Unstable Features

Experimental Cargo features are only available on the nightly channel. You typically use one of the -Z flags to enable them. Run cargo -Z help to see a list of flags available.

-Z unstable-options is a generic flag for enabling other unstable command-line flags. Options requiring this will be called out below.

Some unstable features will require you to specify the cargo-features key in Cargo.toml.


The -Z no-index-update flag ensures that Cargo does not attempt to update the registry index. This is intended for tools such as Crater that issue many Cargo commands, and you want to avoid the network latency for updating the index each time.


  • Original Issue: #6477
  • Cache usage meta tracking issue: #7150

The -Z mtime-on-use flag is an experiment to have Cargo update the mtime of used files to make it easier for tools like cargo-sweep to detect which files are stale. For many workflows this needs to be set on all invocations of cargo. To make this more practical setting the unstable.mtime_on_use flag in .cargo/config or the corresponding ENV variable will apply the -Z mtime-on-use to all invocations of nightly cargo. (the config flag is ignored by stable)


  • Original Issue: #4988
  • Stabilization Issue: #5133

When running commands such as cargo install or cargo build, Cargo currently requires dev-dependencies to be downloaded, even if they are not used. The -Z avoid-dev-deps flag allows Cargo to avoid downloading dev-dependencies if they are not needed. The Cargo.lock file will not be generated if dev-dependencies are skipped.


Note: It is not recommended to use this feature. Because it enforces minimal versions for all transitive dependencies, its usefulness is limited since not all external dependencies declare proper lower version bounds. It is intended that it will be changed in the future to only enforce minimal versions for direct dependencies.

When a Cargo.lock file is generated, the -Z minimal-versions flag will resolve the dependencies to the minimum semver version that will satisfy the requirements (instead of the greatest version).

The intended use-case of this flag is to check, during continuous integration, that the versions specified in Cargo.toml are a correct reflection of the minimum versions that you are actually using. That is, if Cargo.toml says foo = "1.0.0" that you don't accidentally depend on features added only in foo 1.5.0.


This feature allows you to specify the directory where artifacts will be copied to after they are built. Typically artifacts are only written to the target/release or target/debug directories. However, determining the exact filename can be tricky since you need to parse JSON output. The --out-dir flag makes it easier to predictably access the artifacts. Note that the artifacts are copied, so the originals are still in the target directory. Example:

cargo +nightly build --out-dir=out -Z unstable-options

This can also be specified in .cargo/config files.

out-dir = "out"


This flag changes cargo test's behavior when handling doctests when a target is passed. Currently, if a target is passed that is different from the host cargo will simply skip testing doctests. If this flag is present, cargo will continue as normal, passing the tests to doctest, while also passing it a --target option, as well as enabling -Zunstable-features --enable-per-target-ignores and passing along information from .cargo/config. See the rustc issue for more information.

cargo test --target foo -Zdoctest-xcompile

Custom named profiles

With this feature you can define custom profiles having new names. With the custom profile enabled, build artifacts can be emitted by default to directories other than release or debug, based on the custom profile's name.

For example:

cargo-features = ["named-profiles"]

inherits = "release"
lto = true

An inherits key is used in order to receive attributes from other profiles, so that a new custom profile can be based on the standard dev or release profile presets. Cargo emits errors in case inherits loops are detected. When considering inheritance hierarchy, all profiles directly or indirectly inherit from either from release or from dev.

Valid profile names are: must not be empty, use only alphanumeric characters or - or _.

Passing --profile with the profile's name to various Cargo commands, directs operations to use the profile's attributes. Overrides that are specified in the profiles from which the custom profile inherits are inherited too.

For example, using cargo build with --profile and the manifest from above:

cargo +nightly build --profile release-lto -Z unstable-options

When a custom profile is used, build artifcats go to a different target by default. In the example above, you can expect to see the outputs under target/release-lto.

New dir-name attribute

Some of the paths generated under target/ have resulted in a de-facto "build protocol", where cargo is invoked as a part of a larger project build. So, to preserve the existing behavior, there is also a new attribute dir-name, which when left unspecified, defaults to the name of the profile. For example:

inherits = "release"
dir-name = "lto"  # Emits to target/lto instead of target/release-lto
lto = true

Namespaced features

Currently, it is not possible to have a feature and a dependency with the same name in the manifest. If you set namespaced-features to true, the namespaces for features and dependencies are separated. The effect of this is that, in the feature requirements, dependencies have to be prefixed with crate:. Like this:

namespaced-features = true

bar = ["crate:baz", "foo"]
foo = []

baz = { version = "0.1", optional = true }

To prevent unnecessary boilerplate from having to explicitly declare features for each optional dependency, implicit features get created for any optional dependencies where a feature of the same name is not defined. However, if a feature of the same name as a dependency is defined, that feature must include the dependency as a requirement, as foo = ["crate:foo"].


