If you want to hack on Miri yourself, great! Here are some resources you might find useful.
Check out the issues on this GitHub repository for some ideas. In particular,
look for the green E-*
labels which mark issues that should be rather
well-suited for onboarding. For more ideas or help with hacking on Miri, you can
contact us (oli-obk
and RalfJ
) on the Rust Zulip.
Miri heavily relies on internal and unstable rustc interfaces to execute MIR, which means it is important that you install a version of rustc that Miri actually works with.
The rust-version
file contains the commit hash of rustc that Miri is currently
tested against. Other versions will likely not work. After installing
rustup-toolchain-install-master
, you can run the following command to
install that exact version of rustc as a toolchain:
./miri toolchain
This will set up a rustup toolchain called miri
and set it as an override for
the current directory.
You can also create a .auto-everything
file (contents don't matter, can be empty), which
will cause any ./miri
command to automatically call ./miri toolchain
, clippy
and rustfmt
for you. If you don't want all of these to happen, you can add individual .auto-toolchain
,
.auto-clippy
and .auto-fmt
files respectively.
Invoking Miri requires getting a bunch of flags right and setting up a custom
sysroot. The miri
script takes care of that for you. With the
build environment prepared, compiling Miri is just one command away:
./miri build
Run ./miri
without arguments to see the other commands our build tool
supports.
The Miri driver compiled from src/bin/miri.rs
is the "heart" of Miri: it is
basically a version of rustc
that, instead of compiling your code, runs it.
It accepts all the same flags as rustc
(though the ones only affecting code
generation and linking obviously will have no effect) and more.
For example, you can (cross-)run the driver on a particular file by doing
./miri run tests/pass/format.rs
./miri run tests/pass/hello.rs --target i686-unknown-linux-gnu
Tests in pass-dep
need to be run using ./miri run --dep <filename>
.
For example:
./miri run --dep tests/pass-dep/shims/libc-fs.rs
You can (cross-)run the entire test suite using:
./miri test
./miri test --target i686-unknown-linux-gnu
./miri test FILTER
only runs those tests that contain FILTER
in their filename (including the
base directory, e.g. ./miri test fail
will run all compile-fail tests). Multiple filters are
supported: ./miri test FILTER1 FILTER2
runs all tests that contain either string.
You can get a trace of which MIR statements are being executed by setting the
MIRI_LOG
environment variable. For example:
MIRI_LOG=info ./miri run tests/pass/vec.rs
Setting MIRI_LOG
like this will configure logging for Miri itself as well as
the rustc_middle::mir::interpret
and rustc_mir::interpret
modules in rustc. You
can also do more targeted configuration, e.g. the following helps debug the
stacked borrows implementation:
MIRI_LOG=rustc_mir::interpret=info,miri::stacked_borrows ./miri run tests/pass/vec.rs
Note that you will only get info
, warn
or error
messages if you use a prebuilt compiler.
In order to get debug
and trace
level messages, you need to build miri with a locally built
compiler that has debug=true
set in config.toml
.
You can set MIRI_BACKTRACE=1
to get a backtrace of where an
evaluation error was originally raised.
We use ui-testing in Miri, meaning we generate .stderr
and .stdout
files for the output
produced by Miri. You can use ./miri test --bless
to automatically (re)generate these files when
you add new tests or change how Miri presents certain output.
Note that when you also use MIRIFLAGS
to change optimizations and similar, the ui output
will change in unexpected ways. In order to still be able
to run the other checks while ignoring the ui output, use MIRI_SKIP_UI_CHECKS=1 ./miri test
.
For more info on how to configure ui tests see the documentation on the ui test crate
Working with the driver directly gives you full control, but you also lose all the convenience provided by cargo. Once your test case depends on a crate, it is probably easier to test it with the cargo wrapper. You can install your development version of Miri using
./miri install
and then you can use it as if it was installed by rustup
as a component of the
miri
toolchain. Note that the miri
and cargo-miri
executables are placed
in the miri
toolchain's sysroot to prevent conflicts with other toolchains.
The Miri binaries in the cargo
bin directory (usually ~/.cargo/bin
) are managed by rustup.
There's a test for the cargo wrapper in the test-cargo-miri
directory; run ./run-test.py
in
there to execute it. You can pass --target
to execute the test for another target.
Miri re-builds the standard library into a custom sysroot, so it is fairly easy
to test Miri against a modified standard library -- you do not even have to
build Miri yourself, the Miri shipped by rustup
will work. All you have to do
is set the MIRI_LIB_SRC
environment variable to the library
folder of a
rust-lang/rust
repository checkout. Note that changing files in that directory
does not automatically trigger a re-build of the standard library; you have to
clear the Miri build cache manually (on Linux, rm -rf ~/.cache/miri
;
on Windows, rmdir /S "%LOCALAPPDATA%\rust-lang\miri\cache"
;
and on macOS, rm -rf ~/Library/Caches/org.rust-lang.miri
).
Miri comes with a few benchmarks; you can run ./miri bench
to run them with the locally built
Miri. Note: this will run ./miri install
as a side-effect. Also requires hyperfine
to be
installed (cargo install hyperfine
).
To configure rust-analyzer
and VS Code for working on Miri, save the following
to .vscode/settings.json
in your local Miri clone:
{
"rust-analyzer.rustc.source": "discover",
"rust-analyzer.linkedProjects": [
"Cargo.toml",
"cargo-miri/Cargo.toml",
"miri-script/Cargo.toml",
],
"rust-analyzer.check.overrideCommand": [
"env",
"MIRI_AUTO_OPS=no",
"./miri",
"cargo",
"clippy", // make this `check` when working with a locally built rustc
"--message-format=json",
"--all-targets",
],
// Contrary to what the name suggests, this also affects proc macros.
