Are ESLint rules around performance worth advocating?

[johnnyreilly]

Hey @DanielRosenwasser,

I do hope you're well. First up, this is not directly an issue with the wiki at all, but I thought this was a good place to have a conversation about your wiki page on performance.

I was just talking with @dragomirtitian about me writing a blog post which would show how to take a few of the tips in here and apply them with ESLint; there’s ESLint rules out there that can help with this. Essentially the main ones of interest are:

• Prefer Interface over Type https://github.com/cartant/eslint-plugin-etc/blob/main/docs/rules/prefer-interface.md
• Enforce return types https://github.com/typescript-eslint/typescript-eslint/blob/master/packages/eslint-plugin/docs/rules/explicit-function-return-type.md

Titian posed the (entirely helpful) question:

I'm sure these make difference.. By at what point.. 1K loc? 10Kloc? 100? If this is a concern for huge scale projects is it really worth it for smaller projects? I mean shaving off 0.2s in a build doesn't seem worth it if I have to maintain a type and an inferable return type.

This is a great question. And I have to state a personal stylistic preference fortypes over interfaces, as well as preferring return type inference. Much less noisy. But perf is royalty and style is secondary.

My own feeling is that if you want to go with these rules, it’s easier if you start with them at the beginning of a project that you expect to grow. As retro-fixing all of these in a codebase would not be fun!

Anyway, rambling done. I have two questions for you if have a moment:

1. Is it worth blogging to advocate that people use the above ESLint rules? Or is this potentially transient advice? Something that’s useful now but could change as the compiler evolves?

2. Do you have a feeling for what level of difference that following these standards can make? The answer will no doubt be "it depends", but I'm curious if you happen to have an idea for what the boundaries of success could look like here. So, is it possible to say something like "enforcing return types on functions has been observed to improve type checking performance by up to 40%", or "preferring interfaces over types has been observed to improve type checking performance by up to 20%"? Actually having an idea of the level of difference that could be made is a super helpful piece of info. (Essentially this is Titian's question 😄 )

I hope you don't mind me posting the question here. I did consider DMing you instead, but I thought I'd do a @swyx and "learn in public" 😄