Fetch without an object filter when the workspace shares the mirror

[piob-io]

Summary

A filter: input (or the blob:none that sparse-checkout implies) is now dropped whenever the workspace verifiably shares the mirror's object store (sharesMirrorObjects, the same gate the shallow negotiation uses). Without a mirror the filter is passed through unchanged.

Why. With the mirror attached as an alternate, every object a filter would omit is already local, so the filter cannot save any transfer — but it does make the workspace a partial clone, and that is expensive with an alternate behind it. Since git 2.48, fetch-pack runs index-pack --promisor for a filtered fetch, and index-pack --promisor repacks every locally available, non-promisor object linked from the received pack into a new promisor pack (repack_local_links → pack-objects --exclude-promisor-objects-best-effort). The fetched commit's parent lives in the mirror, so that repack walks the mirror's history reachable from it and writes it into the workspace's .git/objects. There is no config to turn this off.

Reproduced on an 8.5 GB mirror with git 2.51 (trace2), --filter=blob:none --depth=2 of a merge commit whose parent is the mirror tip:

upload-pack + pack-objects (server):          0.30 s, 17 objects, 62 KB      <- the actual fetch
index-pack --promisor
  └ pack-objects --exclude-promisor-objects-best-effort: 100.9 s
      -> pack-c271….pack  3.7 GB, 7,260,634 objects, .promisor
same fetch without --filter:                  0.59 s total

In production this shows up as a flat ~60 s per filtered fetch that receives any objects, independent of depth or ref count (proportional to mirror size), plus several GB written to the runner's root disk per job; unfiltered fetches of the same shape take ~1 s. A filtered fetch that receives no objects (ref already in the mirror) is unaffected, which is why the cost is bimodal.

Change (git-source-provider.ts):

const sharesMirrorObjects = !!cacheInfo && await blacksmithCache.sharesMirrorObjects(repositoryPath, mirrorPath)
const filter = settings.filter ?? (settings.sparseCheckout ? 'blob:none' : undefined)
if (filter && sharesMirrorObjects) core.info(`[git-mirror] Fetching without --filter=${filter}: ...`)
else if (filter) fetchOptions.filter = filter

Side effect: fetch-depth: 0 + filter: now qualifies for the direct mirror ref copy (fetchRefsFromMirror, previously gated on !fetchOptions.filter), so it no longer does a +refs/heads/* network fetch either. The resulting workspace is a regular clone (no remote.origin.promisor / partialclonefilter); the only behavioural difference for later steps is that objects the mirror lacks are not lazily fetched — with a full mirror behind the alternates there are none, and dissociate: true is unaffected.

Tests

• __test__/mirror-filter-fetch-git.test.ts (real git): unfiltered full and depth-2 fetches into an alternates workspace store ≤ 6 objects, write no .promisor, leave no partial-clone config, and check out the whole tree; a blob:none fetch into the same workspace does produce a promisor pack, and on git ≥ 2.48 its object count exceeds the mirror's history. Passes on git 2.34 and 2.51.
• test-blacksmith.yml: fetch-depth: 0 + filter: blob:none and a sparse-checkout checkout on the real mirror; asserts no promisor packs, no partial-clone config, < 1000 local objects, fsck clean, sparse cone applied.
• README / action.yml filter description updated.

Link to Devin session: https://app.devin.ai/sessions/a3b42ef15226438faf3f9cdde9662918
Open in Devin Desktop: https://app.devin.ai/desktop/session/a3b42ef15226438faf3f9cdde9662918?variant=devin
Requested by: @piob-io

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled. (Staging)}

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring