17 KiB
title | author | date |
---|---|---|
Rust | Matthias Beyer | 2017-03-05 |
Rust
To install the rust compiler and cargo put
rustc
cargo
into the environment.systemPackages
or bring them into
scope with nix-shell -p rustc cargo
.
For daily builds (beta and nightly) use either rustup from nixpkgs or use the Rust nightlies overlay.
Compiling Rust applications with Cargo
Rust applications are packaged by using the buildRustPackage
helper from rustPlatform
:
rustPlatform.buildRustPackage rec {
pname = "ripgrep";
version = "11.0.2";
src = fetchFromGitHub {
owner = "BurntSushi";
repo = pname;
rev = version;
sha256 = "1iga3320mgi7m853la55xip514a3chqsdi1a1rwv25lr9b1p7vd3";
};
cargoSha256 = "17ldqr3asrdcsh4l29m3b5r37r5d0b3npq1lrgjmxb6vlx6a36qh";
meta = with stdenv.lib; {
description = "A fast line-oriented regex search tool, similar to ag and ack";
homepage = "https://github.com/BurntSushi/ripgrep";
license = licenses.unlicense;
maintainers = [ maintainers.tailhook ];
};
}
buildRustPackage
requires a cargoSha256
attribute which is computed over
all crate sources of this package. Currently it is obtained by inserting a
fake checksum into the expression and building the package once. The correct
checksum can then be taken from the failed build.
Per the instructions in the Cargo Book
best practices guide, Rust applications should always commit the Cargo.lock
file in git to ensure a reproducible build. However, a few packages do not, and
Nix depends on this file, so if it missing you can use cargoPatches
to apply
it in the patchPhase
. Consider sending a PR upstream with a note to the
maintainer describing why it's important to include in the application.
The fetcher will verify that the Cargo.lock
file is in sync with the src
attribute, and fail the build if not. It will also will compress the vendor
directory into a tar.gz archive.
Cross compilation
By default, Rust packages are compiled for the host platform, just like any
other package is. The --target
passed to rust tools is computed from this.
By default, it takes the stdenv.hostPlatform.config
and replaces components
where they are known to differ. But there are ways to customize the argument:
-
To choose a different target by name, define
stdenv.hostPlatform.rustc.arch.config
as that name (a string), and that name will be used instead.For example:
import <nixpkgs> { crossSystem = (import <nixpkgs/lib>).systems.examples.armhf-embedded // { rustc.arch.config = "thumbv7em-none-eabi"; }; }
will result in:
--target thumbv7em-none-eabi
-
To pass a completely custom target, define
stdenv.hostPlatform.rustc.arch.config
with its name, andstdenv.hostPlatform.rustc.arch.custom
with the value. The value will be serialized to JSON in a file called${stdenv.hostPlatform.rustc.arch.config}.json
, and the path of that file will be used instead.For example:
import <nixpkgs> { crossSystem = (import <nixpkgs/lib>).systems.examples.armhf-embedded // { rustc.arch.config = "thumb-crazy"; rustc.arch.custom = { foo = ""; bar = ""; }; }; } will result in: ```shell --target /nix/store/asdfasdfsadf-thumb-crazy.json # contains {"foo":"","bar":""}
Finally, as an ad-hoc escape hatch, a computed target (string or JSON file
path) can be passed directly to buildRustPackage
:
pkgs.rustPlatform.buildRustPackage {
(...)
target = "x86_64-fortanix-unknown-sgx";
}
This is useful to avoid rebuilding Rust tools, since they are actually target
agnostic and don't need to be rebuilt. But in the future, we should always
build the Rust tools and standard library crates separately so there is no
reason not to take the stdenv.hostPlatform.rustc
-modifying approach, and the
ad-hoc escape hatch to buildRustPackage
can be removed.
Note that currently custom targets aren't compiled with std
, so cargo test
will fail. This can be ignored by adding doCheck = false;
to your derivation.
Running package tests
When using buildRustPackage
, the checkPhase
is enabled by default and runs
cargo test
on the package to build. To make sure that we don't compile the
sources twice and to actually test the artifacts that will be used at runtime,
the tests will be ran in the release
mode by default.
However, in some cases the test-suite of a package doesn't work properly in the
release
mode. For these situations, the mode for checkPhase
can be changed like
so:
rustPlatform.buildRustPackage {
/* ... */
checkType = "debug";
}
Please note that the code will be compiled twice here: once in release
mode
for the buildPhase
, and again in debug
mode for the checkPhase
.
Tests relying on the structure of the target/
directory
Some tests may rely on the structure of the target/
directory. Those tests
are likely to fail because we use cargo --target
during the build. This means that
the artifacts
are stored in target/<architecture>/release/
,
rather than in target/release/
.
This can only be worked around by patching the affected tests accordingly.
Disabling package-tests
In some instances, it may be necessary to disable testing altogether (with doCheck = false;
):
- If no tests exist -- the
checkPhase
should be explicitly disabled to skip unnecessary build steps to speed up the build. - If tests are highly impure (e.g. due to network usage).
There will obviously be some corner-cases not listed above where it's sensible to disable tests. The above are just guidelines, and exceptions may be granted on a case-by-case basis.
However, please check if it's possible to disable a problematic subset of the test suite and leave a comment explaining your reasoning.
Setting test-threads
buildRustPackage
will use parallel test threads by default,
sometimes it may be necessary to disable this so the tests run consecutively.
rustPlatform.buildRustPackage {
/* ... */
cargoParallelTestThreads = false;
}
Building a package in debug
mode
By default, buildRustPackage
will use release
mode for builds. If a package
should be built in debug
mode, it can be configured like so:
rustPlatform.buildRustPackage {
/* ... */
buildType = "debug";
}
In this scenario, the checkPhase
will be ran in debug
mode as well.
Custom build
/install
-procedures
Some packages may use custom scripts for building/installing, e.g. with a Makefile
.
In these cases, it's recommended to override the buildPhase
/installPhase
/checkPhase
.
Otherwise, some steps may fail because of the modified directory structure of target/
.
Building a crate with an absent or out-of-date Cargo.lock file
buildRustPackage
needs a Cargo.lock
file to get all dependencies in the
source code in a reproducible way. If it is missing or out-of-date one can use
the cargoPatches
attribute to update or add it.
{ lib, rustPlatform, fetchFromGitHub }:
rustPlatform.buildRustPackage rec {
(...)
cargoPatches = [
# a patch file to add/update Cargo.lock in the source code
./add-Cargo.lock.patch
];
}
Compiling Rust crates using Nix instead of Cargo
Simple operation
When run, cargo build
produces a file called Cargo.lock
,
containing pinned versions of all dependencies. Nixpkgs contains a
tool called carnix
(nix-env -iA nixos.carnix
), which can be used
to turn a Cargo.lock
into a Nix expression.
That Nix expression calls rustc
directly (hence bypassing Cargo),
and can be used to compile a crate and all its dependencies. Here is
an example for a minimal hello
crate:
$ cargo new hello
$ cd hello
$ cargo build
Compiling hello v0.1.0 (file:///tmp/hello)
Finished dev [unoptimized + debuginfo] target(s) in 0.20 secs
$ carnix -o hello.nix --src ./. Cargo.lock --standalone
$ nix-build hello.nix -A hello_0_1_0
Now, the file produced by the call to carnix
, called hello.nix
, looks like:
# Generated by carnix 0.6.5: carnix -o hello.nix --src ./. Cargo.lock --standalone
{ lib, stdenv, buildRustCrate, fetchgit }:
let kernel = stdenv.buildPlatform.parsed.kernel.name;
# ... (content skipped)
in
rec {
hello = f: hello_0_1_0 { features = hello_0_1_0_features { hello_0_1_0 = f; }; };
hello_0_1_0_ = { dependencies?[], buildDependencies?[], features?[] }: buildRustCrate {
crateName = "hello";
version = "0.1.0";
authors = [ "pe@pijul.org <pe@pijul.org>" ];
src = ./.;
inherit dependencies buildDependencies features;
};
hello_0_1_0 = { features?(hello_0_1_0_features {}) }: hello_0_1_0_ {};
hello_0_1_0_features = f: updateFeatures f (rec {
hello_0_1_0.default = (f.hello_0_1_0.default or true);
}) [ ];
}
In particular, note that the argument given as --src
is copied
verbatim to the source. If we look at a more complicated
dependencies, for instance by adding a single line libc="*"
to our
Cargo.toml
, we first need to run cargo build
to update the
Cargo.lock
. Then, carnix
needs to be run again, and produces the
following nix file:
# Generated by carnix 0.6.5: carnix -o hello.nix --src ./. Cargo.lock --standalone
{ lib, stdenv, buildRustCrate, fetchgit }:
let kernel = stdenv.buildPlatform.parsed.kernel.name;