The --build-plan argument for the build command will output JSON with information about which commands would be run without actually executing anything. This can be useful when integrating with another build tool. Example:

cargo +nightly build --build-plan -Z unstable-options


Metabuild is a feature to have declarative build scripts. Instead of writing a build.rs script, you specify a list of build dependencies in the metabuild key in Cargo.toml. A build script is automatically generated that runs each build dependency in order. Metabuild packages can then read metadata from Cargo.toml to specify their behavior.

Include cargo-features at the top of Cargo.toml, a metabuild key in the package, list the dependencies in build-dependencies, and add any metadata that the metabuild packages require under package.metadata. Example:

cargo-features = ["metabuild"]

name = "mypackage"
version = "0.0.1"
metabuild = ["foo", "bar"]

foo = "1.0"
bar = "1.0"

extra-info = "qwerty"

Metabuild packages should have a public function called metabuild that performs the same actions as a regular build.rs script would perform.


The 'public-dependency' feature allows marking dependencies as 'public' or 'private'. When this feature is enabled, additional information is passed to rustc to allow the 'exported_private_dependencies' lint to function properly.

This requires the appropriate key to be set in cargo-features:

cargo-features = ["public-dependency"]

my_dep = { version = "1.2.3", public = true }
private_dep = "2.0.0" # Will be 'private' by default


  • Tracking Repository: https://github.com/rust-lang/wg-cargo-std-aware

The build-std feature enables Cargo to compile the standard library itself as part of a crate graph compilation. This feature has also historically been known as "std-aware Cargo". This feature is still in very early stages of development, and is also a possible massive feature addition to Cargo. This is a very large feature to document, even in the minimal form that it exists in today, so if you're curious to stay up to date you'll want to follow the tracking repository and its set of issues.

The functionality implemented today is behind a flag called -Z build-std. This flag indicates that Cargo should compile the standard library from source code using the same profile as the main build itself. Note that for this to work you need to have the source code for the standard library available, and at this time the only supported method of doing so is to add the rust-src rust rustup component:

$ rustup component add rust-src --toolchain nightly

It is also required today that the -Z build-std flag is combined with the --target flag. Note that you're not forced to do a cross compilation, you're just forced to pass --target in one form or another.

Usage looks like:

$ cargo new foo
$ cd foo
$ cargo +nightly run -Z build-std --target x86_64-unknown-linux-gnu
   Compiling core v0.0.0 (...)
   Compiling foo v0.1.0 (...)
    Finished dev [unoptimized + debuginfo] target(s) in 21.00s
     Running `target/x86_64-unknown-linux-gnu/debug/foo`
Hello, world!

Here we recompiled the standard library in debug mode with debug assertions (like src/main.rs is compiled) and everything was linked together at the end.

Using -Z build-std will implicitly compile the stable crates core, std, alloc, and proc_macro. If you're using cargo test it will also compile the test crate. If you're working with an environment which does not support some of these crates, then you can pass an argument to -Zbuild-std as well:

$ cargo +nightly build -Z build-std=core,alloc

The value here is a comma-separated list of standard library crates to build.


As a summary, a list of requirements today to use -Z build-std are:

  • You must install libstd's source code through rustup component add rust-src
  • You must pass --target
  • You must use both a nightly Cargo and a nightly rustc
  • The -Z build-std flag must be passed to all cargo invocations.

Reporting bugs and helping out

The -Z build-std feature is in the very early stages of development! This feature for Cargo has an extremely long history and is very large in scope, and this is just the beginning. If you'd like to report bugs please either report them to:

  • Cargo - https://github.com/rust-lang/cargo/issues/new - for implementation bugs
  • The tracking repository - https://github.com/rust-lang/wg-cargo-std-aware/issues/new - for larger design questions.

Also if you'd like to see a feature that's not yet implemented and/or if something doesn't quite work the way you'd like it to, feel free to check out the issue tracker of the tracking repository, and if it's not there please file a new issue!


The timings feature gives some information about how long each compilation takes, and tracks concurrency information over time.

cargo +nightly build -Z timings

The -Ztimings flag can optionally take a comma-separated list of the following values:

  • html — Saves a file called cargo-timing.html to the current directory with a report of the compilation. Files are also saved with a timestamp in the filename if you want to look at older runs.
  • info — Displays a message to stdout after each compilation finishes with how long it took.
  • json — Emits some JSON information about timing information.

The default if none are specified is html,info.

Reading the graphs

There are two graphs in the output. The "unit" graph shows the duration of each unit over time. A "unit" is a single compiler invocation. There are lines that show which additional units are "unlocked" when a unit finishes. That is, it shows the new units that are now allowed to run because their dependencies are all finished. Hover the mouse over a unit to highlight the lines. This can help visualize the critical path of dependencies. This may change between runs because the units may finish in different orders.

The "codegen" times are highlighted in a lavender color. In some cases, build pipelining allows units to start when their dependencies are performing code generation. This information is not always displayed (for example, binary units do not show when code generation starts).

