chore: replace fast-glob with tinyglobby

[SuperchupuDev]

Description

tinyglobby is meant to be a drop-in replacement to globby and fast-glob, and it only uses two subdependencies - fdir and picomatch. as discussed in the discord server, this PR switches to it

expandDirectories is explicitly disabled for better fast-glob compatibility

funny enough the vitest monorepo already uses it in some way, due to unocss and vite-plugin-pwa using it

Please don't delete this checklist! Before submitting the PR, please make sure you do the following:

• It's really useful if your PR references an issue where it is discussed ahead of time. If the feature is substantial or introduces breaking changes without a discussion, PR might be closed.
• Ideally, include a test that fails without this PR but passes with it.
• Please, don't make changes to pnpm-lock.yaml unless you introduce a new test example.

Tests

• Run the tests with pnpm test:ci.

Documentation

• If you introduce new functionality, document it. You can run documentation with pnpm run docs command.

Changesets

• Changes in changelog are generated from PR name. Please, make sure that it explains your changes in an understandable manner. Please, prefix changeset messages with feat:, fix:, perf:, docs:, or chore:.

[netlify]

✅ Deploy Preview for vitest-dev ready!

Name	Link
🔨 Latest commit	ab3aded
🔍 Latest deploy log	https://app.netlify.com/sites/vitest-dev/deploys/66daf80c3975a8000868d95b
😎 Deploy Preview	https://deploy-preview-6274--vitest-dev.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify site configuration.

[SuperchupuDev]

the tests seem to fail some snapshots, but only because they have a different order. should i update the snapshots or do a manual .sort?

[sheremet-va]

the tests seem to fail some snapshots, but only because they have a different order. should i update the snapshots or do a manual .sort?

Why is the order different?

[SuperchupuDev]

unsure why until i debug a bit more, but it looks like it's alphabetically sorted now, maybe it's the opposite in fast-glob? ideally vitest should report errors in the same order regardless of the order the glob results came in. from earlier tests i did weeks ago tinyglobby's globs aren't guaranteed to be sorted. iirc globs in general follow a non-deterministic order even if fast-glob in particular doesn't

[sheremet-va]

unsure why until i debug a bit more, but it looks like it's alphabetically sorted now, maybe it's the opposite in fast-glob? ideally vitest should report errors in the same order regardless of the order the glob results came in. from earlier tests i did weeks ago tinyglobby's globs aren't guaranteed to be sorted. iirc globs in general follow a non-deterministic order even if fast-glob in particular doesn't

Vitest has a built in sequencer that does sorting: https://vitest.dev/config/#sequence-sequencer

[SuperchupuDev]

i'm having a bit of trouble debugging due to vitest running tests using itself

[sheremet-va]

Hm, looks like we sort by names only when sharding. Normally, we use fs.stat and fallback to the existing sorting

[SuperchupuDev]

okay, so after debugging i can confirm that on that particular test case, fast-glob returns the two results reverse sorted, while tinyglobby does sort them correctly. should i just change the snapshots?

[SuperchupuDev]

pushed the snapshot sorting update, i don't see why it should keep the unsorted order it had, and if anything that can always be changed in a followup pr

[sheremet-va]

okay, so after debugging i can confirm that on that particular test case, fast-glob returns the two results reverse sorted, while tinyglobby does sort them correctly. should i just change the snapshots?

What determines “correctly”? Why the previous sorting is incorrect and what are the consequences of that change? Is this flaky? Will test order be non-deterministic with tinyglobby?

[SuperchupuDev]

from my testing, the array of two elements produced from globbing in that test is not alphabetically sorted, and it is in tinyglobby, every single time i've ran these tests. by correct i meant it being sorted at all. despite that, i don't have evidence that globbing is deterministic with tinyglobby. the proper fix would be to just let vitest sort the reports itself

[sheremet-va]

by correct i meant it being sorted at all.

You said it returns them in reversed order, so that's sorted, no? Just in reverse.

i have no idea why the tests are failing now, the only thing changed after force pushing was that snapshot in particular.

Tests bail out early, so if there are more errors, you won't see them unless all tests have run

[sheremet-va]

Sorry for being picky here. I just remember that the sorting is important for filtering for example and while I appreciate the smaller size and dependencies count (and I agree that we should keep making it smaller), we should keep it as compatible as possible.

[SuperchupuDev]

that array of two elements in particular seems to be reverse sorted, not because fast-glob returns globs reverse sorted but because an array of two can only ever be either sorted or reverse sorted :P weird wording on my part

[SuperchupuDev]

it seems like the config in that test in particular breaks tinyglobby, and it'd be complex and hacky enough to add. thankfully it's the only remaining failing test

https://github.com/vitest-dev/vitest/blob/eae3bf4a72c1ab0b0b8ea0e13a3ed8a415ebfc66/test/reporters/reportTest2/custom-reporter-path.vitest.config.ts

it looks like a very hacky config? if the root is __dirname, it makes no sense to include tests outside that root. do users actually do that instead of defining a broader root (like path.join(__dirname, '..'))?

[sheremet-va]

it seems like the config in that test in particular breaks tinyglobby, and it'd be complex and hacky enough to add. thankfully it's the only remaining failing test

eae3bf4/test/reporters/reportTest2/custom-reporter-path.vitest.config.ts

it looks like a very hacky config? if the root is __dirname, it makes no sense to include tests outside that root. do users actually do that instead of defining a broader root (like path.join(__dirname, '..'))?

It does seem weird, but if it worked before then someone might've used this in the wild. But I haven't seen any usage of this pattern to be honest. We'll discuss with the team how we can proceed here in the next meeting. Thank you for the PR!

[SuperchupuDev]

after thinking a lot about the problem, it is something important to have regardless of its use in vitest. i have a not that hacky fix almost ready, but it makes using patterns with leading ../ slower in some cases (cases that can be made a lot faster using the ignore option). ill publish a new version with the fix very soon

[SuperchupuDev]

opened a fix SuperchupuDev/tinyglobby#18, will wait until you have the meeting you mentioned though

[sheremet-va]

The team thinks it's fine to drop support for ../

[SuperchupuDev]

after refactoring out the ../ in that one test tests seem to pass now, except one in macos that's failing in other prs too

[SuperchupuDev]

ended up adding ../ support in 0.2.3 anyways, some use cases from an unrelated package were breaking without it