Enable parserOptions.allowAutomaticSingleRunInference

[JoshuaKGoldberg]

Coming over from https://twitter.com/andhaveaniceday/status/1738986013320794313: typescript-eslint has a relatively undocumented allowAutomaticSingleRunInference feature that in many cases speeds up single-run linting. That includes CI.

Because eslint/rfcs#102 wasn't accepted (😞) and one last bug with it has yet to be explained (typescript-eslint/typescript-eslint#3851), we've been cautious to enable it by default in typescript-eslint. But as @jakebailey noted, [DT doesn't] have project references and it's all single run outside of the editor.

[jakebailey]

I misspoke; we do run in the editor, I actually just meant that the big single run is the "slow" run. Does this negatively impact the editor? Should this be set in the runner instead?

[JoshuaKGoldberg]

Does this negatively impact the editor?

I don't believe it does, no. @bradzacher can probably keep me honest - but to my knowledge the only known issue is the mysterious one from the case that uses project references.

[bradzacher]

Note that if you've got custom scripts to run eslint we won't be able to infer it as a CLI run and so the option will do nothing. You'll need to set the env var to force it on for that case.

[jakebailey]

What's the env var? I must have missed it trying to look through the docs.

[bradzacher]

TSESTREE_SINGLE_RUN=true should do it for you

[jakebailey]

Thanks! I think we should probably just set that right at the start of dtslint's main entrypoint, and skip the setting (given it won't work). When I (or someone else) gets a chance, I'll be curious to benchmark it.

I also want to try the ProjectService, though I suspect that there will be some OOMing because the functionality to ask it to drop files is probably not there yet.

[jakebailey]

So, it turns out that we already do this:

DefinitelyTyped-tools/packages/dtslint/src/lint.ts

Line 22 in d64f518

process.env.TSESTREE_SINGLE_RUN = "true";

Removing that, then benchmarking reenabling it, shows that it unfortunately does not help:

Benchmark 1: pnpm test gensync
  Time (mean ± σ):      5.412 s ±  0.022 s    [User: 8.465 s, System: 0.588 s]
  Range (min … max):    5.372 s …  5.435 s    10 runs
 
Benchmark 2: TSESTREE_SINGLE_RUN=true pnpm test gensync
  Time (mean ± σ):      5.444 s ±  0.050 s    [User: 8.517 s, System: 0.599 s]
  Range (min … max):    5.374 s …  5.521 s    10 runs
 
Summary
  'pnpm test gensync' ran
    1.01 ± 0.01 times faster than 'TSESTREE_SINGLE_RUN=true pnpm test gensync'

So, I guess this issue can be closed as we're already enabling it... but is it suspicious that it doesn't actually help?

[JoshuaKGoldberg]

Heh, maybe the issue should be tracking why it doesn't actually help?

But we're also planning on enabling it by default in v7 (typescript-eslint/typescript-eslint#8121). And this issue isn't very high on my priority queue to investigate. So I'm probably not going to be taking action soon tbh.