMPUTATION);
}

into this code:

#![feature(inline_const)]

fn add_one(x: i32) -> i32 { x + 1 }
fn main() {
    let x = add_one(const { 1 + 2 * 3 / 4 });
}

You can also use inline constant expressions in patterns:


#![allow(unused)]
#![feature(inline_const)]

fn main() {
const fn one() -> i32 { 1 }

let some_int = 3;
match some_int {
    const { 1 + 2 } => println!("Matched 1 + 2"),
    const { one() } => println!("Matched const fn returning 1"),
    _ => println!("Didn't match anything :("),
}
}

intrinsics

The tracking issue for this feature is: None.

Intrinsics are never intended to be stable directly, but intrinsics are often exported in some sort of stable manner. Prefer using the stable interfaces to the intrinsic directly when you can.


These are imported as if they were FFI functions, with the special rust-intrinsic ABI. For example, if one was in a freestanding context, but wished to be able to transmute between types, and perform efficient pointer arithmetic, one would import those functions via a declaration like

#![feature(intrinsics)]
fn main() {}

extern "rust-intrinsic" {
    fn transmute<T, U>(x: T) -> U;

    fn offset<T>(dst: *const T, offset: isize) -> *const T;
}

As with any other FFI functions, these are always unsafe to call.

isa_attribute

The tracking issue for this feature is: #74727


label_break_value

The tracking issue for this feature is: #48594


lang_items

The tracking issue for this feature is: None.


The rustc compiler has certain pluggable operations, that is, functionality that isn't hard-coded into the language, but is implemented in libraries, with a special marker to tell the compiler it exists. The marker is the attribute #[lang = "..."] and there are various different values of ..., i.e. various different 'lang items'.

For example, Box pointers require two lang items, one for allocation and one for deallocation. A freestanding program that uses the Box sugar for dynamic allocations via malloc and free:

#![feature(lang_items, box_syntax, start, libc, core_intrinsics)]
#![no_std]
use core::intrinsics;
use core::panic::PanicInfo;

extern crate libc;

#[lang = "owned_box"]
pub struct Box<T>(*mut T);

#[lang = "exchange_malloc"]
unsafe fn allocate(size: usize, _align: usize) -> *mut u8 {
    let p = libc::malloc(size as libc::size_t) as *mut u8;

    // Check if `malloc` failed:
    if p as usize == 0 {
        intrinsics::abort();
    }

    p
}

#[lang = "box_free"]
unsafe fn box_free<T: ?Sized>(ptr: *mut T) {
    libc::free(ptr as *mut libc::c_void)
}

#[start]
fn main(_argc: isize, _argv: *const *const u8) -> isize {
    let _x = box 1;

    0
}

#[lang = "eh_personality"] extern fn rust_eh_personality() {}
#[lang = "panic_impl"] extern fn rust_begin_panic(info: &PanicInfo) -> ! { unsafe { intrinsics::abort() } }
#[no_mangle] pub extern fn rust_eh_register_frames () {}
#[no_mangle] pub extern fn rust_eh_unregister_frames () {}

Note the use of abort: the exchange_malloc lang item is assumed to return a valid pointer, and so needs to do the check internally.

Other features provided by lang items include:

  • overloadable operators via traits: the traits corresponding to the ==, <, dereferencing (*) and + (etc.) operators are all marked with lang items; those specific four are eq, ord, deref, and add respectively.
  • stack unwinding and general failure; the eh_personality, panic and panic_bounds_checks lang items.
  • the traits in std::marker used to indicate types of various kinds; lang items send, sync and copy.
  • the marker types and variance indicators found in std::marker; lang items covariant_type, contravariant_lifetime, etc.

Lang items are loaded lazily by the compiler; e.g. if one never uses Box then there is no need to define functions for exchange_malloc and box_free. rustc will emit an error when an item is needed but not found in the current crate or any that it depends on.

Most lang items are defined by libcore, but if you're trying to build an executable without the standard library, you'll run into the need for lang items. The rest of this page focuses on this use-case, even though lang items are a bit broader than that.

Using libc

In order to build a #[no_std] executable we will need libc as a dependency. We can specify this using our Cargo.toml file:

[dependencies]
libc = { version = "0.2.14", default-features = false }

Note that the default features have been disabled. This is a critical step - the default features of libc include the standard library and so must be disabled.

Writing an executable without stdlib

Controlling the entry point is possible in two ways: the #[start] attribute, or overriding the default shim for the C main function with your own.

The function marked #[start] is passed the command line parameters in the same format as C:

#![feature(lang_items, core_intrinsics)]
#![feature(start)]
#![no_std]
use core::intrinsics;
use core::panic::PanicInfo;

// Pull in the system libc library for what crt0.o likely requires.
extern crate libc;

// Entry point for this program.
#[start]
fn start(_argc: isize, _argv: *const *const u8) -> isize {
    0
}

// These functions are used by the compiler, but not
// for a bare-bones hello world. These are normally
// provided by libstd.
#[lang = "eh_personality"]
#[no_mangle]
pub extern fn rust_eh_personality() {
}

#[lang = "panic_impl"]
#[no_mangle]
pub extern fn rust_begin_panic(info: &PanicInfo) -> ! {
    unsafe { intrinsics::abort() }
}

To override the compiler-inserted main shim, one has to disable it with #![no_main] and then create the appropriate symbol with the correct ABI and the correct name, which requires overriding the compiler's name mangling too:

#![feature(lang_items, core_intrinsics)]
#![feature(start)]
#![no_std]
#![no_main]
use core::intrinsics;
use core::panic::PanicInfo;

// Pull in the system libc library for what crt0.o likely requires.
extern crate libc;

// Entry point for this program.
#[no_mangle] // ensure that this symbol is called `main` in the output
pub extern fn main(_argc: i32, _argv: *const *const u8) -> i32 {
    0
}

// These functions are used by the compiler, but not
// for a bare-bones hello world. These are normally
// provided by libstd.
#[lang = "eh_personality"]
#[no_mangle]
pub extern fn rust_eh_personality() {
}

#[lang = "panic_impl"]
#[no_mangle]
pub extern fn rust_begin_panic(info: &PanicInfo) -> ! {
    unsafe { intrinsics::abort() }
}

In many cases, you may need to manually link to the compiler_builtins crate when building a no_std binary. You may observe this via linker error messages such as "undefined reference to `__rust_probestack'".

More about the language items

The compiler currently makes a few assumptions about symbols which are available in the executable to call. Normally these functions are provided by the standard library, but without it you must define your own. These symbols are called "language items", and they each have an internal name, and then a signature that an implementation must conform to.

The first of these functions, rust_eh_personality, is used by the failure mechanisms of the compiler. This is often mapped to GCC's personality function (see the libstd implementation for more information), but crates which do not trigger a panic can be assured that this function is never called. The language item's name is eh_personality.

The second function, rust_begin_panic, is also used by the failure mechanisms of the compiler. When a panic happens, this controls the message that's displayed on the screen. While the language item's name is panic_impl, the symbol name is rust_begin_panic.

Finally, a eh_catch_typeinfo static is needed for certain targets which implement Rust panics on top of C++ exceptions.

List of all language items

This is a list of all language items in Rust along with where they are located in the source code.

  • Primitives
    • i8: libcore/num/mod.rs
    • i16: libcore/num/mod.rs
    • i32: libcore/num/mod.rs
    • i64: libcore/num/mod.rs
    • i128: libcore/num/mod.rs
    • isize: libcore/num/mod.rs
    • u8: libcore/num/mod.rs
    • u16: libcore/num/mod.rs
    • u32: libcore/num/mod.rs
    • u64: libcore/num/mod.rs
    • u128: libcore/num/mod.rs
    • usize: libcore/num/mod.rs
    • f32: libstd/f32.rs
    • f64: libstd/f64.rs
    • char: libcore/char.rs
    • slice: liballoc/slice.rs
    • str: liballoc/str.rs
    • const_ptr: libcore/ptr.rs
    • mut_ptr: libcore/ptr.rs
    • unsafe_cell: libcore/cell.rs
  • Runtime
    • start: libstd/rt.rs
    • eh_personality: libpanic_unwind/emcc.rs (EMCC)
    • eh_personality: libpanic_unwind/gcc.rs (GNU)
    • eh_personality: libpanic_unwind/seh.rs (SEH)
    • eh_catch_typeinfo: libpanic_unwind/emcc.rs (EMCC)
    • panic: libcore/panicking.rs
    • panic_bounds_check: libcore/panicking.rs
    • panic_impl: libcore/panicking.rs
    • panic_impl: libstd/panicking.rs
  • Allocations
    • owned_box: liballoc/boxed.rs
    • exchange_malloc: liballoc/heap.rs
    • box_free: liballoc/heap.rs
  • Operands
    • not: libcore/ops/bit.rs
    • bitand: libcore/ops/bit.rs
    • bitor: libcore/ops/bit.rs
    • bitxor: libcore/ops/bit.rs
    • shl: libcore/ops/bit.rs
    • shr: libcore/ops/bit.rs
    • bitand_assign: libcore/ops/bit.rs
    • bitor_assign: libcore/ops/bit.rs
    • bitxor_assign: libcore/ops/bit.rs
    • shl_assign: libcore/ops/bit.rs
    • shr_assign: libcore/ops/bit.rs
    • deref: libcore/ops/deref.rs
    • deref_mut: libcore/ops/deref.rs
    • index: libcore/ops/index.rs
    • index_mut: libcore/ops/index.rs
    • add: libcore/ops/arith.rs
    • sub: libcore/ops/arith.rs
    • mul: libcore/ops/arith.rs
    • div: libcore/ops/arith.rs
    • rem: libcore/ops/arith.rs
    • neg: libcore/ops/arith.rs
    • add_assign: libcore/ops/arith.rs
    • sub_assign: libcore/ops/arith.rs
    • mul_assign: libcore/ops/arith.rs
    • div_assign: libcore/ops/arith.rs
    • rem_assign: libcore/ops/arith.rs
    • eq: libcore/cmp.rs
    • ord: libcore/cmp.rs
  • Functions
    • fn: libcore/ops/function.rs
    • fn_mut: libcore/ops/function.rs
    • fn_once: libcore/ops/function.rs
    • generator_state: libcore/ops/generator.rs
    • generator: libcore/ops/generator.rs
  • Other
    • coerce_unsized: libcore/ops/unsize.rs
    • drop: libcore/ops/drop.rs
    • drop_in_place: libcore/ptr.rs
    • clone: libcore/clone.rs
    • copy: libcore/marker.rs
    • send: libcore/marker.rs
    • sized: libcore/marker.rs
    • unsize: libcore/marker.rs
    • sync: libcore/marker.rs
    • phantom_data: libcore/marker.rs
    • discriminant_kind: libcore/marker.rs
    • freeze: libcore/marker.rs
    • debug_trait: libcore/fmt/mod.rs
    • non_zero: libcore/nonzero.rs
    • arc: liballoc/sync.rs
    • rc: liballoc/rc.rs

lazy_normalization_consts

The tracking issue for this feature is: #72219


let_chains

The tracking issue for this feature is: #53667


link_args

The tracking issue for this feature is: #29596


You can tell rustc how to customize linking, and that is via the link_args attribute. This attribute is applied to extern blocks and specifies raw flags which need to get passed to the linker when producing an artifact. An example usage would be:

#![feature(link_args)]

#[link_args = "-foo -bar -baz"]
extern {}
fn main() {}

Note that this feature is currently hidden behind the feature(link_args) gate because this is not a sanctioned way of performing linking. Right now rustc shells out to the system linker (gcc on most systems, link.exe on MSVC), so it makes sense to provide extra command line arguments, but this will not always be the case. In the future rustc may use LLVM directly to link native libraries, in which case link_args will have no meaning. You can achieve the same effect as the link_args attribute with the -C link-args argument to rustc.

It is highly recommended to not use this attribute, and rather use the more formal #[link(...)] attribute on extern blocks instead.

link_cfg

This feature is internal to the Rust compiler and is not intended for general use.


link_llvm_intrinsics

The tracking issue for this feature is: #29602


linkage

The tracking issue for this feature is: #29603


lint_reasons

The tracking issue for this feature is: #54503


main

The tracking issue for this feature is: #29634


marker_trait_attr

The tracking issue for this feature is: #29864


Normally, Rust keeps you from adding trait implementations that could overlap with each other, as it would be ambiguous which to use. This feature, however, carves out an exception to that rule: a trait can opt-in to having overlapping implementations, at the cost that those implementations are not allowed to override anything (and thus the trait itself cannot have any associated items, as they're pointless when they'd need to do the same thing for every type anyway).


#![allow(unused)]
#![feature(marker_trait_attr)]

fn main() {
#[marker] trait CheapToClone: Clone {}

impl<T: Copy> CheapToClone for T {}

// These could potentially overlap with the blanket implementation above,
// so are only allowed because CheapToClone is a marker trait.
impl<T: CheapToClone, U: CheapToClone> CheapToClone for (T, U) {}
impl<T: CheapToClone> CheapToClone for std::ops::Range<T> {}

fn cheap_clone<T: CheapToClone>(t: T) -> T {
    t.clone()
}
}

This is expected to replace the unstable overlapping_marker_traits feature, which applied to all empty traits (without needing an opt-in).

member_constraints

The tracking issue for this feature is: #61997


The member_constraints feature gate lets you use impl Trait syntax with multiple unrelated lifetime parameters.

A simple example is:

#![feature(member_constraints)]

trait Trait<'a, 'b> { }
impl<T> Trait<'_, '_> for T {}

fn foo<'a, 'b>(x: &'a u32, y: &'b u32) -> impl Trait<'a, 'b> {
  (x, y)
}

fn main() { }

Without the member_constraints feature gate, the above example is an error because both 'a and 'b appear in the impl Trait bounds, but neither outlives the other.

min_const_generics

The tracking issue for this feature is: #74878


min_specialization

The tracking issue for this feature is: #31844


mips_target_feature

The tracking issue for this feature is: #44839


movbe_target_feature

The tracking issue for this feature is: #44839


naked_functions

The tracking issue for this feature is: #32408


needs_panic_runtime

The tracking issue for this feature is: #32837


negative_impls

The tracking issue for this feature is #68318.


With the feature gate negative_impls, you can write negative impls as well as positive ones:


#![allow(unused)]
#![feature(negative_impls)]
fn main() {
trait DerefMut { }
impl<T: ?Sized> !DerefMut for &T { }
}

Negative impls indicate a semver guarantee that the given trait will not be implemented for the given types. Negative impls play an additional purpose for auto traits, described below.

Negative impls have the following characteristics:

  • They do not have any items.
  • They must obey the orphan rules as if they were a positive impl.
  • They cannot "overlap" with any positive impls.

Semver interaction

It is a breaking change to remove a negative impl. Negative impls are a commitment not to implement the given trait for the named types.

Orphan and overlap rules

Negative impls must obey the same orphan rules as a positive impl. This implies you cannot add a negative impl for types defined in upstream crates and so forth.

Similarly, negative impls cannot overlap with positive impls, again using the same "overlap" check that we ordinarily use to determine if two impls overlap. (Note that positive impls typically cannot overlap with one another either, except as permitted by specialization.)

Interaction with auto traits

Declaring a negative impl impl !SomeAutoTrait for SomeType for an auto-trait serves two purposes:

  • as with any trait, it declares that SomeType will never implement SomeAutoTrait;
  • it disables the automatic SomeType: SomeAutoTrait impl that would otherwise have been generated.

Note that, at present, there is no way to indicate that a given type does not implement an auto trait but that it may do so in the future. For ordinary types, this is done by simply not declaring any impl at all, but that is not an option for auto traits. A workaround is that one could embed a marker type as one of the fields, where the marker type is !AutoTrait.

Immediate uses

Negative impls are used to declare that &T: !DerefMut and &mut T: !Clone, as required to fix the soundness of Pin described in #66544.

This serves two purposes:

  • For proving the correctness of unsafe code, we can use that impl as evidence that no DerefMut or Clone impl exists.
  • It prevents downstream crates from creating such impls.

never_type

The tracking issue for this feature is: #35121


never_type_fallback

The tracking issue for this feature is: #65992


nll

The tracking issue for this feature is: #43234


no_core

The tracking issue for this feature is: #29639


no_niche

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


no_sanitize

The tracking issue for this feature is: #39699


The no_sanitize attribute can be used to selectively disable sanitizer instrumentation in an annotated function. This might be useful to: avoid instrumentation overhead in a performance critical function, or avoid instrumenting code that contains constructs unsupported by given sanitizer.

The precise effect of this annotation depends on particular sanitizer in use. For example, with no_sanitize(thread), the thread sanitizer will no longer instrument non-atomic store / load operations, but it will instrument atomic operations to avoid reporting false positives and provide meaning full stack traces.

Examples


#![allow(unused)]
#![feature(no_sanitize)]

fn main() {
#[no_sanitize(address)]
fn foo() {
  // ...
}
}

non_ascii_idents

The tracking issue for this feature is: #55467


The non_ascii_idents feature adds support for non-ASCII identifiers.

Examples


#![allow(unused)]
#![feature(non_ascii_idents)]

fn main() {
const ε: f64 = 0.00001f64;
const Π: f64 = 3.14f64;
}

Changes to the language reference

Lexer:
IDENTIFIER :
      XID_start XID_continue*
   | _ XID_continue+

An identifier is any nonempty Unicode string of the following form:

Either

Or

  • The first character is _
  • The identifier is more than one character, _ alone is not an identifier
  • The remaining characters have property XID_continue

that does not occur in the set of strict keywords.

Note: XID_start and XID_continue as character properties cover the character ranges used to form the more familiar C and Java language-family identifiers.

object_safe_for_dispatch

The tracking issue for this feature is: #43561


omit_gdb_pretty_printer_section

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


optimize_attribute

The tracking issue for this feature is: #54882


optin_builtin_traits

The tracking issue for this feature is #13231


The optin_builtin_traits feature gate allows you to define auto traits.

Auto traits, like Send or Sync in the standard library, are marker traits that are automatically implemented for every type, unless the type, or a type it contains, has explicitly opted out via a negative impl. (Negative impls are separately controlled by the negative_impls feature.)

impl !Trait for Type

Example:

#![feature(negative_impls)]
#![feature(optin_builtin_traits)]

auto trait Valid {}

struct True;
struct False;

impl !Valid for False {}

struct MaybeValid<T>(T);

fn must_be_valid<T: Valid>(_t: T) { }

fn main() {
    // works
    must_be_valid( MaybeValid(True) );
                
    // compiler error - trait bound not satisfied
    // must_be_valid( MaybeValid(False) );
}

Automatic trait implementations

When a type is declared as an auto trait, we will automatically create impls for every struct/enum/union, unless an explicit impl is provided. These automatic impls contain a where clause for each field of the form T: AutoTrait, where T is the type of the field and AutoTrait is the auto trait in question. As an example, consider the struct List and the auto trait Send:


#![allow(unused)]
fn main() {
struct List<T> {
  data: T,
  next: Option<Box<List<T>>>,
}
}

Presuming that there is no explicit impl of Send for List, the compiler will supply an automatic impl of the form:


#![allow(unused)]
fn main() {
struct List<T> {
  data: T,
  next: Option<Box<List<T>>>,
}

unsafe impl<T> Send for List<T>
where
  T: Send, // from the field `data`
  Option<Box<List<T>>>: Send, // from the field `next`
{ }
}

Explicit impls may be either positive or negative. They take the form:

impl<...> AutoTrait for StructName<..> { }
impl<...> !AutoTrait for StructName<..> { }

Coinduction: Auto traits permit cyclic matching

Unlike ordinary trait matching, auto traits are coinductive. This means, in short, that cycles which occur in trait matching are considered ok. As an example, consider the recursive struct List introduced in the previous section. In attempting to determine whether List: Send, we would wind up in a cycle: to apply the impl, we must show that Option<Box<List>>: Send, which will in turn require Box<List>: Send and then finally List: Send again. Under ordinary trait matching, this cycle would be an error, but for an auto trait it is considered a successful match.

Items

Auto traits cannot have any trait items, such as methods or associated types. This ensures that we can generate default implementations.

Supertraits

Auto traits cannot have supertraits. This is for soundness reasons, as the interaction of coinduction with implied bounds is difficult to reconcile.

or_patterns

The tracking issue for this feature is: #54883


The or_pattern language feature allows | to be arbitrarily nested within a pattern, for example, Some(A(0) | B(1 | 2)) becomes a valid pattern.

Examples

#![feature(or_patterns)]

pub enum Foo {
    Bar,
    Baz,
    Quux,
}

pub fn example(maybe_foo: Option<Foo>) {
    match maybe_foo {
        Some(Foo::Bar | Foo::Baz) => {
            println!("The value contained `Bar` or `Baz`");
        }
        Some(_) => {
            println!("The value did not contain `Bar` or `Baz`");
        }
        None => {
            println!("The value was `None`");
        }
    }
}

panic_runtime

The tracking issue for this feature is: #32837


platform_intrinsics

The tracking issue for this feature is: #27731


plugin

The tracking issue for this feature is: #29597

This feature is part of "compiler plugins." It will often be used with the plugin_registrar and rustc_private features.


rustc can load compiler plugins, which are user-provided libraries that extend the compiler's behavior with new lint checks, etc.

A plugin is a dynamic library crate with a designated registrar function that registers extensions with rustc. Other crates can load these extensions using the crate attribute #![plugin(...)]. See the rustc_driver::plugin documentation for more about the mechanics of defining and loading a plugin.

In the vast majority of cases, a plugin should only be used through #![plugin] and not through an extern crate item. Linking a plugin would pull in all of librustc_ast and librustc as dependencies of your crate. This is generally unwanted unless you are building another plugin.

The usual practice is to put compiler plugins in their own crate, separate from any macro_rules! macros or ordinary Rust code meant to be used by consumers of a library.

Lint plugins

Plugins can extend Rust's lint infrastructure with additional checks for code style, safety, etc. Now let's write a plugin lint-plugin-test.rs that warns about any item named lintme.

#![feature(plugin_registrar)]
#![feature(box_syntax, rustc_private)]

extern crate rustc_ast;

// Load rustc as a plugin to get macros
extern crate rustc_driver;
#[macro_use]
extern crate rustc_lint;
#[macro_use]
extern crate rustc_session;

use rustc_driver::plugin::Registry;
use rustc_lint::{EarlyContext, EarlyLintPass, LintArray, LintContext, LintPass};
use rustc_ast::ast;
declare_lint!(TEST_LINT, Warn, "Warn about items named 'lintme'");

declare_lint_pass!(Pass => [TEST_LINT]);

impl EarlyLintPass for Pass {
    fn check_item(&mut self, cx: &EarlyContext, it: &ast::Item) {
        if it.ident.name.as_str() == "lintme" {
            cx.lint(TEST_LINT, |lint| {
                lint.build("item is named 'lintme'").set_span(it.span).emit()
            });
        }
    }
}

#[plugin_registrar]
pub fn plugin_registrar(reg: &mut Registry) {
    reg.lint_store.register_lints(&[&TEST_LINT]);
    reg.lint_store.register_early_pass(|| box Pass);
}

Then code like

#![feature(plugin)]
#![plugin(lint_plugin_test)]

fn lintme() { }

will produce a compiler warning:

foo.rs:4:1: 4:16 warning: item is named 'lintme', #[warn(test_lint)] on by default
foo.rs:4 fn lintme() { }
         ^~~~~~~~~~~~~~~

The components of a lint plugin are:

  • one or more declare_lint! invocations, which define static Lint structs;

  • a struct holding any state needed by the lint pass (here, none);

  • a LintPass implementation defining how to check each syntax element. A single LintPass may call span_lint for several different Lints, but should register them all through the get_lints method.

Lint passes are syntax traversals, but they run at a late stage of compilation where type information is available. rustc's built-in lints mostly use the same infrastructure as lint plugins, and provide examples of how to access type information.

Lints defined by plugins are controlled by the usual attributes and compiler flags, e.g. #[allow(test_lint)] or -A test-lint. These identifiers are derived from the first argument to declare_lint!, with appropriate case and punctuation conversion.

You can run rustc -W help foo.rs to see a list of lints known to rustc, including those provided by plugins loaded by foo.rs.

plugin_registrar

The tracking issue for this feature is: #29597

This feature is part of "compiler plugins." It will often be used with the plugin and rustc_private features as well. For more details, see their docs.


powerpc_target_feature

The tracking issue for this feature is: #44839


precise_pointer_size_matching

The tracking issue for this feature is: #56354


prelude_import

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


proc_macro_hygiene

The tracking issue for this feature is: #54727


profiler_runtime

The tracking issue for this feature is: #42524.


raw_dylib

The tracking issue for this feature is: #58713


raw_ref_op

The tracking issue for this feature is: #64490


register_attr

The tracking issue for this feature is: #66080


register_tool

The tracking issue for this feature is: #66079


repr_simd

The tracking issue for this feature is: #27731


repr128

The tracking issue for this feature is: #56071


The repr128 feature adds support for #[repr(u128)] on enums.


#![allow(unused)]
#![feature(repr128)]

fn main() {
#[repr(u128)]
enum Foo {
    Bar(u64),
}
}

riscv_target_feature

The tracking issue for this feature is: #44839


rtm_target_feature

The tracking issue for this feature is: #44839


rustc_attrs

This feature has no tracking issue, and is therefore internal to the compiler, not being intended for general use.

Note: rustc_attrs enables many rustc-internal attributes and this page only discuss a few of them.


The rustc_attrs feature allows debugging rustc type layouts by using #[rustc_layout(...)] to debug layout at compile time (it even works with cargo check) as an alternative to rustc -Z print-type-sizes that is way more verbose.

Options provided by #[rustc_layout(...)] are debug, size, align, abi. Note that it only works on sized types without generics.

Examples

#![feature(rustc_attrs)]

#[rustc_layout(abi, size)]
pub enum X {
    Y(u8, u8, u8),
    Z(isize),
}

When that is compiled, the compiler will error with something like

error: abi: Aggregate { sized: true }
 --> src/lib.rs:4:1
  |
4 | / pub enum T {
5 | |     Y(u8, u8, u8),
6 | |     Z(isize),
7 | | }
  | |_^

error: size: Size { raw: 16 }
 --> src/lib.rs:4:1
  |
4 | / pub enum T {
5 | |     Y(u8, u8, u8),
6 | |     Z(isize),
7 | | }
  | |_^

error: aborting due to 2 previous errors

rustc_private

The tracking issue for this feature is: #27812


simd_ffi

The tracking issue for this feature is: #27731


specialization

The tracking issue for this feature is: #31844


sse4a_target_feature

The tracking issue for this feature is: #44839


staged_api

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


start

The tracking issue for this feature is: #29633


static_nobundle

The tracking issue for this feature is: #37403


stmt_expr_attributes

The tracking issue for this feature is: #15701


structural_match

The tracking issue for this feature is: #31434


target_feature_11

The tracking issue for this feature is: #69098


tbm_target_feature

The tracking issue for this feature is: #44839


test_2018_feature

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


thread_local

The tracking issue for this feature is: #29594


trait_alias

The tracking issue for this feature is: #41517


The trait_alias feature adds support for trait aliases. These allow aliases to be created for one or more traits (currently just a single regular trait plus any number of auto-traits), and used wherever traits would normally be used as either bounds or trait objects.

#![feature(trait_alias)]

trait Foo = std::fmt::Debug + Send;
trait Bar = Foo + Sync;

// Use trait alias as bound on type parameter.
fn foo<T: Foo>(v: &T) {
    println!("{:?}", v);
}

pub fn main() {
    foo(&1);

    // Use trait alias for trait objects.
    let a: &Bar = &123;
    println!("{:?}", a);
    let b = Box::new(456) as Box<dyn Foo>;
    println!("{:?}", b);
}

transparent_unions

The tracking issue for this feature is #60405


The transparent_unions feature allows you mark unions as #[repr(transparent)]. A union may be #[repr(transparent)] in exactly the same conditions in which a struct may be #[repr(transparent)] (generally, this means the union must have exactly one non-zero-sized field). Some concrete illustrations follow.


#![allow(unused)]
#![feature(transparent_unions)]

fn main() {
// This union has the same representation as `f32`.
#[repr(transparent)]
union SingleFieldUnion {
    field: f32,
}

// This union has the same representation as `usize`.
#[repr(transparent)]
union MultiFieldUnion {
    field: usize,
    nothing: (),
}
}

For consistency with transparent structs, unions must have exactly one non-zero-sized field. If all fields are zero-sized, the union must not be #[repr(transparent)]:


#![allow(unused)]
#![feature(transparent_unions)]

fn main() {
// This (non-transparent) union is already valid in stable Rust:
pub union GoodUnion {
    pub nothing: (),
}

// Error: transparent union needs exactly one non-zero-sized field, but has 0
// #[repr(transparent)]
// pub union BadUnion {
//     pub nothing: (),
// }
}

The one exception is if the union is generic over T and has a field of type T, it may be #[repr(transparent)] even if T is a zero-sized type:


#![allow(unused)]
#![feature(transparent_unions)]

fn main() {
// This union has the same representation as `T`.
#[repr(transparent)]
pub union GenericUnion<T: Copy> { // Unions with non-`Copy` fields are unstable.
    pub field: T,
    pub nothing: (),
}

// This is okay even though `()` is a zero-sized type.
pub const THIS_IS_OKAY: GenericUnion<()> = GenericUnion { field: () };
}

Like transarent structs, a transparent union of type U has the same layout, size, and ABI as its single non-ZST field. If it is generic over a type T, and all its fields are ZSTs except for exactly one field of type T, then it has the same layout and ABI as T (even if T is a ZST when monomorphized).

Like transparent structs, transparent unions are FFI-safe if and only if their underlying representation type is also FFI-safe.

A union may not be eligible for the same nonnull-style optimizations that a struct or enum (with the same fields) are eligible for. Adding #[repr(transparent)] to union does not change this. To give a more concrete example, it is unspecified whether size_of::<T>() is equal to size_of::<Option<T>>(), where T is a union (regardless of whether or not it is transparent). The Rust compiler is free to perform this optimization if possible, but is not required to, and different compiler versions may differ in their application of these optimizations.

trivial_bounds

The tracking issue for this feature is: #48214


try_blocks

The tracking issue for this feature is: #31436


The try_blocks feature adds support for try blocks. A try block creates a new scope one can use the ? operator in.


#![allow(unused)]
#![feature(try_blocks)]

fn main() {
use std::num::ParseIntError;

let result: Result<i32, ParseIntError> = try {
    "1".parse::<i32>()?
        + "2".parse::<i32>()?
        + "3".parse::<i32>()?
};
assert_eq!(result, Ok(6));

let result: Result<i32, ParseIntError> = try {
    "1".parse::<i32>()?
        + "foo".parse::<i32>()?
        + "3".parse::<i32>()?
};
assert!(result.is_err());
}

type_alias_impl_trait

The tracking issue for this feature is: #63063


type_ascription

The tracking issue for this feature is: #23416


unboxed_closures

The tracking issue for this feature is #29625

See Also: fn_traits


The unboxed_closures feature allows you to write functions using the "rust-call" ABI, required for implementing the Fn* family of traits. "rust-call" functions must have exactly one (non self) argument, a tuple representing the argument list.

#![feature(unboxed_closures)]

extern "rust-call" fn add_args(args: (u32, u32)) -> u32 {
    args.0 + args.1
}

fn main() {}

unsafe_block_in_unsafe_fn

The tracking issue for this feature is: #71668


unsized_locals

The tracking issue for this feature is: #48055


This implements RFC1909. When turned on, you can have unsized arguments and locals:

#![feature(unsized_locals)]

use std::any::Any;

fn main() {
    let x: Box<dyn Any> = Box::new(42);
    let x: dyn Any = *x;
    //  ^ unsized local variable
    //               ^^ unsized temporary
    foo(x);
}

fn foo(_: dyn Any) {}
//     ^^^^^^ unsized argument

The RFC still forbids the following unsized expressions:

#![feature(unsized_locals)]

use std::any::Any;

struct MyStruct<T: ?Sized> {
    content: T,
}

struct MyTupleStruct<T: ?Sized>(T);

fn answer() -> Box<dyn Any> {
    Box::new(42)
}

fn main() {
    // You CANNOT have unsized statics.
    static X: dyn Any = *answer();  // ERROR
    const Y: dyn Any = *answer();  // ERROR

    // You CANNOT have struct initialized unsized.
    MyStruct { content: *answer() };  // ERROR
    MyTupleStruct(*answer());  // ERROR
    (42, *answer());  // ERROR

    // You CANNOT have unsized return types.
    fn my_function() -> dyn Any { *answer() }  // ERROR

    // You CAN have unsized local variables...
    let mut x: dyn Any = *answer();  // OK
    // ...but you CANNOT reassign to them.
    x = *answer();  // ERROR

    // You CANNOT even initialize them separately.
    let y: dyn Any;  // OK
    y = *answer();  // ERROR

    // Not mentioned in the RFC, but by-move captured variables are also Sized.
    let x: dyn Any = *answer();
    (move || {  // ERROR
        let y = x;
    })();

    // You CAN create a closure with unsized arguments,
    // but you CANNOT call it.
    // This is an implementation detail and may be changed in the future.
    let f = |x: dyn Any| {};
    f(*answer());  // ERROR
}

By-value trait objects

With this feature, you can have by-value self arguments without Self: Sized bounds.

#![feature(unsized_locals)]

trait Foo {
    fn foo(self) {}
}

impl<T: ?Sized> Foo for T {}

fn main() {
    let slice: Box<[i32]> = Box::new([1, 2, 3]);
    <[i32] as Foo>::foo(*slice);
}

And Foo will also be object-safe.

#![feature(unsized_locals)]

trait Foo {
    fn foo(self) {}
}

impl<T: ?Sized> Foo for T {}

fn main () {
    let slice: Box<dyn Foo> = Box::new([1, 2, 3]);
    // doesn't compile yet
    <dyn Foo as Foo>::foo(*slice);
}

One of the objectives of this feature is to allow Box<dyn FnOnce>.

Variable length arrays

The RFC also describes an extension to the array literal syntax: [e; dyn n]. In the syntax, n isn't necessarily a constant expression. The array is dynamically allocated on the stack and has the type of [T], instead of [T; n].

#![feature(unsized_locals)]

fn mergesort<T: Ord>(a: &mut [T]) {
    let mut tmp = [T; dyn a.len()];
    // ...
}

fn main() {
    let mut a = [3, 1, 5, 6];
    mergesort(&mut a);
    assert_eq!(a, [1, 3, 5, 6]);
}

VLAs are not implemented yet. The syntax isn't final, either. We may need an alternative syntax for Rust 2015 because, in Rust 2015, expressions like [e; dyn(1)] would be ambiguous. One possible alternative proposed in the RFC is [e; n]: if n captures one or more local variables, then it is considered as [e; dyn n].

Advisory on stack usage

It's advised not to casually use the #![feature(unsized_locals)] feature. Typical use-cases are:

  • When you need a by-value trait objects.
  • When you really need a fast allocation of small temporary arrays.

Another pitfall is repetitive allocation and temporaries. Currently the compiler simply extends the stack frame every time it encounters an unsized assignment. So for example, the code

#![feature(unsized_locals)]

fn main() {
    let x: Box<[i32]> = Box::new([1, 2, 3, 4, 5]);
    let _x = {{{{{{{{{{*x}}}}}}}}}};
}

and the code

#![feature(unsized_locals)]

fn main() {
    for _ in 0..10 {
        let x: Box<[i32]> = Box::new([1, 2, 3, 4, 5]);
        let _x = *x;
    }
}

will unnecessarily extend the stack frame.

unsized_tuple_coercion

The tracking issue for this feature is: #42877


This is a part of RFC0401. According to the RFC, there should be an implementation like this:

impl<..., T, U: ?Sized> Unsized<(..., U)> for (..., T) where T: Unsized<U> {}

This implementation is currently gated behind #[feature(unsized_tuple_coercion)] to avoid insta-stability. Therefore you can use it like this:

#![feature(unsized_tuple_coercion)]

fn main() {
    let x : ([i32; 3], [i32; 3]) = ([1, 2, 3], [4, 5, 6]);
    let y : &([i32; 3], [i32]) = &x;
    assert_eq!(y.1[0], 4);
}

untagged_unions

The tracking issue for this feature is: #55149


unwind_attributes

The tracking issue for this feature is: #58760


wasm_target_feature

The tracking issue for this feature is: #44839


Library Features

alloc_error_hook

The tracking issue for this feature is: #51245


alloc_internals

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


alloc_layout_extra

The tracking issue for this feature is: #55724


alloc_prelude

The tracking issue for this feature is: #58935


allocator_api

The tracking issue for this feature is #32838


Sometimes you want the memory for one collection to use a different allocator than the memory for another collection. In this case, replacing the global allocator is not a workable option. Instead, you need to pass in an instance of an AllocRef to each collection for which you want a custom allocator.

TBD

arc_mutate_strong_count

The tracking issue for this feature is: #71983


arc_new_cyclic

The tracking issue for this feature is: #75861


array_chunks

The tracking issue for this feature is: #74985


array_error_internals

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


array_from_ref

The tracking issue for this feature is: #77101


array_map

The tracking issue for this feature is: #75243


array_methods

The tracking issue for this feature is: #76118


array_value_iter

The tracking issue for this feature is: #65798


array_windows

The tracking issue for this feature is: #75027


asm

The tracking issue for this feature is: #72016


For extremely low-level manipulations and performance reasons, one might wish to control the CPU directly. Rust supports using inline assembly to do this via the asm! macro.

Guide-level explanation

Rust provides support for inline assembly via the asm! macro. It can be used to embed handwritten assembly in the assembly output generated by the compiler. Generally this should not be necessary, but might be where the required performance or timing cannot be otherwise achieved. Accessing low level hardware primitives, e.g. in kernel code, may also demand this functionality.

Note: the examples here are given in x86/x86-64 assembly, but other architectures are also supported.

Inline assembly is currently supported on the following architectures:

  • x86 and x86-64
  • ARM
  • AArch64
  • RISC-V
  • NVPTX
  • Hexagon
  • MIPS32r2 and MIPS64r2

Basic usage

Let us start with the simplest possible example:


#![allow(unused)]
fn main() {
#![feature(asm)]
unsafe {
    asm!("nop");
}
}

This will insert a NOP (no operation) instruction into the assembly generated by the compiler. Note that all asm! invocations have to be inside an unsafe block, as they could insert arbitrary instructions and break various invariants. The instructions to be inserted are listed in the first argument of the asm! macro as a string literal.

Inputs and outputs

Now inserting an instruction that does nothing is rather boring. Let us do something that actually acts on data:


#![allow(unused)]
fn main() {
#![feature(asm)]
let x: u64;
unsafe {
    asm!("mov {}, 5", out(reg) x);
}
assert_eq!(x, 5);
}

This will write the value 5 into the u64 variable x. You can see that the string literal we use to specify instructions is actually a template string. It is governed by the same rules as Rust format strings. The arguments that are inserted into the template however look a bit different then you may be familiar with. First we need to specify if the variable is an input or an output of the inline assembly. In this case it is an output. We declared this by writing out. We also need to specify in what kind of register the assembly expects the variable. In this case we put it in an arbitrary general purpose register by specifying reg. The compiler will choose an appropriate register to insert into the template and will read the variable from there after the inline assembly finishes executing.

Let us see another example that also uses an input:


#![allow(unused)]
fn main() {
#![feature(asm)]
let i: u64 = 3;
let o: u64;
unsafe {
    asm!(
        "mov {0}, {1}",
        "add {0}, {number}",
        out(reg) o,
        in(reg) i,
        number = const 5,
    );
}
assert_eq!(o, 8);
}

This will add 5 to the input in variable i and write the result to variable o. The particular way this assembly does this is first copying the value from i to the output, and then adding 5 to it.

The example shows a few things:

First, we can see that asm! allows multiple template string arguments; each one is treated as a separate line of assembly code, as if they were all joined together with newlines between them. This makes it easy to format assembly code.

Second, we can see that inputs are declared by writing in instead of out.

Third, one of our operands has a type we haven't seen yet, const. This tells the compiler to expand this argument to value directly inside the assembly template. This is only possible for constants and literals.

Fourth, we can see that we can specify an argument number, or name as in any format string. For inline assembly templates this is particularly useful as arguments are often used more than once. For more complex inline assembly using this facility is generally recommended, as it improves readability, and allows reordering instructions without changing the argument order.

We can further refine the above example to avoid the mov instruction:


#![allow(unused)]
fn main() {
#![feature(asm)]
let mut x: u64 = 3;
unsafe {
    asm!("add {0}, {number}", inout(reg) x, number = const 5);
}
assert_eq!(x, 8);
}

We can see that inout is used to specify an argument that is both input and output. This is different from specifying an input and output separately in that it is guaranteed to assign both to the same register.

It is also possible to specify different variables for the input and output parts of an inout operand:


#![allow(unused)]
fn main() {
#![feature(asm)]
let x: u64 = 3;
let y: u64;
unsafe {
    asm!("add {0}, {number}", inout(reg) x => y, number = const 5);
}
assert_eq!(y, 8);
}

Late output operands

The Rust compiler is conservative with its allocation of operands. It is assumed that an out can be written at any time, and can therefore not share its location with any other argument. However, to guarantee optimal performance it is important to use as few registers as possible, so they won't have to be saved and reloaded around the inline assembly block. To achieve this Rust provides a lateout specifier. This can be used on any output that is written only after all inputs have been consumed. There is also a inlateout variant of this specifier.

Here is an example where inlateout cannot be used:


#![allow(unused)]
fn main() {
#![feature(asm)]
let mut a: u64 = 4;
let b: u64 = 4;
let c: u64 = 4;
unsafe {
    asm!(
        "add {0}, {1}",
        "add {0}, {2}",
        inout(reg) a,
        in(reg) b,
        in(reg) c,
    );
}
assert_eq!(a, 12);
}

Here the compiler is free to allocate the same register for inputs b and c since it knows they have the same value. However it must allocate a separate register for a since it uses inout and not inlateout. If inlateout was used, then a and c could be allocated to the same register, in which case the first instruction to overwrite the value of c and cause the assembly code to produce the wrong result.

However the following example can use inlateout since the output is only modified after all input registers have been read:


#![allow(unused)]
fn main() {
#![feature(asm)]
let mut a: u64 = 4;
let b: u64 = 4;
unsafe {
    asm!("add {0}, {1}", inlateout(reg) a, in(reg) b);
}
assert_eq!(a, 8);
}

As you can see, this assembly fragment will still work correctly if a and b are assigned to the same register.

Explicit register operands

Some instructions require that the operands be in a specific register. Therefore, Rust inline assembly provides some more specific constraint specifiers. While reg is generally available on any architecture, these are highly architecture specific. E.g. for x86 the general purpose registers eax, ebx, ecx, edx, ebp, esi, and edi among others can be addressed by their name.


#![allow(unused)]
fn main() {
#![feature(asm)]
let cmd = 0xd1;
unsafe {
    asm!("out 0x64, eax", in("eax") cmd);
}
}

In this example we call the out instruction to output the content of the cmd variable to port 0x64. Since the out instruction only accepts eax (and its sub registers) as operand we had to use the eax constraint specifier.

Note that unlike other operand types, explicit register operands cannot be used in the template string: you can't use {} and should write the register name directly instead. Also, they must appear at the end of the operand list after all other operand types.

Consider this example which uses the x86 mul instruction:


#![allow(unused)]
fn main() {
#![feature(asm)]
fn mul(a: u64, b: u64) -> u128 {
    let lo: u64;
    let hi: u64;

    unsafe {
        asm!(
            // The x86 mul instruction takes rax as an implicit input and writes
            // the 128-bit result of the multiplication to rax:rdx.
            "mul {}",
            in(reg) a,
            inlateout("rax") b => lo,
            lateout("rdx") hi
        );
    }

    ((hi as u128) << 64) + lo as u128
}
}

This uses the mul instruction to multiply two 64-bit inputs with a 128-bit result. The only explicit operand is a register, that we fill from the variable a. The second operand is implicit, and must be the rax register, which we fill from the variable b. The lower 64 bits of the result are stored in rax from which we fill the variable lo. The higher 64 bits are stored in rdx from which we fill the variable hi.

Clobbered registers

In many cases inline assembly will modify state that is not needed as an output. Usually this is either because we have to use a scratch register in the assembly, or instructions modify state that we don't need to further examine. This state is generally referred to as being "clobbered". We need to tell the compiler about this since it may need to save and restore this state around the inline assembly block.


#![allow(unused)]
fn main() {
#![feature(asm)]
let ebx: u32;
let ecx: u32;

unsafe {
    asm!(
        "cpuid",
        // EAX 4 selects the "Deterministic Cache Parameters" CPUID leaf
        inout("eax") 4 => _,
        // ECX 0 selects the L0 cache information.
        inout("ecx") 0 => ecx,
        lateout("ebx") ebx,
        lateout("edx") _,
    );
}

println!(
    "L1 Cache: {}",
    ((ebx >> 22) + 1) * (((ebx >> 12) & 0x3ff) + 1) * ((ebx & 0xfff) + 1) * (ecx + 1)
);
}

In the example above we use the cpuid instruction to get the L1 cache size. This instruction writes to eax, ebx, ecx, and edx, but for the cache size we only care about the contents of ebx and ecx.

However we still need to tell the compiler that eax and edx have been modified so that it can save any values that were in these registers before the asm. This is done by declaring these as outputs but with _ instead of a variable name, which indicates that the output value is to be discarded.

This can also be used with a general register class (e.g. reg) to obtain a scratch register for use inside the asm code:


#![allow(unused)]
fn main() {
#![feature(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);
}

Symbol operands

A special operand type, sym, allows you to use the symbol name of a fn or static in inline assembly code. This allows you to call a function or access a global variable without needing to keep its address in a register.