The "custom build" units are build.rs scripts, which when run are highlighted in orange.

The second graph shows Cargo's concurrency over time. The three lines are:

  • "Waiting" (red) — This is the number of units waiting for a CPU slot to open.
  • "Inactive" (blue) — This is the number of units that are waiting for their dependencies to finish.
  • "Active" (green) — This is the number of units currently running.

Note: This does not show the concurrency in the compiler itself. rustc coordinates with Cargo via the "job server" to stay within the concurrency limit. This currently mostly applies to the code generation phase.

Tips for addressing compile times:

  • Look for slow dependencies.
    • Check if they have features that you may wish to consider disabling.
    • Consider trying to remove the dependency completely.
  • Look for a crate being built multiple times with different versions. Try to remove the older versions from the dependency graph.
  • Split large crates into smaller pieces.
  • If there are a large number of crates bottlenecked on a single crate, focus your attention on improving that one crate to improve parallelism.


  • Tracking rustc issue: #63012

The -Z binary-dep-depinfo flag causes Cargo to forward the same flag to rustc which will then cause rustc to include the paths of all binary dependencies in the "dep info" file (with the .d extension). Cargo then uses that information for change-detection (if any binary dependency changes, then the crate will be rebuilt). The primary use case is for building the compiler itself, which has implicit dependencies on the standard library that would otherwise be untracked for change-detection.


The -Z panic-abort-tests flag will enable nightly support to compile test harness crates with -Cpanic=abort. Without this flag Cargo will compile tests, and everything they depend on, with -Cpanic=unwind because it's the only way test-the-crate knows how to operate. As of rust-lang/rust#64158, however, the test crate supports -C panic=abort with a test-per-process, and can help avoid compiling crate graphs multiple times.

It's currently unclear how this feature will be stabilized in Cargo, but we'd like to stabilize it somehow!


The --config CLI option allows arbitrary config values to be passed in via the command-line. The argument should be in TOML syntax of KEY=VALUE:

cargo +nightly -Zunstable-options --config net.git-fetch-with-cli=true fetch

The --config option may be specified multiple times, in which case the values are merged in left-to-right order, using the same merging logic that multiple config files use. CLI values take precedence over environment variables, which take precedence over config files.

Some examples of what it looks like using Bourne shell syntax:

# Most shells will require escaping.
cargo --config http.proxy=\"http://example.com\" …

# Spaces may be used.
cargo --config "net.git-fetch-with-cli = true" …

# TOML array example. Single quotes make it easier to read and write.
cargo --config 'build.rustdocflags = ["--html-in-header", "header.html"]' …

# Example of a complex TOML key.
cargo --config "target.'cfg(all(target_arch = \"arm\", target_os = \"none\"))'.runner = 'my-runner'" …

# Example of overriding a profile setting.
cargo --config profile.dev.package.image.opt-level=3 …


The include key in a config file can be used to load another config file. It takes a string for a path to another file relative to the config file, or a list of strings. It requires the -Zconfig-include command-line option.

# .cargo/config
include = '../../some-common-config.toml'

The config values are first loaded from the include path, and then the config file's own values are merged on top of it.

This can be paired with config-cli to specify a file to load from the command-line. Pass a path to a config file as the argument to --config:

cargo +nightly -Zunstable-options -Zconfig-include --config somefile.toml build

CLI paths are relative to the current working directory.


The -Zfeatures option causes Cargo to use a new feature resolver that can resolve features differently from before. It takes a comma separated list of options to indicate which new behaviors to enable. With no options, it should behave the same as without the flag.

cargo +nightly -Zfeatures=itarget,build_dep

The available options are:

  • itarget — Ignores features for target-specific dependencies for targets that don't match the current compile target. For example:

    version = "1.0"
    features = ["f1"]
    version = "1.0"
    features = ["f2"]

    When building this example for a non-Windows platform, the f2 feature will not be enabled.

  • build_dep — Prevents features enabled on build dependencies from being enabled for normal dependencies. For example:

    log = "0.4"
    log = {version = "0.4", features=['std']}

    When building the build script, the log crate will be built with the std feature. When building the library of your package, it will not enable the feature.

  • dev_dep — Prevents features enabled on dev dependencies from being enabled for normal dependencies. For example:

    serde = {version = "1.0", default-features = false}
    serde = {version = "1.0", features = ["std"]}

    In this example, the library will normally link against serde without the std feature. However, when built as a test or example, it will include the std feature.

    This mode is ignored if you are building any test, bench, or example. That is, dev dependency features will still be unified if you run commands like cargo test or cargo build --all-targets.

  • all — Enable all feature options (itarget,build_dep,dev_dep).

  • compare — This option compares the resolved features to the old resolver, and will print any differences.


The -Z crate-versions flag will make cargo doc include appropriate crate versions for the current crate and all of its dependencies (unless --no-deps was provided) in the compiled documentation.

You can find an example screenshot for the cargo itself in the tracking issue.