# ... (content skipped)
in
rec {
hello = f: hello_0_1_0 { features = hello_0_1_0_features { hello_0_1_0 = f; }; };
hello_0_1_0_ = { dependencies?[], buildDependencies?[], features?[] }: buildRustCrate {
crateName = "hello";
version = "0.1.0";
authors = [ "pe@pijul.org <pe@pijul.org>" ];
src = ./.;
inherit dependencies buildDependencies features;
};
libc_0_2_36_ = { dependencies?[], buildDependencies?[], features?[] }: buildRustCrate {
crateName = "libc";
version = "0.2.36";
authors = [ "The Rust Project Developers" ];
sha256 = "01633h4yfqm0s302fm0dlba469bx8y6cs4nqc8bqrmjqxfxn515l";
inherit dependencies buildDependencies features;
};
hello_0_1_0 = { features?(hello_0_1_0_features {}) }: hello_0_1_0_ {
dependencies = mapFeatures features ([ libc_0_2_36 ]);
};
hello_0_1_0_features = f: updateFeatures f (rec {
hello_0_1_0.default = (f.hello_0_1_0.default or true);
libc_0_2_36.default = true;
}) [ libc_0_2_36_features ];
libc_0_2_36 = { features?(libc_0_2_36_features {}) }: libc_0_2_36_ {
features = mkFeatures (features.libc_0_2_36 or {});
};
libc_0_2_36_features = f: updateFeatures f (rec {
libc_0_2_36.default = (f.libc_0_2_36.default or true);
libc_0_2_36.use_std =
(f.libc_0_2_36.use_std or false) ||
(f.libc_0_2_36.default or false) ||
(libc_0_2_36.default or false);
}) [];
}
Here, the libc
crate has no src
attribute, so buildRustCrate
will fetch it from crates.io. A sha256
attribute is still needed for Nix purity.
Handling external dependencies
Some crates require external libraries. For crates from
crates.io, such libraries can be specified in
defaultCrateOverrides
package in nixpkgs itself.
Starting from that file, one can add more overrides, to add features or build inputs by overriding the hello crate in a seperate file.
with import <nixpkgs> {};
((import ./hello.nix).hello {}).override {
crateOverrides = defaultCrateOverrides // {
hello = attrs: { buildInputs = [ openssl ]; };
};
}
Here, crateOverrides
is expected to be a attribute set, where the
key is the crate name without version number and the value a function.
The function gets all attributes passed to buildRustCrate
as first
argument and returns a set that contains all attribute that should be
overwritten.
For more complicated cases, such as when parts of the crate's
derivation depend on the crate's version, the attrs
argument of
the override above can be read, as in the following example, which
patches the derivation:
with import <nixpkgs> {};
((import ./hello.nix).hello {}).override {
crateOverrides = defaultCrateOverrides // {
hello = attrs: lib.optionalAttrs (lib.versionAtLeast attrs.version "1.0") {
postPatch = ''
substituteInPlace lib/zoneinfo.rs \
--replace "/usr/share/zoneinfo" "${tzdata}/share/zoneinfo"
'';
};
};
}
Another situation is when we want to override a nested
dependency. This actually works in the exact same way, since the
crateOverrides
parameter is forwarded to the crate's
dependencies. For instance, to override the build inputs for crate
libc
in the example above, where libc
is a dependency of the main
crate, we could do:
with import <nixpkgs> {};
((import hello.nix).hello {}).override {
crateOverrides = defaultCrateOverrides // {
libc = attrs: { buildInputs = []; };
};
}
Options and phases configuration
Actually, the overrides introduced in the previous section are more general. A number of other parameters can be overridden:
-
The version of rustc used to compile the crate:
(hello {}).override { rust = pkgs.rust; };
-
Whether to build in release mode or debug mode (release mode by default):
(hello {}).override { release = false; };
-
Whether to print the commands sent to rustc when building (equivalent to
--verbose
in cargo:(hello {}).override { verbose = false; };
-
Extra arguments to be passed to
rustc
:(hello {}).override { extraRustcOpts = "-Z debuginfo=2"; };
-
Phases, just like in any other derivation, can be specified using the following attributes:
preUnpack
,postUnpack
,prePatch
,patches
,postPatch
,preConfigure
(in the case of a Rust crate, this is run before calling the "build" script),postConfigure
(after the "build" script),preBuild
,postBuild
,preInstall
andpostInstall
. As an example, here is how to create a new module before running the build script:(hello {}).override { preConfigure = '' echo "pub const PATH=\"${hi.out}\";" >> src/path.rs" ''; };
Features
One can also supply features switches. For example, if we want to
compile diesel_cli
only with the postgres
feature, and no default
features, we would write:
(callPackage ./diesel.nix {}).diesel {
default = false;
postgres = true;
}
Where diesel.nix
is the file generated by Carnix, as explained above.
