Add support for partial checkout filters

[finleygn]

A change to add support for sparse checkouts has been added recently (#1369). This extends upon that to allow support for other filters, and the ability to filter without a sparse checkout.

- uses: actions/checkout@v3
  with:
    filter: tree:0

Inspired by this blog post -> https://github.blog/2020-12-21-get-up-to-speed-with-partial-clone-and-shallow-clone

[gabrielyotoo]

bump +1 this would be nice to use with Yarn

[Rycieos]

This is an often requested feature:
#172, #663, #1152

I would personally love to see this added.

[petergaultney]

i've used this fork/branch successfully with both tree:0 and blob:none. Thank you!

[flying-sheep]

I think filter: 'tree:0' would be a much better default than fetch-depth: 1.

It would be almost exactly as small, but without breaking the Git history.

[gabrielyotoo]

did you miss exit 1 here?

[finleygn]

yep, good spot 😳

[jakobmerrild]

This is exactly what we need in our repo. Speeds up checkout by a factor of 4 without breaking git history so our existing logic that uses git diff and git blame will still work ❤️

[cory-miller]

Thanks for your contribution.

Do you mind adding an update to https://github.com/actions/checkout/blob/72f2cec99f417b1a1c5e2e88945068983b7965f9/action.yml#L56C19-L60 with the new setting with a note that it will override the sparseCheckout option if also set?

The rest LGTM.

[finleygn]

Will take a look as to whats causing the test to fail soon...

[finleygn]

@cory-miller the test actions should pass now 🙏

[cory-miller]

Thanks for working on this!

I will build a release soon.

[Kurt-von-Laven]

I think filter: 'tree:0' would be a much better default than fetch-depth: 1.

It would be almost exactly as small, but without breaking the Git history.

This is a cool feature, but please don't make the default checkout larger for everyone since most users don't need the Git history in their CI pipelines. We do often need the history, so we tried switching from fetch-depth: 0 to filter: tree:0, and our CI pipelines became twice as slow overall. Then we tried filter: blob:none, and that still slowed down our CI pipelines by 50%. I suspect this is happening because some tool (e.g., Gitleaks) is causing Git to fetch the missing data piecemeal rather than more efficiently all at once by attempting to access trees or blobs that aren't available locally. I don't mean this as a knock on partial clones; everyone should always measure the impact of performance changes in their own specific use case. I would guess that treeless clones are a net performance benefit if and only if no trees or blobs from previous commits are needed, because GitHub has already put so much effort into optimizing checkout times in GitHub Actions. Similarly, I would guess that blobless clones are a net performance benefit if and only if no blobs from previous commits are needed.

[jakebailey]

I think there's something missing from the docs here; you still need to explicitly set fetch-depth: 0 in order to get all of history, making a full workflow look like:

- use: actions/checkout@v4
  with:
    fetch-depth: 0
    filter: blob:none

Without setting it, you still only get the most recent commit.

For the TypeScript repo:

• fetch-depth: null, filter: null: 12s, but only one commit.
• fetch-depth: null, filter: blob:none: 30s, but still only one commit!
• fetch-depth: 0, filter: null, 4m6s: full history
• fetch-depth: 0, filter: blob:none, 1m37s: full history

For example, https://github.com/microsoft/TypeScript/actions/runs/6359689867/job/17272959471?pr=55923 (though this link will eventually break)

[Kurt-von-Laven]

@jakebailey, did you mean filter: blob:none rather than filter: filter:none for the last bullet?

[jakebailey]

Yes, that's just a typo on my part.