If you're interested in adding a feature or fixing a bug, thank you! Please start by reading this document and opening a Github discussion or bug about the thing you want to do, to avoid wasted work. void wasted work, and feel free to talk or ask about the approach.
Please also read the DESIGN.md file for technical information not specifically about posting contributions.
This project is conducted in accord with the Rust Code of Conduct.
One of the most helpful things you can do is to try cargo-mutants on a new tree: either your own project or an important open source project:
- Did cargo-mutants hang, error, or otherwise fail to run? If so a bug with reproduction instructions would be very helpful.
- Were the mutants interesting or helpful in understanding coverage or test quality? Did it generate any new tests that were accepted into the tree?
Generally, variables and parameters should be the snake_case version of their
type name: source_tree: SourceTree. However if that would be unclear or ambiguous, use a different name that does not repeat the type: src: &Path, dest: &Path.
Try to keep one major class or separation of concern per mod, with
implementation details being private. However, fields that would have trivial
getters and that don't break the abstraction can be public, since this crate does not provide a library API. public is used mostly as a marker that something is an implementation detail of a module.
Please run cargo fmt and cargo clippy. These are checked in CI.
Of course, please add tests for new features or bug fixes, and see the Testing section of the design doc.
Tests should have names that read like English sentences (or subsentences)
asserting a fact about how the program behaves, like
copy_testdata_doesnt_include_build_artifacts. It's fine if this makes the
test function names relatively long.
However, also try to avoid "noise" or low-value words in test names.
show_version is clear enough and does not need to be
when_run_with_show_version_it_prints_the_version_to_stdout.
As with other code, if you feel you need to add a comment to explain what it does, then first consider whether the test can have a better name.
If the test exercises a particular test tree, option, or function, make sure that name literally occurs within the test name.
Many tests use Insta to assert the expected output. Insta makes it easy to review and accept changes to expected output, either when there's a real functional change or when for example line numbers or output formatting changes.
To conveniently review changed output, cargo install cargo-insta and then run cargo insta test --review etc.
CLI tests spawn a new process which can be slightly slow, especially on Windows, and especially if they actually test the mutants.
Try to keep testdata trees reasonably minimal for whatever they're intended to test.
When the tests run, cargo test runs a test suite binary from target, which then runs cargo-mutants as a subprocess. (Or, in some cases, it runs cargo which runs cargo-mutants.)
As a result, attaching a debugger to the test binary will let you see the code that launches the subprocess and that inspects the output, but it won't let you step through cargo-mutants itself, which is probably the most interesting part.
Probably the easiest path is to just make note of the command run by the test, and then run that command yourself, under a debugger, outside of the test suite. For example, ./target/debug/cargo-mutants -d ./testdata/tree/factorial --list.
You may wish to turn off the timeouts with -t 0.
The largest area for new work at the moment is in generating new mutations. Most of the code for this is in visit.rs.
If you look in mutants.out/debug.log you can see messages like Return type is not recognized, trying Default.
These might be good places to add a new more specific pattern.
Also mutants.out/unviable.txt might suggest ways to generate new patterns that are viable.