ReviewOS

also looking at this

stacks/bun-query-builder

fix(ci): address the pantry action by its real owner, not a rename redirect

#1103
Merged glennmichael123 wants to merge fix/pantry-action-owner into main
2 files +2 -2
10 days ago

Both workflows reference home-lang/pantry/packages/action@main:

  • .github/workflows/ci.yml:124
  • .github/workflows/release.yml:23

home-lang/pantry does not exist. The reference resolves only through GitHub's rename redirect:

$ gh api repos/home-lang/pantry --jq .full_name
pantry-pm/pantry

Meanwhile the home-lang org is live, with 7 repos (home, home-os, pantry-registry, …) and no pantry among them. A GitHub rename redirect survives only while the old name stays free at the old location, so this indirection is load-bearing for both CI and the release path.

That matters more on the release side: the pantry step runs at release.yml:23, ahead of the publish steps, in a job holding id-token: write (:14), NPM_TOKEN (:42) and GITHUB_TOKEN (:57).

This PR points both workflows at pantry-pm/pantry directly — where they already resolve to. Same action, same ref, one less indirection. No behavior change; this PR's own CI exercises it, since publish-commit runs the step from the PR head.

What this deliberately does not do

It leaves the @main ref alone. That is a separate decision with a real trade-off and it is yours, not mine:

  • @main (today) — upstream fixes arrive immediately; upstream breakage does too, unreviewed. This is what reddened main on Aug 14, when an SDK change promoted sqlite: latest from silently-skipped to fatally-unresolvable with no commit on our side (fixed in 0671de8 by moving deps.yaml to ranges).
  • Pin to a tag — v0.11.29 is current; tags are published, so this is available today.
  • Pin to a SHA — safest, needs a bot or a human to bump.

Two things worth knowing before choosing:

  1. pantry-pm/pantry@main moved this morning (19255bff7301, 11:04Z). The ref is genuinely live, not dormant.
  2. Pinning uses: alone is only half the job — the action also resolves a pantry binary version, so pinning one and not the other leaves the part that actually runs publish:commit still floating.

Happy to do whichever you prefer as a follow-up. It is also worth checking whether buddy-bot is configured to bump action refs, since that determines whether SHA pinning is sustainable here or turns into manual work.

🤖 Generated with Claude Code

2 changed files on the files tab, with 0 review threads.

Sign in to comment on this pull request.