#![allow(unused)]
fn main() {
#![feature(asm)]
extern "C" fn foo(arg: i32) {
    println!("arg = {}", arg);
}

fn call_foo(arg: i32) {
    unsafe {
        asm!(
            "call {}",
            sym foo,
            // 1st argument in rdi, which is caller-saved
            inout("rdi") arg => _,
            // All caller-saved registers must be marked as clobberred
            out("rax") _, out("rcx") _, out("rdx") _, out("rsi") _,
            out("r8") _, out("r9") _, out("r10") _, out("r11") _,
            out("xmm0") _, out("xmm1") _, out("xmm2") _, out("xmm3") _,
            out("xmm4") _, out("xmm5") _, out("xmm6") _, out("xmm7") _,
            out("xmm8") _, out("xmm9") _, out("xmm10") _, out("xmm11") _,
            out("xmm12") _, out("xmm13") _, out("xmm14") _, out("xmm15") _,
        )
    }
}
}

Note that the fn or static item does not need to be public or #[no_mangle]: the compiler will automatically insert the appropriate mangled symbol name into the assembly code.

Register template modifiers

In some cases, fine control is needed over the way a register name is formatted when inserted into the template string. This is needed when an architecture's assembly language has several names for the same register, each typically being a "view" over a subset of the register (e.g. the low 32 bits of a 64-bit register).

By default the compiler will always choose the name that refers to the full register size (e.g. rax on x86-64, eax on x86, etc).

This default can be overriden by using modifiers on the template string operands, just like you would with format strings:


#![allow(unused)]
fn main() {
#![feature(asm)]
let mut x: u16 = 0xab;

unsafe {
    asm!("mov {0:h}, {0:l}", inout(reg_abcd) x);
}

assert_eq!(x, 0xabab);
}

In this example, we use the reg_abcd register class to restrict the register allocator to the 4 legacy x86 register (ax, bx, cx, dx) of which the first two bytes can be addressed independently.

Let us assume that the register allocator has chosen to allocate x in the ax register. The h modifier will emit the register name for the high byte of that register and the l modifier will emit the register name for the low byte. The asm code will therefore be expanded as mov ah, al which copies the low byte of the value into the high byte.

If you use a smaller data type (e.g. u16) with an operand and forget the use template modifiers, the compiler will emit a warning and suggest the correct modifier to use.

Memory address operands

Sometimes assembly instructions require operands passed via memory addresses/memory locations. You have to manually use the memory address syntax specified by the respectively architectures. For example, in x86/x86_64 and intel assembly syntax, you should wrap inputs/outputs in [] to indicate they are memory operands:


#![allow(unused)]
fn main() {
#![feature(asm, llvm_asm)]
fn load_fpu_control_word(control: u16) {
unsafe {
    asm!("fldcw [{}]", in(reg) &control, options(nostack));

    // Previously this would have been written with the deprecated `llvm_asm!` like this
    llvm_asm!("fldcw $0" :: "m" (control) :: "volatile");
}
}
}

Options

By default, an inline assembly block is treated the same way as an external FFI function call with a custom calling convention: it may read/write memory, have observable side effects, etc. However in many cases, it is desirable to give the compiler more information about what the assembly code is actually doing so that it can optimize better.

Let's take our previous example of an add instruction:


#![allow(unused)]
fn main() {
#![feature(asm)]
let mut a: u64 = 4;
let b: u64 = 4;
unsafe {
    asm!(
        "add {0}, {1}",
        inlateout(reg) a, in(reg) b,
        options(pure, nomem, nostack),
    );
}
assert_eq!(a, 8);
}

Options can be provided as an optional final argument to the asm! macro. We specified three options here:

  • pure means that the asm code has no observable side effects and that its output depends only on its inputs. This allows the compiler optimizer to call the inline asm fewer times or even eliminate it entirely.
  • nomem means that the asm code does not read or write to memory. By default the compiler will assume that inline assembly can read or write any memory address that is accessible to it (e.g. through a pointer passed as an operand, or a global).
  • nostack means that the asm code does not push any data onto the stack. This allows the compiler to use optimizations such as the stack red zone on x86-64 to avoid stack pointer adjustments.

These allow the compiler to better optimize code using asm!, for example by eliminating pure asm! blocks whose outputs are not needed.

See the reference for the full list of available options and their effects.

Reference-level explanation

Inline assembler is implemented as an unsafe macro asm!(). The first argument to this macro is a template string literal used to build the final assembly. The following arguments specify input and output operands. When required, options are specified as the final argument.

The following ABNF specifies the general syntax:

dir_spec := "in" / "out" / "lateout" / "inout" / "inlateout"
reg_spec := <register class> / "<explicit register>"
operand_expr := expr / "_" / expr "=>" expr / expr "=>" "_"
reg_operand := dir_spec "(" reg_spec ")" operand_expr
operand := reg_operand / "const" const_expr / "sym" path
option := "pure" / "nomem" / "readonly" / "preserves_flags" / "noreturn" / "nostack" / "att_syntax"
options := "options(" option *["," option] [","] ")"
asm := "asm!(" format_string *("," format_string) *("," [ident "="] operand) ["," options] [","] ")"

The macro will initially be supported only on ARM, AArch64, Hexagon, x86, x86-64 and RISC-V targets. Support for more targets may be added in the future. The compiler will emit an error if asm! is used on an unsupported target.

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, named arguments must appear after positional arguments. Explicit register operands must appear at the end of the operand list, after named arguments if any.

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.

The 5 targets specified in this RFC (x86, ARM, AArch64, RISC-V, Hexagon) all use the assembly code syntax 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.

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 is written to 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 an in.
    • You should only write to the register after all inputs are read, otherwise you may clobber an input.
  • 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 is written to 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 is written to 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.
  • inlateout(<reg>) <expr> / inlateout(<reg>) <in expr> => <out expr>
    • Identical to inout except that the register allocator can reuse a register allocated to an in (this can happen if the compiler knows the in has the same initial value as the inlateout).
    • You should only write to the register after all inputs are read, otherwise you may clobber an input.
  • const <expr>
    • <expr> must be an integer or floating-point constant expression.
    • The value of the expression is formatted as a string and substituted directly into the asm template string.
  • sym <path>
    • <path> must refer to a fn or static.
    • 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.

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.

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). Using string literals for register names enables support for architectures that use special characters in register names, such as MIPS ($0, $1, etc).

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 implement Copy). This includes architecture-specific vector types defined in std::arch such as __m128 (x86) or int8x16_t (ARM).

Here is the list of currently supported register classes:

ArchitectureRegister classRegistersLLVM constraint code
x86regax, bx, cx, dx, si, di, r[8-15] (x86-64 only)r
x86reg_abcdax, bx, cx, dxQ
x86-32reg_byteal, bl, cl, dl, ah, bh, ch, dhq
x86-64reg_byteal, bl, cl, dl, sil, dil, r[8-15]b, ah*, bh*, ch*, dh*q
x86xmm_regxmm[0-7] (x86) xmm[0-15] (x86-64)x
x86ymm_regymm[0-7] (x86) ymm[0-15] (x86-64)x
x86zmm_regzmm[0-7] (x86) zmm[0-31] (x86-64)v
x86kregk[1-7]Yk
AArch64regx[0-28], x30r
AArch64vregv[0-31]w
AArch64vreg_low16v[0-15]x
ARMregr[0-5] r7*, r[8-10], r11*, r12, r14r
ARM (Thumb)reg_thumbr[0-r7]l
ARM (ARM)reg_thumbr[0-r10], r12, r14l
ARMsregs[0-31]t
ARMsreg_low16s[0-15]x
ARMdregd[0-31]w
ARMdreg_low16d[0-15]t
ARMdreg_low8d[0-8]x
ARMqregq[0-15]w
ARMqreg_low8q[0-7]t
ARMqreg_low4q[0-3]x
MIPSreg$[2-25]r
MIPSfreg$f[0-31]f
NVPTXreg16None*h
NVPTXreg32None*r
NVPTXreg64None*l
RISC-Vregx1, x[5-7], x[9-15], x[16-31] (non-RV32E)r
RISC-Vfregf[0-31]f
Hexagonregr[0-28]r

Note: On x86 we treat reg_byte differently from reg because the compiler can allocate al and ah separately whereas reg reserves the whole register.

Note #2: On x86-64 the high byte registers (e.g. ah) are only available when used as an explicit register. Specifying the reg_byte register class for an operand will always allocate a low byte register.

Note #3: NVPTX doesn't have a fixed register set, so named registers are not supported.

Note #4: On ARM the frame pointer is either r7 or r11 depending on the platform.

Additional register classes may be added in the future based on demand (e.g. MMX, x87, etc).

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.

ArchitectureRegister classTarget featureAllowed types
x86-32regNonei16, i32, f32
x86-64regNonei16, i32, f32, i64, f64
x86reg_byteNonei8
x86xmm_regssei32, f32, i64, f64,
i8x16, i16x8, i32x4, i64x2, f32x4, f64x2
x86ymm_regavxi32, f32, i64, f64,
i8x16, i16x8, i32x4, i64x2, f32x4, f64x2
i8x32, i16x16, i32x8, i64x4, f32x8, f64x4
x86zmm_regavx512fi32, f32, i64, f64,
i8x16, i16x8, i32x4, i64x2, f32x4, f64x2
i8x32, i16x16, i32x8, i64x4, f32x8, f64x4
i8x64, i16x32, i32x16, i64x8, f32x16, f64x8
x86kregaxv512fi8, i16
x86kregaxv512bwi32, i64
AArch64regNonei8, i16, i32, f32, i64, f64
AArch64vregfpi8, i16, i32, f32, i64, f64,
i8x8, i16x4, i32x2, i64x1, f32x2, f64x1,
i8x16, i16x8, i32x4, i64x2, f32x4, f64x2
ARMregNonei8, i16, i32, f32
ARMsregvfp2i32, f32
ARMdregvfp2i64, f64, i8x8, i16x4, i32x2, i64x1, f32x2
ARMqregneoni8x16, i16x8, i32x4, i64x2, f32x4
MIPS32regNonei8, i16, i32, f32
MIPS32fregNonef32, f64
MIPS64regNonei8, i16, i32, i64, f32, f64
MIPS64fregNonef32, f64
NVPTXreg16Nonei8, i16
NVPTXreg32Nonei8, i16, i32, f32
NVPTXreg64Nonei8, i16, i32, f32, i64, f64
RISC-V32regNonei8, i16, i32, f32
RISC-V64regNonei8, i16, i32, f32, i64, f64
RISC-Vfregff32
RISC-Vfregdf64
HexagonregNonei8, i16, i32, f32

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:

ArchitectureBase registerAliases
x86axeax, rax
x86bxebx, rbx
x86cxecx, rcx
x86dxedx, rdx
x86siesi, rsi
x86diedi, rdi
x86bpbpl, ebp, rbp
x86spspl, esp, rsp
x86ipeip, rip
x86st(0)st
x86r[8-15]r[8-15]b, r[8-15]w, r[8-15]d
x86xmm[0-31]ymm[0-31], zmm[0-31]
AArch64x[0-30]w[0-30]
AArch64x29fp
AArch64x30lr
AArch64spwsp
AArch64xzrwzr
AArch64v[0-31]b[0-31], h[0-31], s[0-31], d[0-31], q[0-31]
ARMr[0-3]a[1-4]
ARMr[4-9]v[1-6]
ARMr9rfp
ARMr10sl
ARMr11fp
ARMr12ip
ARMr13sp
ARMr14lr
ARMr15pc
RISC-Vx0zero
RISC-Vx1ra
RISC-Vx2sp
RISC-Vx3gp
RISC-Vx4tp
RISC-Vx[5-7]t[0-2]
RISC-Vx8fp, s0
RISC-Vx9s1
RISC-Vx[10-17]a[0-7]
RISC-Vx[18-27]s[2-11]
RISC-Vx[28-31]t[3-6]
RISC-Vf[0-7]ft[0-7]
RISC-Vf[8-9]fs[0-1]
RISC-Vf[10-17]fa[0-7]
RISC-Vf[18-27]fs[2-11]
RISC-Vf[28-31]ft[8-11]
Hexagonr29sp
Hexagonr30fr
Hexagonr31lr

Some registers cannot be used for input or output operands:

ArchitectureUnsupported registerReason
AllspThe stack pointer must be restored to its original value at the end of an asm code block.
Allbp (x86), x29 (AArch64), x8 (RISC-V), fr (Hexagon), $fp (MIPS)The frame pointer cannot be used as an input or output.
ARMr7 or r11On 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.
ARMr6r6 is used internally by LLVM as a base pointer and therefore cannot be used as an input or output.
x86k0This is a constant zero register which can't be modified.
x86ipThis is the program counter, not a real register.
x86mm[0-7]MMX registers are not currently supported (but may be in the future).
x86st([0-7])x87 registers are not currently supported (but may be in the future).
AArch64xzrThis is a constant zero register which can't be modified.
ARMpcThis is the program counter, not a real register.
MIPS$0 or $zeroThis is a constant zero register which can't be modified.
MIPS$1 or $atReserved for assembler.
MIPS$26/$k0, $27/$k1OS-reserved registers.
MIPS$28/$gpGlobal pointer cannot be used as inputs or outputs.
MIPS$raReturn address cannot be used as inputs or outputs.
RISC-Vx0This is a constant zero register which can't be modified.
RISC-Vgp, tpThese registers are reserved and cannot be used as inputs or outputs.
HexagonlrThis is the link register which cannot be used as an input or output.