Setting Up nix-shell
Oftentimes you want to develop code from within nix-shell
. Unfortunately
buildRustCrate
does not support common nix-shell
operations directly
(see this issue)
so we will use stdenv.mkDerivation
instead.
Using the example hello
project above, we want to do the following:
- Have access to
cargo
andrustc
- Have the
openssl
library available to a crate through it's normal compilation mechanism (pkg-config
).
A typical shell.nix
might look like:
with import <nixpkgs> {};
stdenv.mkDerivation {
name = "rust-env";
nativeBuildInputs = [
rustc cargo
# Example Build-time Additional Dependencies
pkgconfig
];
buildInputs = [
# Example Run-time Additional Dependencies
openssl
];
# Set Environment Variables
RUST_BACKTRACE = 1;
}
You should now be able to run the following:
$ nix-shell --pure
$ cargo build
$ cargo test
Controlling Rust Version Inside nix-shell
To control your rust version (i.e. use nightly) from within shell.nix
(or
other nix expressions) you can use the following shell.nix
# Latest Nightly
with import <nixpkgs> {};
let src = fetchFromGitHub {
owner = "mozilla";
repo = "nixpkgs-mozilla";
# commit from: 2019-05-15
rev = "9f35c4b09fd44a77227e79ff0c1b4b6a69dff533";
sha256 = "18h0nvh55b5an4gmlgfbvwbyqj91bklf1zymis6lbdh75571qaz0";
};
in
with import "${src.out}/rust-overlay.nix" pkgs pkgs;
stdenv.mkDerivation {
name = "rust-env";
buildInputs = [
# Note: to use use stable, just replace `nightly` with `stable`
latest.rustChannels.nightly.rust
# Add some extra dependencies from `pkgs`
pkgconfig openssl
];
# Set Environment Variables
RUST_BACKTRACE = 1;
}
Now run:
$ rustc --version
rustc 1.26.0-nightly (188e693b3 2018-03-26)
To see that you are using nightly.
Using the Rust nightlies overlay
Mozilla provides an overlay for nixpkgs to bring a nightly version of Rust into scope. This overlay can also be used to install recent unstable or stable versions of Rust, if desired.
To use this overlay, clone
nixpkgs-mozilla,
and create a symbolic link to the file
rust-overlay.nix
in the ~/.config/nixpkgs/overlays
directory.
$ git clone https://github.com/mozilla/nixpkgs-mozilla.git
$ mkdir -p ~/.config/nixpkgs/overlays
$ ln -s $(pwd)/nixpkgs-mozilla/rust-overlay.nix ~/.config/nixpkgs/overlays/rust-overlay.nix
The latest version can be installed with the following command:
$ nix-env -Ai nixos.latest.rustChannels.stable.rust
Or using the attribute with nix-shell:
$ nix-shell -p nixos.latest.rustChannels.stable.rust
To install the beta or nightly channel, "stable" should be substituted by "nightly" or "beta", or use the function provided by this overlay to pull a version based on a build date.
The overlay automatically updates itself as it uses the same source as rustup.