Developing Rust

Hacking on the Pants engine in Rust.

---

We welcome contributions to Rust! We use Rust to implement the Pants engine in a performant, safe, and ergonomic way.

Still learning Rust? Ask to get added to reviews

We'd be happy to ping you on Rust changes we make for you to see how Rust is used in the wild. Please message us on the #engine channel in Slack to let us know your interest.

Recommendation: share your plan first

Because changes to Rust deeply impact how Pants runs, it is especially helpful to share any plans to work on Rust before making changes. Please message us on Slack in the #engine channel or open a GitHub issue.

Code organization

The code for the top-level Pants Rust crate lives in src/rust/engine. The top-level Cargo.toml file at src/rust/engine/Cargo.toml defines a cargo workspace containing a number of other subcrates, which live in subdirectories of src/rust/engine. Defining multiple subcrates in this way allows changes affecting one subcrate to avoid affecting other subcrates and triggering more recompilation than is necessary.

Several of the particularly important subcrates are:

• graph: the core of Pants's rule graph implementation.
• ui: the dynamic UI.
• sharded_lmdb: custom wrappers around the crates.io lmdb crate, which provides bindings to lmdb.
• fs: manipulating the filesystem.
• process_execution: running local and remote processes.

Rust <-> Python interaction

Pants is best conceptualized as a Python program that makes frequent foreign function interface (FFI) calls into Rust code.

The top-level engine Rust crate gets compiled into a library named native_engine.so, which Python code knows how to interact with. We use the Rust cpython crate to manage foreign function interaction.

The C FFI functions that Rust code exposes as a public interface live in src/rust/engine/src/externs/interface.rs. On the Python side, the native module at src/python/pants/engine/internals/native.py contains wrappers around invocations of Rust functions and other FFI boilerplate.

Rust can also invoke Python functions and object constructors thanks to the cpython crate. Most of our helper code lives in src/rust/engine/src/externs/mod.rs.

We are planning to port additional functionality from Python to Rust, generally for performance reasons.

Common commands

Rather than using a global installation of Cargo, use the ./cargo script.

Compile

To check that the Rust code is valid, use ./cargo check. To check that it integrates correctly with Pants' Python code, use MODE=debug ./pants ... as usual (which will compile first, and is slower than check).

Set MODE=debug when iterating on the boundary between Rust and Python

As described in Setting up Pants, we default to compiling Rust in release mode, rather than debug mode.

When working on the boundary between Rust and Python, you typically should set the environment variable MODE=debug for substantially faster compiles.

Run tests

To run tests for all crates, run:

./cargo test

To run for a specific crate, such as the fs create, run:

./cargo test -p fs

To run for a specific test, use Cargo's filtering mechanism, e.g.:

./cargo test -p fs read_file_missing

Tip: enabling logging in tests

When debugging, it can be helpful to capture logs with env_logger.

To enable logging:

1. Add env_logger = "..." to dev-dependencies in the crate's Cargo.toml, replacing the ... with the relevant version. Search for the version used in other crates.
2. At the start of your test, add let _logger = env_logger::try_init();.
3. Add log statements wherever you'd like using log::info!() et al.
4. Run your test with RUST_LOG=trace ./cargo test -p $crate test_name -- --nocapture, using one of error, warn, info, debug, or trace.

Autoformat

./cargo fmt

To run in lint mode, add --check.

Run Clippy

./cargo clippy

The fs_util tool

fs_util is a utility that enables you to interact with Snapshots from the command line. You can use it to help debug issues with snapshotted files.

To build it, run this from the root of the repository:

$ cd src/rust/engine && ../../../build-support/bin/native/cargo build -p fs_util

That will produce src/rust/engine/target/debug/fs_util.

To inspect a particular snapshot, you'll need to tell fs_util where the storage is and the digest and length of the snapshot to inspect. You can use the --local-store-path flag for that.

For example, this command pretty prints the recursive file list of a directory through the directory subcommand.

$ src/rust/engine/target/debug/fs_util --local-store-path=${HOME}/.cache/pants/lmdb_store directory cat-proto --output-format=recursive-file-list <digesh> <len>

Pass the --help flag to see other ways of using fs_util, along with its subcommands. Each subcommand can be passed the --help flag.