In some cases LLVM will allocate a "reserved register" for reg operands even though this register cannot be explicitly specified. Assembly code making use of reserved registers should be careful since reg operands may alias with those registers. Reserved registers are:

  • The frame pointer on all architectures.
  • r6 on ARM.

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.

ArchitectureRegister classModifierExample outputLLVM modifier
x86-32regNoneeaxk
x86-64regNoneraxq
x86-32reg_abcdlalb
x86-64reglalb
x86reg_abcdhahh
x86regxaxw
x86regeeaxk
x86-64regrraxq
x86reg_byteNoneal / ahNone
x86xmm_regNonexmm0x
x86ymm_regNoneymm0t
x86zmm_regNonezmm0g
x86*mm_regxxmm0x
x86*mm_regyymm0t
x86*mm_regzzmm0g
x86kregNonek1None
AArch64regNonex0x
AArch64regww0w
AArch64regxx0x
AArch64vregNonev0None
AArch64vregvv0None
AArch64vregbb0b
AArch64vreghh0h
AArch64vregss0s
AArch64vregdd0d
AArch64vregqq0q
ARMregNoner0None
ARMsregNones0None
ARMdregNoned0P
ARMqregNoneq0q
ARMqrege / fd0 / d1e / f
MIPSregNone$2None
MIPSfregNone$f0None
NVPTXreg16Noners0None
NVPTXreg32Noner0None
NVPTXreg64Nonerd0None
RISC-VregNonex1None
RISC-VfregNonef0None
HexagonregNoner0None

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: the x, t and g 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.

Options

Flags are used to further influence the behavior of the inline assembly block. Currently the following options are defined:

  • pure: The asm block has no side effects, 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 the nomem options is also set). This allows the compiler to execute the asm 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.
  • nomem: The asm blocks does not read or write to any memory. This allows the compiler to cache the values of modified global variables in registers across the asm block since it knows that they are not read or written to by the asm.
  • readonly: The asm block does not write to any memory. This allows the compiler to cache the values of unmodified global variables in registers across the asm block since it knows that they are not written to by the asm.
  • preserves_flags: The asm block does not modify the flags register (defined in the rules below). This allows the compiler to avoid recomputing the condition flags after the asm block.
  • noreturn: The asm 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. A noreturn asm block behaves just like a function which doesn't return; notably, local variables in scope are not dropped before it is invoked.
  • nostack: The asm 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 %.

The compiler performs some additional checks on options:

  • The nomem and readonly options are mutually exclusive: it is a compile-time error to specify both.
  • The pure option must be combined with either the nomem or readonly options, otherwise a compile-time error is emitted.
  • 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.

Rules for inline assembly

  • 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).
  • 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 an in, 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 the 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 (outside the scope of this RFC).
  • 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 the asm has side-effects other than its direct outputs. Behavior is also undefined if two executions of the asm code with the same inputs result in different outputs.
    • When used with the nomem option, "inputs" are just the direct inputs of the asm!.
    • When used with the readonly option, "inputs" comprise the direct inputs of the asm! and any memory that the asm! block is allowed to read.
  • These flags registers must be restored upon exiting the asm block if the preserves_flags option is set:
    • x86
      • Status flags in EFLAGS (CF, PF, AF, ZF, SF, OF).
      • Floating-point status word (all).
      • Floating-point exception flags in MXCSR (PE, UE, OE, ZE, DE, IE).
    • ARM
      • Condition flags in CPSR (N, Z, C, V)
      • Saturation flag in CPSR (Q)
      • Greater than or equal flags in CPSR (GE).
      • Condition flags in FPSCR (N, Z, C, V)
      • Saturation flag in FPSCR (QC)
      • Floating-point exception flags in FPSCR (IDC, IXC, UFC, OFC, DZC, IOC).
    • AArch64
      • Condition flags (NZCV register).
      • Floating-point status (FPSR register).
    • RISC-V
      • Floating-point exception flags in fcsr (fflags).
  • On x86, the direction flag (DF in EFLAGS) is clear on entry to an asm block and must be clear on exit.
    • Behavior is undefined if the direction flag is set on exiting an asm block.
  • The requirement of restoring the stack pointer and non-output registers to their original value only applies when exiting an asm! block.
    • This means that asm! blocks that never return (even if not marked noreturn) don't need to preserve these registers.
    • When returning to a different asm! block than you entered (e.g. for context switching), these registers must contain the value they had upon entering the asm! block that you are exiting.
      • You cannot exit an asm! block that has not been entered. Neither can you exit an asm! block that has already been exited.
      • You are responsible for switching any target-specific state (e.g. thread-local storage, stack bounds).
      • The set of memory locations that you may access is the intersection of those allowed by the asm! blocks you entered and exited.
  • You cannot assume that an asm! block will appear exactly once in the output binary. The compiler is allowed to instantiate multiple copies of the asm! block, for example when the function containing it is inlined in multiple places.
    • As a consequence, you should only use local labels inside inline assembly code. Defining symbols in assembly code may lead to assembler and/or linker errors due to duplicate symbol definitions.

Note: As a general rule, the flags covered by preserves_flags are those which are not preserved when performing a function call.

assoc_char_consts

The tracking issue for this feature is: #71763


assoc_char_funcs

The tracking issue for this feature is: #71763


atomic_from_mut

The tracking issue for this feature is: #76314


atomic_mut_ptr

The tracking issue for this feature is: #66893


available_concurrency

The tracking issue for this feature is: #74479


backtrace

The tracking issue for this feature is: #53487


binary_heap_drain_sorted

The tracking issue for this feature is: #59278


binary_heap_into_iter_sorted

The tracking issue for this feature is: #59278


binary_heap_retain

The tracking issue for this feature is: #71503


bool_to_option

The tracking issue for this feature is: #64260


bound_cloned

The tracking issue for this feature is: #61356


box_into_boxed_slice

The tracking issue for this feature is: #71582


box_into_pin

The tracking issue for this feature is: #62370


btree_drain_filter

The tracking issue for this feature is: #70530


bufreader_seek_relative

The tracking issue for this feature is: #31100


c_void_variant

This feature is internal to the Rust compiler and is not intended for general use.


can_vector

The tracking issue for this feature is: #69941


cell_leak

The tracking issue for this feature is: #69099


cell_update

The tracking issue for this feature is: #50186


cfg_accessible

The tracking issue for this feature is: #64797


char_error_internals

This feature is internal to the Rust compiler and is not intended for general use.


char_internals

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


clamp

The tracking issue for this feature is: #44095


cmp_min_max_by

The tracking issue for this feature is: #64460


coerce_unsized

The tracking issue for this feature is: #27732


command_access

The tracking issue for this feature is: #44434


concat_idents

The tracking issue for this feature is: #29599


The concat_idents feature adds a macro for concatenating multiple identifiers into one identifier.

Examples

#![feature(concat_idents)]

fn main() {
    fn foobar() -> u32 { 23 }
    let f = concat_idents!(foo, bar);
    assert_eq!(f(), 23);
}

const_align_of_val

The tracking issue for this feature is: #46571


const_alloc_layout

The tracking issue for this feature is: #67521


const_assume

The tracking issue for this feature is: #76972


const_btree_new

The tracking issue for this feature is: #71835


const_caller_location

The tracking issue for this feature is: #76156


const_checked_int_methods

The tracking issue for this feature is: #53718


const_cow_is_borrowed

The tracking issue for this feature is: #65143


const_cstr_unchecked

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


const_cttz

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


const_discriminant

The tracking issue for this feature is: #69821


const_euclidean_int_methods

The tracking issue for this feature is: #53718


const_float_bits_conv

The tracking issue for this feature is: #72447


const_float_classify

The tracking issue for this feature is: #72505


const_int_pow

The tracking issue for this feature is: #53718


const_int_unchecked_arith

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


const_ip

The tracking issue for this feature is: #76205


const_ipv4

The tracking issue for this feature is: #76205


const_ipv6

The tracking issue for this feature is: #76205


const_likely

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


const_maybe_uninit_as_ptr

The tracking issue for this feature is: #75251


const_nonnull_slice_from_raw_parts

The tracking issue for this feature is: #71941


const_option

The tracking issue for this feature is: #67441


const_overflowing_int_methods

The tracking issue for this feature is: #53718


const_pin

The tracking issue for this feature is: #76654


const_pref_align_of

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


const_ptr_is_null

The tracking issue for this feature is: #74939


const_ptr_offset

The tracking issue for this feature is: #71499


const_ptr_offset_from

The tracking issue for this feature is: #41079


const_raw_ptr_comparison

The tracking issue for this feature is: #53020


const_size_of_val

The tracking issue for this feature is: #46571


const_slice_from_raw_parts

The tracking issue for this feature is: #67456


const_slice_ptr_len

The tracking issue for this feature is: #71146


const_str_from_utf8_unchecked

The tracking issue for this feature is: #75196


const_type_id

The tracking issue for this feature is: #77125


const_type_name

The tracking issue for this feature is: #63084


const_unreachable_unchecked

The tracking issue for this feature is: #53188


const_wrapping_int_methods

The tracking issue for this feature is: #53718


constctlz

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


container_error_extra

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


control_flow_enum

The tracking issue for this feature is: #75744


convert_float_to_int

The tracking issue for this feature is: #67057


core_intrinsics

This feature is internal to the Rust compiler and is not intended for general use.


core_panic

This feature is internal to the Rust compiler and is not intended for general use.


core_private_bignum

This feature is internal to the Rust compiler and is not intended for general use.


core_private_diy_float

This feature is internal to the Rust compiler and is not intended for general use.


cow_is_borrowed

The tracking issue for this feature is: #65143


cstring_from_vec_with_nul

The tracking issue for this feature is: #73179


deadline_api

The tracking issue for this feature is: #46316


debug_non_exhaustive

The tracking issue for this feature is: #67364


dec2flt

This feature is internal to the Rust compiler and is not intended for general use.


default_free_fn

The tracking issue for this feature is: #73014


Adds a free default() function to the std::default module. This function just forwards to Default::default(), but may remove repetition of the word "default" from the call site.

