# Contributing to `gimli`
Hi! We'd love to have your contributions! If you want help or mentorship, reach
out to us in a GitHub issue, or ping `fitzgen` in `#rust` on `irc.mozilla.org`.
* [Code of Conduct](#coc)
* [Filing an Issue](#issues)
* [Building `gimli`](#building)
* [Testing `gimli`](#testing)
* [Test Coverage](#coverage)
* [Using `test-assembler`](#test-assembler)
* [Benchmarking](#benchmarking)
* [Style](#style)
## <a id="coc"></a> Code of Conduct
We abide by the
[Rust Code of Conduct](https://www.rust-lang.org/en-US/conduct.html) and ask
that you do as well.
## <a id="issues"></a> Filing an Issue
Think you've found a bug? File an issue! To help us understand and reproduce the
issue, provide us with:
* The (preferably minimal) test case
* Steps to reproduce the issue using the test case
* The expected result of following those steps
* The actual result of following those steps
Definitely file an issue if you see an unexpected panic originating from within
`gimli`! `gimli` should never panic unless it is explicitly documented to panic
in the specific circumstances provided.
## <a id="building"></a> Building `gimli`
`gimli` should always build on stable `rustc`, but we recommend using
[`rustup`](https://www.rustup.rs/) so you can switch to nightly `rustc` and run
benchmarks.
To build `gimli`:
```
$ cargo build
```
## <a id="testing"></a> Testing `gimli`
Run the tests with `cargo`:
```
$ cargo test
```
### <a id="coverage"></a> Test Coverage
If you have `kcov` installed under linux, then you can generate code coverage
results using the `coverage` script in the root of the repository, and view them
at `target/kcov/index.html`. Otherwise you can create a pull request and view
the coverage results on coveralls.io.
```
$ ./coverage
```
The ideal we aim to reach is having our unit tests exercise every branch in
`gimli`. We allow an exception for branches which propagate errors inside a
`try!(..)` invocation, but we *do* want to exercise the original error paths.
Pull requests adding new code should ensure that this ideal is met.
At the time of writing we have 94% test coverage according to our coveralls.io
continuous integration. That number should generally stay the same or go up ;)
This is a bit subjective, because -.001% is just noise and doesn't matter.
### <a id="test-assembler"></a> Using `test-assembler`
We use the awesome
[`test-assembler`](https://github.com/luser/rust-test-assembler) crate to
construct binary test data. It makes building complex test cases readable.
[Here is an example usage in `gimli`](https://github.com/gimli-rs/gimli/blob/156451f3fe6eeb2fa62b84b362c33fcb176e1171/src/loc.rs#L263)
## <a id="benchmarking"></a> Benchmarking
The benchmarks require nightly `rustc`, so use `rustup`:
```
$ rustup run nightly cargo bench
```
We aim to be the fastest DWARF library. Period.
Please provide before and after benchmark results with your pull requests. You
may also find [`cargo benchcmp`](https://github.com/BurntSushi/cargo-benchcmp)
handy for comparing results.
Pull requests adding `#[bench]` micro-benchmarks that exercise a new edge case
are very welcome!
## <a id="style"></a> Style
We use `rustfmt` to automatically format and style all of our code.
To install `rustfmt`:
```
$ rustup component add rustfmt-preview
```
To run `rustfmt` on `gimli`:
```
$ cargo fmt
```