I had started writing a 1-for-1 Pylint port in Rust (almost nowhere right now), but Ruff has gotten there way faster. Props!

Pylint is rough to make faster, party from the existing structure, but honestly at one point you are just paying a huge cost for traversing wide ASTs and doing all this work, and then paying extra for the Python object model.

In my profiling of Pylint, I've found some fixes, and I am curious about what something like slots or mypyc could do to it... but I'm also very partial to just having a system language implementation.

People talk about ease of contributing as a big advantage for these kinds of tools to be written in the language they are working on. Personally I've found that people who like messing with tools like the idea of learning languages to contribute.

My one kind of reservation about Ruff is that the code base is designed as "run each checker as its own function". My suspicion is that this leads to a performance ceilings as each checker is traversing the AST looking for its flavor of thing. Visitor patterns (theoretically!) allow for us to run checkers in parallel nicely. Granted the Ruff pattern allows for parallel work as well (and you can share all of the AST data between threads anyways), so maybe its "right". It definitely looks easier to maintain!