Here is an example:

#![feature(default_free_fn)]
use std::default::default;

#[derive(Default)]
struct AppConfig {
    foo: FooConfig,
    bar: BarConfig,
}

#[derive(Default)]
struct FooConfig {
    foo: i32,
}

#[derive(Default)]
struct BarConfig {
    bar: f32,
    baz: u8,
}

fn main() {
    let options = AppConfig {
        foo: default(),
        bar: BarConfig {
            bar: 10.1,
            ..default()
        },
    };
}

deque_range

The tracking issue for this feature is: #74217


derive_clone_copy

This feature is internal to the Rust compiler and is not intended for general use.


derive_eq

This feature is internal to the Rust compiler and is not intended for general use.


discriminant_kind

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


dispatch_from_dyn

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


div_duration

The tracking issue for this feature is: #63139


drain_filter

The tracking issue for this feature is: #43244


duration_constants

The tracking issue for this feature is: #57391


duration_consts_2

The tracking issue for this feature is: #72440


duration_saturating_ops

The tracking issue for this feature is: #76416


duration_zero

The tracking issue for this feature is: #73544


entry_insert

The tracking issue for this feature is: #65225


error_iter

The tracking issue for this feature is: #58520


error_type_id

The tracking issue for this feature is: #60784


exact_size_is_empty

The tracking issue for this feature is: #35428


extend_one

The tracking issue for this feature is: #72631


fd

This feature is internal to the Rust compiler and is not intended for general use.


fd_read

This feature is internal to the Rust compiler and is not intended for general use.


fixed_size_array

The tracking issue for this feature is: #27778


flt2dec

This feature is internal to the Rust compiler and is not intended for general use.


fmt_as_str

The tracking issue for this feature is: #74442


fmt_internals

This feature is internal to the Rust compiler and is not intended for general use.


fn_traits

The tracking issue for this feature is #29625

See Also: unboxed_closures


The fn_traits feature allows for implementation of the Fn* traits for creating custom closure-like types.

#![feature(unboxed_closures)]
#![feature(fn_traits)]

struct Adder {
    a: u32
}

impl FnOnce<(u32, )> for Adder {
    type Output = u32;
    extern "rust-call" fn call_once(self, b: (u32, )) -> Self::Output {
        self.a + b.0
    }
}

fn main() {
    let adder = Adder { a: 3 };
    assert_eq!(adder(2), 5);
}

foo

The tracking issue for this feature is: #1234


forget_unsized

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


format_args_nl

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


future_poll_fn

The tracking issue for this feature is: #72302


gen_future

The tracking issue for this feature is: #50547


generator_trait

The tracking issue for this feature is: #43122


get_mut_unchecked

The tracking issue for this feature is: #63292


global_asm

The tracking issue for this feature is: #35119


The global_asm! macro allows the programmer to write arbitrary assembly outside the scope of a function body, passing it through rustc and llvm to the assembler. The macro is a no-frills interface to LLVM's concept of module-level inline assembly. That is, all caveats applicable to LLVM's module-level inline assembly apply to global_asm!.

global_asm! fills a role not currently satisfied by either asm! or #[naked] functions. The programmer has all features of the assembler at their disposal. The linker will expect to resolve any symbols defined in the inline assembly, modulo any symbols marked as external. It also means syntax for directives and assembly follow the conventions of the assembler in your toolchain.

A simple usage looks like this:

#![feature(global_asm)]
you also need relevant target_arch cfgs
global_asm!(include_str!("something_neato.s"));

And a more complicated usage looks like this:

#![feature(global_asm)]
#![cfg(any(target_arch = "x86", target_arch = "x86_64"))]

pub mod sally {
    global_asm!(r#"
        .global foo
      foo:
        jmp baz
    "#);

    #[no_mangle]
    pub unsafe extern "C" fn baz() {}
}

// the symbols `foo` and `bar` are global, no matter where
// `global_asm!` was used.
extern "C" {
    fn foo();
    fn bar();
}

pub mod harry {
    global_asm!(r#"
        .global bar
      bar:
        jmp quux
    "#);

    #[no_mangle]
    pub unsafe extern "C" fn quux() {}
}

You may use global_asm! multiple times, anywhere in your crate, in whatever way suits you. The effect is as if you concatenated all usages and placed the larger, single usage in the crate root.


If you don't need quite as much power and flexibility as global_asm! provides, and you don't mind restricting your inline assembly to fn bodies only, you might try the asm feature instead.

hash_drain_filter

The tracking issue for this feature is: #59618


hash_raw_entry

The tracking issue for this feature is: #56167


hash_set_entry

The tracking issue for this feature is: #60896


hashmap_internals

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


inplace_iteration

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


int_bits_const

The tracking issue for this feature is: #76904


int_error_internals

This feature is internal to the Rust compiler and is not intended for general use.


int_error_matching

The tracking issue for this feature is: #22639


integer_atomics

The tracking issue for this feature is: #32976


into_future

The tracking issue for this feature is: #67644


io_slice_advance

The tracking issue for this feature is: #62726


ip

The tracking issue for this feature is: #27709


is_sorted

The tracking issue for this feature is: #53485


Add the methods is_sorted, is_sorted_by and is_sorted_by_key to [T]; add the methods is_sorted, is_sorted_by and is_sorted_by_key to Iterator.

iter_advance_by

The tracking issue for this feature is: #77404


iter_is_partitioned

The tracking issue for this feature is: #62544


iter_map_while

The tracking issue for this feature is: #68537


iter_order_by

The tracking issue for this feature is: #64295


iter_partition_in_place

The tracking issue for this feature is: #62543


iterator_fold_self

The tracking issue for this feature is: #68125


layout_for_ptr

The tracking issue for this feature is: #69835


liballoc_internals

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


libstd_io_internals

This feature is internal to the Rust compiler and is not intended for general use.


libstd_sys_internals

This feature is internal to the Rust compiler and is not intended for general use.


libstd_thread_internals

This feature is internal to the Rust compiler and is not intended for general use.


linked_list_cursors

The tracking issue for this feature is: #58533


linked_list_extras

The tracking issue for this feature is: #27794


linked_list_prepend

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


linked_list_remove

The tracking issue for this feature is: #69210


llvm_asm

The tracking issue for this feature is: #70173


For extremely low-level manipulations and performance reasons, one might wish to control the CPU directly. Rust supports using inline assembly to do this via the llvm_asm! macro.

llvm_asm!(assembly template
   : output operands
   : input operands
   : clobbers
   : options
   );

Any use of llvm_asm is feature gated (requires #![feature(llvm_asm)] on the crate to allow) and of course requires an unsafe block.

Note: the examples here are given in x86/x86-64 assembly, but all platforms are supported.

Assembly template

The assembly template is the only required parameter and must be a literal string (i.e. "")

#![feature(llvm_asm)]

#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
fn foo() {
    unsafe {
        llvm_asm!("NOP");
    }
}

// Other platforms:
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
fn foo() { /* ... */ }

fn main() {
    // ...
    foo();
    // ...
}

(The feature(llvm_asm) and #[cfg]s are omitted from now on.)

Output operands, input operands, clobbers and options are all optional but you must add the right number of : if you skip them:

#![feature(llvm_asm)]
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
fn main() { unsafe {
llvm_asm!("xor %eax, %eax"
    :
    :
    : "eax"
   );
} }
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
fn main() {}

Whitespace also doesn't matter:

#![feature(llvm_asm)]
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
fn main() { unsafe {
llvm_asm!("xor %eax, %eax" ::: "eax");
} }
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
fn main() {}

Operands

Input and output operands follow the same format: : "constraints1"(expr1), "constraints2"(expr2), ...". Output operand expressions must be mutable place, or not yet assigned:

#![feature(llvm_asm)]
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
fn add(a: i32, b: i32) -> i32 {
    let c: i32;
    unsafe {
        llvm_asm!("add $2, $0"
             : "=r"(c)
             : "0"(a), "r"(b)
             );
    }
    c
}
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
fn add(a: i32, b: i32) -> i32 { a + b }

fn main() {
    assert_eq!(add(3, 14159), 14162)
}

If you would like to use real operands in this position, however, you are required to put curly braces {} around the register that you want, and you are required to put the specific size of the operand. This is useful for very low level programming, where which register you use is important:


#![allow(unused)]
fn main() {
#![feature(llvm_asm)]
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
unsafe fn read_byte_in(port: u16) -> u8 {
let result: u8;
llvm_asm!("in %dx, %al" : "={al}"(result) : "{dx}"(port));
result
}
}

Clobbers

Some instructions modify registers which might otherwise have held different values so we use the clobbers list to indicate to the compiler not to assume any values loaded into those registers will stay valid.

#![feature(llvm_asm)]
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
fn main() { unsafe {
// Put the value 0x200 in eax:
llvm_asm!("mov $$0x200, %eax" : /* no outputs */ : /* no inputs */ : "eax");
} }
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
fn main() {}

Input and output registers need not be listed since that information is already communicated by the given constraints. Otherwise, any other registers used either implicitly or explicitly should be listed.

If the assembly changes the condition code register cc should be specified as one of the clobbers. Similarly, if the assembly modifies memory, memory should also be specified.

Options

The last section, options is specific to Rust. The format is comma separated literal strings (i.e. :"foo", "bar", "baz"). It's used to specify some extra info about the inline assembly:

Current valid options are:

  1. volatile - specifying this is analogous to __asm__ __volatile__ (...) in gcc/clang.
  2. alignstack - certain instructions expect the stack to be aligned a certain way (i.e. SSE) and specifying this indicates to the compiler to insert its usual stack alignment code
  3. intel - use intel syntax instead of the default AT&T.
#![feature(llvm_asm)]
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
fn main() {
let result: i32;
unsafe {
   llvm_asm!("mov eax, 2" : "={eax}"(result) : : : "intel")
}
println!("eax is currently {}", result);
}
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
fn main() {}

More Information

The current implementation of the llvm_asm! macro is a direct binding to LLVM's inline assembler expressions, so be sure to check out their documentation as well for more information about clobbers, constraints, etc.

