Thanks to a very helpful pairing with @jakebailey, I filed two three issues on TypeScript:

• ⚡ Performance: Project service spends excess time cleaning client files when called synchronously microsoft/TypeScript#59335
• ⚡ Performance: Project service doesn't cache all fs.statSync microsoft/TypeScript#59338
• ⚡ Performance: Project service doesn't cache all fs.realpath  microsoft/TypeScript#59342

Those seem to be most of the performance discrepancy. ⚡

Removing from the v8 milestone. There are multiple issues filed over on TypeScript and our usage in community projects (e.g. t3-oss/create-t3-app#1936, trpc/trpc#5868) has shown real-world cases tend to be at least as fast.

Status update: microsoft/TypeScript#59335 resolved something like >60-70% of the discrepency. From microsoft/TypeScript#59689:

┌──────────┬───────┬───────────────────────┬───────────────────────┐
│ version  │ files │ project (even layout) │ service (even layout) │
┼──────────┼───────┼───────────────────────┼───────────────────────┤
│ baseline │ 1024  │ '2.524 s ±  0.016 s'  │ '3.338 s ±  0.013 s'  │
┼──────────┼───────┼───────────────────────┼───────────────────────┤
│ modified │ 1024  │ '2.531 s ±  0.010 s'  │ '2.846 s ±  0.010 s'  │
┴──────────┴───────┴───────────────────────┴───────────────────────┘

I'm on @eslint/js": "^9.12.0, just migrated to 9th version and also wanted to compare project vs projectService performance and got these results:

=== project ===
pnpm exec eslint .  15.32s user 1.35s system 158% cpu 10.532 total
pnpm exec eslint .  14.35s user 1.18s system 161% cpu 9.629 total
pnpm exec eslint .  14.21s user 1.02s system 160% cpu 9.475 total
pnpm exec eslint .  15.01s user 1.02s system 164% cpu 9.757 total
pnpm exec eslint .  14.12s user 1.20s system 156% cpu 9.772 total
=== projectService ===
pnpm exec eslint .  18.82s user 1.49s system 146% cpu 13.882 total
pnpm exec eslint .  18.92s user 1.46s system 148% cpu 13.701 total
pnpm exec eslint .  18.98s user 1.48s system 147% cpu 13.883 total
pnpm exec eslint .  19.12s user 1.58s system 147% cpu 14.062 total
pnpm exec eslint .  19.04s user 1.40s system 147% cpu 13.819 total

I checked the Performance document, but we don't have reloads.
I tested it on TS 5.5, then after reading your last comment I updated TS to 5.6, and now it 1-2 seconds slower >_<
All packages related to TS or eslint are up to date.

I notice, that when we added references the tsconfig.json of non-composite projects that add some overlapped we yet able to fix caused a big slowdown. I quite sure about it, since I needed to add a new tsconfig.json of this kind today. If there are some steps I can make to help give more details, tell me what to do (This is a private repo, so it has some limitations)

Oh! Sorry @oriSomething I missed your message.

Without an actual reproduction it's hard to help. The normal steps we recommend are:

1. Read through https://typescript-eslint.io/troubleshooting/typed-linting/performance and try the techniques in there
2. If nothing in that page helps, try to make a minimum reproducible project or playground without any proprietary/secret stuff that you can share publicly

In general, if you have get partial results that aren't already noted in this issue, it can be helpful to post them. But something we can look at ourselves is really the best thing.

1. I already read https://typescript-eslint.io/troubleshooting/typed-linting/performance and didn't help
2. Technically, it will be and issue to reproduce to a small repo, because it's a big monorepo, and I cannot exposed the source anyway - this is why I asked if there are steps I can debug it myself to source of the slowness. (Worth saying, there is no issue using TSC specifically, only typed lint rules with TSESLint)

I can understand if you can't help more than this.

Thanks for your time!

@oriSomething we have the same problem. Changing:

project: ['./tsconfig.json'],

to:

projectService: true,

increases ESLint's run time from 1m17s to 1m46s.

@olegafx If it wasn't clear. projectService: true was slower in my case

@olegafx If it wasn't clear. projectService: true was slower in my case

That’s exactly what I said. Using projectService: true is slower in our case as well – adding +37% (+29s) to the total run time.

@olegafx I'm so sorry. I miss read 🙈