brush

About

brush (Bo(u)rn(e) RUsty SHell) is a shell implementation with aspirations of compatibility with the POSIX Shell specification and bash.

It's generally functional for interactive use and can execute many scripts but still a work in progress. We do not recommend using this in production scenarios; until it's more stable, there's risk that using this implementation in place of your stable shell may result in unexpected behavior.

This project was primarily borne out of curiosity and a desire to learn. If it proves to be interesting or useful, then that's a bonus :).

Contributions and feedback of all kinds are welcome! For more guidance, please consult our contribution guidelines. For more technical details, please consult the documentation in this repo.

License

Available for use and distribution under the MIT license.

Try it out!

We don't yet have binary releases of brush, but if you have a working rust toolchain installed you can simply run:

cargo install --locked brush-shell

Alternatively, you can clone this repo and execute cargo run. If you don't have rust installed, we recommend installing it via rustup.

When you run brush, it should hopefully look much like bash would on your system since it processes .bashrc and other usual configuration. If you'd like to customize the look of brush to distinguish it from the other shells installed on your system, then you can also author a ~/.brushrc file.

What's working?

In short, quite a lot. Standard and extended control flow, word expansion, most frequently used builtin commands, pipelines, redirection, variables, etc. The plumbing for completion is present, along with support for common cases (e.g. file/dir completion, basic support for programmable completion such as used with git and other tools).

Known limitations Where you can help!

There's a lot that is working, but there are non-trivial gaps in compatibility. Most notably:

Commands run asynchronously as jobs, job management. You can run some-command & but it's proof-of-concept quality at best. Standard job management via fg, bg, and jobs is not fully implemented. This would be a great area for enthusiastic contributors to dive in :).
Honoring set and shopt options (e.g., set -e). The set builtin is implemented, as is set -x and a few other options, but most of the behaviors aren't there. set -e, for example, will execute but its semantics aren't applied across execution.
Curly brace expansion. Most forms of expansion are implemented; for some reason, we never got around to implementing an expansion that turns {a,b} into a b. There's even a test for this, but it's marked as a known failing test.
Tokenizer and parser edge cases. For simplicity and ease of getting going, the tokenizer for brush was hand-implemented and the parsers were implemented using rust-peg. (Also a huge shout out to that project!) There are some edge cases that got tricky and may not be implemented with 100% fidelity (e.g., complex nested parenthetical expressions in arithmetic expressions, further nested inside command substitutions). All of our tests are passing in these areas, but coverage is limited. Augmenting test coverage would be a great starting point.
Anything tagged with a TODO comment or where error::unimp() is used to return a "not implemented" error. These aren't all tracked with GitHub issues right now, but there's many of these scattered throughout the code base. Some are indicative of missing functionality that may be straightforward to implement; others may entail cross-cutting challenges that require design work. These include shell built-ins. Some are completely and fully implemented (e.g. echo), while some only support their most commonly used options. Some aren't implemented at all.

There's certainly more gaps; with time we'll find a way to represent the gaps in some understandable way. Ideally, we'd like to evolve the test suites to add tests for all known missing pieces. That will let us focus on just "fixing the tests".

We'd absolutely love your help with any of the above, with broadening test coverage, deeper compatibility evaluation, or really any other opportunities you can find to help make this project better.

Testing strategy

This project is primarily tested by comparing its behavior with other existing shells, leveraging the latter as test oracles. The integration tests implemented in this repo include 300+ test cases run on both this shell and an oracle, comparing standard output and exit codes.

For more details, please consult the reference documentation on integration testing.

Credits

There's a long list of OSS crates whose shoulders this project rests on. Notably, the following crates are directly relied on for major portions of shell functionality:

rustyline - for readline input and interactive usage
clap - command-line parsing, used both by the top-level brush CLI as well as built-in commands
fancy-regex - relied on for everything regex
tokio - async, well, everything

Huge kudos and thanks also to pprof and criterion projects for enabling awesome flamegraphs in smooth integration with cargo bench's standard benchmarking facilities.

Links: other shell implementations

This is certainly not the first attempt to implement a feature-rich POSIX-ish shell in a non-C/C++ implementation language. Some examples include:

We're sure there are plenty more; we're happy to include links to them as well.

lu-zero/brush