If you need more power and don't mind losing some of the niceties of llvm_asm!, check out global_asm.

log_syntax

The tracking issue for this feature is: #29598


map_entry_replace

The tracking issue for this feature is: #44286


map_first_last

The tracking issue for this feature is: #62924


map_into_keys_values

The tracking issue for this feature is: #75294


maybe_uninit_extra

The tracking issue for this feature is: #63567


maybe_uninit_ref

The tracking issue for this feature is: #63568


maybe_uninit_slice

The tracking issue for this feature is: #63569


maybe_uninit_uninit_array

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


new_uninit

The tracking issue for this feature is: #63291


nonnull_slice_from_raw_parts

The tracking issue for this feature is: #71941


num_as_ne_bytes

The tracking issue for this feature is: #76976


once_cell

The tracking issue for this feature is: #74465


once_poison

The tracking issue for this feature is: #33577


option_expect_none

The tracking issue for this feature is: #62633


option_insert

The tracking issue for this feature is: #78271


option_result_contains

The tracking issue for this feature is: #62358


option_unwrap_none

The tracking issue for this feature is: #62633


option_zip

The tracking issue for this feature is: #70086


or_insert_with_key

The tracking issue for this feature is: #71024


osstring_ascii

The tracking issue for this feature is: #70516


panic_abort

The tracking issue for this feature is: #32837


panic_info_message

The tracking issue for this feature is: #66745


panic_internals

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


panic_unwind

The tracking issue for this feature is: #32837


partition_point

The tracking issue for this feature is: #73831


pattern

The tracking issue for this feature is: #27721


peekable_next_if

The tracking issue for this feature is: #72480


peer_credentials_unix_socket

The tracking issue for this feature is: #42839


pin_static_ref

The tracking issue for this feature is: #78186


poll_map

The tracking issue for this feature is: #63514


print_internals

This feature is internal to the Rust compiler and is not intended for general use.


proc_macro_def_site

The tracking issue for this feature is: #54724


proc_macro_diagnostic

The tracking issue for this feature is: #54140


proc_macro_internals

The tracking issue for this feature is: #27812


proc_macro_is_available

The tracking issue for this feature is: #71436


proc_macro_quote

The tracking issue for this feature is: #54722


proc_macro_span

The tracking issue for this feature is: #54725


proc_macro_tracked_env

The tracking issue for this feature is: #74690


process_exitcode_placeholder

The tracking issue for this feature is: #48711


process_internals

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


profiler_runtime_lib

This feature is internal to the Rust compiler and is not intended for general use.


ptr_as_uninit

The tracking issue for this feature is: #75402


ptr_internals

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


range_bounds_assert_len

The tracking issue for this feature is: #76393


This adds RangeBounds::assert_len.

raw

The tracking issue for this feature is: #27751


raw_ref_macros

The tracking issue for this feature is: #73394


raw_vec_internals

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


read_initializer

The tracking issue for this feature is: #42788


ready_macro

The tracking issue for this feature is: #70922


receiver_trait

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


refcell_take

The tracking issue for this feature is: #71395


renamed_spin_loop

The tracking issue for this feature is: #55002


result_cloned

The tracking issue for this feature is: #63168


result_contains_err

The tracking issue for this feature is: #62358


result_copied

The tracking issue for this feature is: #63168


result_flattening

The tracking issue for this feature is: #70142


rt

This feature is internal to the Rust compiler and is not intended for general use.


seek_convenience

The tracking issue for this feature is: #59359


set_ptr_value

The tracking issue for this feature is: #75091


set_stdio

This feature is internal to the Rust compiler and is not intended for general use.


sgx_platform

The tracking issue for this feature is: #56975


shrink_to

The tracking issue for this feature is: #56431


slice_concat_ext

The tracking issue for this feature is: #27747


slice_concat_trait

The tracking issue for this feature is: #27747


slice_fill

The tracking issue for this feature is: #70758


slice_index_methods

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


slice_internals

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


slice_iter_mut_as_slice

The tracking issue for this feature is: #58957


slice_partition_at_index

The tracking issue for this feature is: #55300


slice_partition_dedup

The tracking issue for this feature is: #54279


slice_ptr_get

The tracking issue for this feature is: #74265


slice_ptr_len

The tracking issue for this feature is: #71146


slice_split_at_unchecked

The tracking issue for this feature is: #76014


slice_strip

The tracking issue for this feature is: #73413


sort_internals

This feature is internal to the Rust compiler and is not intended for general use.


split_inclusive

The tracking issue for this feature is: #72360


std_internals

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


stdsimd

The tracking issue for this feature is: #48556


step_trait

The tracking issue for this feature is: #42168


step_trait_ext

The tracking issue for this feature is: #42168


str_internals

This feature is internal to the Rust compiler and is not intended for general use.


str_split_as_str

The tracking issue for this feature is: #77998


str_split_inclusive_as_str

The tracking issue for this feature is: #77998


str_split_once

The tracking issue for this feature is: #74773


string_drain_as_str

The tracking issue for this feature is: #76905


termination_trait_lib

The tracking issue for this feature is: #43301


test

The tracking issue for this feature is: None.


The internals of the test crate are unstable, behind the test flag. The most widely used part of the test crate are benchmark tests, which can test the performance of your code. Let's make our src/lib.rs look like this (comments elided):

#![feature(test)]

extern crate test;

pub fn add_two(a: i32) -> i32 {
    a + 2
}

#[cfg(test)]
mod tests {
    use super::*;
    use test::Bencher;

    #[test]
    fn it_works() {
        assert_eq!(4, add_two(2));
    }

    #[bench]
    fn bench_add_two(b: &mut Bencher) {
        b.iter(|| add_two(2));
    }
}

Note the test feature gate, which enables this unstable feature.

We've imported the test crate, which contains our benchmarking support. We have a new function as well, with the bench attribute. Unlike regular tests, which take no arguments, benchmark tests take a &mut Bencher. This Bencher provides an iter method, which takes a closure. This closure contains the code we'd like to benchmark.

We can run benchmark tests with cargo bench:

$ cargo bench
   Compiling adder v0.0.1 (file:///home/steve/tmp/adder)
     Running target/release/adder-91b3e234d4ed382a

running 2 tests
test tests::it_works ... ignored
test tests::bench_add_two ... bench:         1 ns/iter (+/- 0)

test result: ok. 0 passed; 0 failed; 1 ignored; 1 measured

Our non-benchmark test was ignored. You may have noticed that cargo bench takes a bit longer than cargo test. This is because Rust runs our benchmark a number of times, and then takes the average. Because we're doing so little work in this example, we have a 1 ns/iter (+/- 0), but this would show the variance if there was one.

Advice on writing benchmarks:

  • Move setup code outside the iter loop; only put the part you want to measure inside
  • Make the code do "the same thing" on each iteration; do not accumulate or change state
  • Make the outer function idempotent too; the benchmark runner is likely to run it many times
  • Make the inner iter loop short and fast so benchmark runs are fast and the calibrator can adjust the run-length at fine resolution
  • Make the code in the iter loop do something simple, to assist in pinpointing performance improvements (or regressions)

Gotcha: optimizations

There's another tricky part to writing benchmarks: benchmarks compiled with optimizations activated can be dramatically changed by the optimizer so that the benchmark is no longer benchmarking what one expects. For example, the compiler might recognize that some calculation has no external effects and remove it entirely.

#![feature(test)]

extern crate test;
use test::Bencher;

#[bench]
fn bench_xor_1000_ints(b: &mut Bencher) {
    b.iter(|| {
        (0..1000).fold(0, |old, new| old ^ new);
    });
}

gives the following results

running 1 test
test bench_xor_1000_ints ... bench:         0 ns/iter (+/- 0)

test result: ok. 0 passed; 0 failed; 0 ignored; 1 measured

The benchmarking runner offers two ways to avoid this. Either, the closure that the iter method receives can return an arbitrary value which forces the optimizer to consider the result used and ensures it cannot remove the computation entirely. This could be done for the example above by adjusting the b.iter call to


#![allow(unused)]
fn main() {
struct X;
impl X { fn iter<T, F>(&self, _: F) where F: FnMut() -> T {} } let b = X;
b.iter(|| {
    // Note lack of `;` (could also use an explicit `return`).
    (0..1000).fold(0, |old, new| old ^ new)
});
}

Or, the other option is to call the generic test::black_box function, which is an opaque "black box" to the optimizer and so forces it to consider any argument as used.

#![feature(test)]

extern crate test;

fn main() {
struct X;
impl X { fn iter<T, F>(&self, _: F) where F: FnMut() -> T {} } let b = X;
b.iter(|| {
    let n = test::black_box(1000);

    (0..n).fold(0, |a, b| a ^ b)
})
}

Neither of these read or modify the value, and are very cheap for small values. Larger values can be passed indirectly to reduce overhead (e.g. black_box(&huge_struct)).

Performing either of the above changes gives the following benchmarking results

running 1 test
test bench_xor_1000_ints ... bench:       131 ns/iter (+/- 3)

test result: ok. 0 passed; 0 failed; 0 ignored; 1 measured

However, the optimizer can still modify a testcase in an undesirable manner even when using either of the above.

thread_id_value

The tracking issue for this feature is: #67939


thread_local_internals

This feature is internal to the Rust compiler and is not intended for general use.


thread_spawn_unchecked

The tracking issue for this feature is: #55132


toowned_clone_into

The tracking issue for this feature is: #41263


total_cmp

The tracking issue for this feature is: #72599


trace_macros

The tracking issue for this feature is #29598.


With trace_macros you can trace the expansion of macros in your code.

Examples

#![feature(trace_macros)]

fn main() {
    trace_macros!(true);
    println!("Hello, Rust!");
    trace_macros!(false);
}

The cargo build output:

note: trace_macro
 --> src/main.rs:5:5
  |
5 |     println!("Hello, Rust!");
  |     ^^^^^^^^^^^^^^^^^^^^^^^^^
  |
  = note: expanding `println! { "Hello, Rust!" }`
  = note: to `print ! ( concat ! ( "Hello, Rust!" , "\n" ) )`
  = note: expanding `print! { concat ! ( "Hello, Rust!" , "\n" ) }`
  = note: to `$crate :: io :: _print ( format_args ! ( concat ! ( "Hello, Rust!" , "\n" ) )
          )`

    Finished dev [unoptimized + debuginfo] target(s) in 0.60 secs

trusted_len

The tracking issue for this feature is: #37572


trusted_random_access

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


try_find

The tracking issue for this feature is: #63178


try_reserve

The tracking issue for this feature is: #48043


try_trait

The tracking issue for this feature is: #42327


This introduces a new trait Try for extending the ? operator to types other than Result (a part of RFC 1859). The trait provides the canonical way to view a type in terms of a success/failure dichotomy. This will allow ? to supplant the try_opt! macro on Option and the try_ready! macro on Poll, among other things.

Here's an example implementation of the trait:

/// A distinct type to represent the `None` value of an `Option`.
///
/// This enables using the `?` operator on `Option`; it's rarely useful alone.
#[derive(Debug)]
#[unstable(feature = "try_trait", issue = "42327")]
pub struct None { _priv: () }

#[unstable(feature = "try_trait", issue = "42327")]
impl<T> ops::Try for Option<T>  {
    type Ok = T;
    type Error = None;

    fn into_result(self) -> Result<T, None> {
        self.ok_or(None { _priv: () })
    }

    fn from_ok(v: T) -> Self {
        Some(v)
    }

    fn from_error(_: None) -> Self {
        None
    }
}

Note the Error associated type here is a new marker. The ? operator allows interconversion between different Try implementers only when the error type can be converted Into the error type of the enclosing function (or catch block). Having a distinct error type (as opposed to just (), or similar) restricts this to where it's semantically meaningful.

type_name_of_val

The tracking issue for this feature is: #66359


unchecked_math

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


unicode_internals

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


unix_socket_peek

The tracking issue for this feature is: #76923


unsafe_cell_get_mut

The tracking issue for this feature is: #76943


unsafe_cell_raw_get

The tracking issue for this feature is: #66358


unsigned_abs

The tracking issue for this feature is: #74913


unsize

The tracking issue for this feature is: #27732


unwrap_infallible

The tracking issue for this feature is: #61695


update_panic_count

This feature is internal to the Rust compiler and is not intended for general use.


variant_count

The tracking issue for this feature is: #73662


vec_into_raw_parts

The tracking issue for this feature is: #65816


vec_remove_item

The tracking issue for this feature is: #40062


vec_spare_capacity

The tracking issue for this feature is: #75017


vecdeque_binary_search

The tracking issue for this feature is: #78021


wake_trait

The tracking issue for this feature is: #69912


wasi_ext

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


windows_by_handle

The tracking issue for this feature is: #63010


windows_c

This feature is internal to the Rust compiler and is not intended for general use.


windows_file_type_ext

This feature has no tracking issue, and is therefore likely internal to the compiler, not being intended for general use.


windows_handle

This feature is internal to the Rust compiler and is not intended for general use.


windows_net

This feature is internal to the Rust compiler and is not intended for general use.


windows_stdio

This feature is internal to the Rust compiler and is not intended for general use.


with_options

The tracking issue for this feature is: #65439


wrapping_int_impl

The tracking issue for this feature is: #32463


wrapping_next_power_of_two

The tracking issue for this feature is: #32463


write_all_vectored

The tracking issue for this feature is: #70436