I prefer to have unused code detected during linting, but sadly, eslint has decided to kill off the APIs that support rules like `no-unused-modules`. Running a separate tool like this one or knip in place of a few lint rules just seems impractical.
eslint is a good example of why coding in javascript is annoying. Your tools just constantly changing wildly over a version upgrade, so you look for a better one and find there's a new Rust linting tool but it's alpha and is missing half the features.
eslint is also a good example why Javascript runtime is bad choice for static analysis tools. The biggest problem is that it's single threaded.
Recent release of concurrency mode in eslint promised approximately 30% linting speed increase.
So now it uses multiple threads instead of one, and you got only 1.3x improvement. In any compiled language like Rust or Go you should expect time improvement that correlates with number of CPU cores engaged.
You can use worker threads in JS, but unfortunately sharing data between threads in context of static analysis, where there is a lot of deeply nested objects (AST) or basically a lot of data in general, is slow because it needs to be serialised and deserialised when it's passed between threads.
Javascript based tools for codebases with 1m+ lines of code becomes unusable :/
+1 for the idea. Enforcing hard boundaries between is surprisingly helpful for AIs to reason about how to structure their changes.
We recently rolled out our own static analysis using oxc-parser and oxc-resolver, and it runs surprisingly fast (<1s for ~100K LOC). For us, it was definitely worth adding this layer of defence against The Slop.
Nice!
I’ve come to similar conclusions recently, with the recent increase in code changes velocity, solid static analysis is more important than ever.
When it comes to the performance, I've learned that reading code from file system and parsing it takes most of the time. Then resolving modules takes a little also.
Once that is done, running different checks is almost instant - like miliseconds.
I prefer to have unused code detected during linting, but sadly, eslint has decided to kill off the APIs that support rules like `no-unused-modules`. Running a separate tool like this one or knip in place of a few lint rules just seems impractical.
eslint is a good example of why coding in javascript is annoying. Your tools just constantly changing wildly over a version upgrade, so you look for a better one and find there's a new Rust linting tool but it's alpha and is missing half the features.
eslint is also a good example why Javascript runtime is bad choice for static analysis tools. The biggest problem is that it's single threaded.
Recent release of concurrency mode in eslint promised approximately 30% linting speed increase.
So now it uses multiple threads instead of one, and you got only 1.3x improvement. In any compiled language like Rust or Go you should expect time improvement that correlates with number of CPU cores engaged.
You can use worker threads in JS, but unfortunately sharing data between threads in context of static analysis, where there is a lot of deeply nested objects (AST) or basically a lot of data in general, is slow because it needs to be serialised and deserialised when it's passed between threads.
Javascript based tools for codebases with 1m+ lines of code becomes unusable :/
Biome has been out of alpha for a few years now and is fantastic :-)
+1 for the idea. Enforcing hard boundaries between is surprisingly helpful for AIs to reason about how to structure their changes.
We recently rolled out our own static analysis using oxc-parser and oxc-resolver, and it runs surprisingly fast (<1s for ~100K LOC). For us, it was definitely worth adding this layer of defence against The Slop.
Nice! I’ve come to similar conclusions recently, with the recent increase in code changes velocity, solid static analysis is more important than ever.
When it comes to the performance, I've learned that reading code from file system and parsing it takes most of the time. Then resolving modules takes a little also.
Once that is done, running different checks is almost instant - like miliseconds.
Looks good; I'm eager to try it. Do you have any questions?