Post

Are there so many rules that look for such specific nodes to make for such a big performance change? That's quite surprising!

It was pretty surprising to us too, actually. I created a tool for analyzing AST node type frequencies in projects. Even in large projects, common things like import statements don't appear in ~15% of files, so we can skip import-related rules for those files. gist.github.com/camchenry/54...

Super interesting! Do you know which rules get removed in practice? Is it like rules that only care about TypeScript types for example?

Are you running the whole visitor for each rule? In my idea you run one visitor pass, where for each node you call the rules that want to visit that node so that if no node matches the rules, it doesn't cost to have it around

Yes 😅 What you described is pretty accurate to the changes we're trying to make. We keep a mapping of (node type) -> rules so we can quickly fetch all the rules that care about that node. The main innovation is being able to quickly determine if a file contains any given node type ahead of time.

Nice! Similar to what TypeScript does too. They have "containsX" bitflags on nodes. So can skip over transformers. E.g. if the code doesn't contain any ES2020 syntax, then that transformer can be skipped entirely.

github.com/oxc-project/... Note that this is still very much a work in progress. Probably won't be merged for a bit as the hard part is ensuring correctness when skipping rules.

Idea: ship it where you make your predictions about the rules that cannot possibly apply and if any is matched you report the transgression via telemetry.