"rust-analyzer.cargo.buildScripts.overrideCommand": [
"env",
"MIRI_AUTO_OPS=no",
"./miri",
"cargo",
"check",
"--message-format=json",
"--all-targets",
],
}
If you are [building Miri with a locally built rustc][], set
rust-analyzer.rustcSource
to the relative path from your Miri clone to the rootCargo.toml
of the locally built rustc. For example, the path might look like../rust/Cargo.toml
.
See the rustc-dev-guide's docs on "Configuring rust-analyzer
for rustc
"
for more information about configuring VS Code and rust-analyzer
.
We described above the simplest way to get a working build environment for Miri,
which is to use the version of rustc indicated by rustc-version
. But
sometimes, that is not enough.
A big part of the Miri driver is shared with rustc, so working on Miri will
sometimes require also working on rustc itself. In this case, you should not
work in a clone of the Miri repository, but in a clone of the
main Rust repository. There is a copy of
Miri located at src/tools/miri
that you can work on directly. A maintainer
will eventually sync those changes back into this repository.
When working on Miri in the rustc tree, here's how you can run tests:
./x.py test miri
--bless
will work, too.
You can also directly run Miri on a Rust source file:
./x.py run miri --stage 1 --args src/tools/miri/tests/pass/hello.rs
We use the josh
proxy to transmit changes between the
rustc and Miri repositories. You can install it as follows:
RUSTFLAGS="--cap-lints=warn" cargo +stable install josh-proxy --git https://github.com/josh-project/josh --tag r23.12.04
Josh will automatically be started and stopped by ./miri
.
Note: this usually happens automatically, so these steps rarely have to be done by hand.
We assume we start on an up-to-date master branch in the Miri repo.
# Fetch and merge rustc side of the history. Takes ca 5 min the first time.
# This will also update the `rustc-version` file.
./miri rustc-pull
# Update local toolchain and apply formatting.
./miri toolchain && ./miri fmt
git commit -am "rustup"
Now push this to a new branch in your Miri fork, and create a PR. It is worth
running ./miri test
locally in parallel, since the test suite in the Miri repo
is stricter than the one on the rustc side, so some small tweaks might be
needed.
We will use the josh proxy to push to your fork of rustc. Run the following in the Miri repo, assuming we are on an up-to-date master branch:
# Push the Miri changes to your rustc fork (substitute your github handle for YOUR_NAME).
./miri rustc-push YOUR_NAME miri
This will create a new branch called miri
in your fork, and the output should include a link that
creates a rustc PR to integrate those changes into the main repository. If that PR has conflicts,
you need to pull rustc changes into Miri first, and then re-do the rustc push.
If this fails due to authentication problems, it can help to make josh push via ssh instead of
https. Add the following to your .gitconfig
:
[url "git@github.com:"]
pushInsteadOf = https://github.com/
The following environment variables are relevant to ./miri
:
MIRI_AUTO_OPS
indicates whether the automatic execution of rustfmt, clippy and toolchain setup (as controlled by the./auto-*
files) should be skipped. If it is set tono
, they are skipped. This is used to allow automated IDE actions to avoid the auto ops.MIRI_LOG
,MIRI_BACKTRACE
control logging and backtrace printing during Miri executions.MIRI_TEST_THREADS
(recognized by./miri test
) sets the number of threads to use for running tests. By default, the number of cores is used.MIRI_SKIP_UI_CHECKS
(recognized by./miri test
) disables checking that thestderr
orstdout
files match the actual output.
Furthermore, the usual environment variables recognized by cargo miri
also work for ./miri
, e.g.
MIRI_LIB_SRC
. Note that MIRIFLAGS
is ignored by ./miri test
as each test controls the flags it
is run with.
The following environment variables are internal and must not be used by anyone but Miri itself. They are used to communicate between different Miri binaries, and as such worth documenting:
CARGO_EXTRA_FLAGS
is understood by./miri
and passed to all host cargo invocations.MIRI_BE_RUSTC
can be set tohost
ortarget
. It tells the Miri driver to actually not interpret the code but compile it like rustc would. Withtarget
, Miri sets some compiler flags to prepare the code for interpretation; withhost
, this is not done. This environment variable is useful to be sure that the compiledrlib
s are compatible with Miri.MIRI_CALLED_FROM_SETUP
is set during the Miri sysroot build, which will re-invokecargo-miri
as therustc
to use for this build.MIRI_CALLED_FROM_RUSTDOC
when set to any value tellscargo-miri
that it is running as a child process ofrustdoc
, which invokes it twice for each doc-test and requires special treatment, most notably a check-only build before interpretation. This is set bycargo-miri
itself when running as arustdoc
-wrapper.MIRI_CWD
when set to any value tells the Miri driver to change to the given directory after loading all the source files, but before commencing interpretation. This is useful if the interpreted program wants a different working directory at run-time than at build-time.MIRI_LOCAL_CRATES
is set bycargo-miri
to tell the Miri driver which crates should be given special treatment in diagnostics, in addition to the crate currently being compiled.MIRI_ORIG_RUSTDOC
is set and read by different phases ofcargo-miri
to remember the value ofRUSTDOC
from before it was overwritten.MIRI_REPLACE_LIBRS_IF_NOT_TEST
when set to any value enables a hack that helps bootstrap run the standard library tests in Miri.MIRI_TEST_TARGET
is set by./miri test
(and./x.py test miri
) to tell the test harness about the chosen target.MIRI_VERBOSE
when set to any value tells the variouscargo-miri
phases to perform verbose logging.MIRI_HOST_SYSROOT
is set by bootstrap to tellcargo-miri
which sysroot to use for host operations.RUSTC_BLESS
is set by./miri test
(and./x.py test miri
) to indicate bless-mode to the test harness.