Skip to content

chore(deps): update projectbluefin/actions digest to b0fbcf6 - #172

Closed
mergeraptor[bot] wants to merge 2 commits into
mainfrom
renovate/projectbluefinactions
Closed

mergeraptor[bot] wants to merge 2 commits into
mainfrom
renovate/projectbluefinactions

Conversation

@mergeraptor

@mergeraptor mergeraptor Bot commented Sep 16, 2026

Copy link
Copy Markdown
Contributor

This PR contains the following updates:

Package Type Update Change
projectbluefin/actions (changelog) workflow digest 6680f74 → b0fbcf6

Configuration

📅 Schedule: (UTC)

  • Branch creation
    • At any time (no schedule defined)
  • Automerge
    • At any time (no schedule defined)

🚦 Automerge: Enabled.

♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.


  • If you want to rebase/retry this PR, check this box

This PR has been generated by Mend Renovate CLI.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@mergeraptor

mergeraptor Bot commented Sep 16, 2026

Copy link
Copy Markdown
Contributor Author

Edited/Blocked Notification

Renovate will not automatically rebase this PR, because it does not recognize the last commit author and assumes somebody else may have edited the PR.

You can manually request rebase by checking the rebase/retry box above.

⚠️ Warning: custom changes will be lost.

@hanthor hanthor left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The digest bump itself is a genuine no-op, and I want to state that first because it makes the rest of this sharper.

6680f74..b0fbcf6 is exactly one commit, b0fbcf6 "fix(promotion): gate queue enrollment on verification (#513)", touching three paths:

.github/workflows/reusable-promote-squash.yml
docs/skills/factory-operations.md
tests/test_promote_gate_order.py

build.yml:179 pins server-installer-test.yml, which is byte-identical across the range (git diff --shortstat 6680f74 b0fbcf6 -- .github/workflows/server-installer-test.yml → empty). Nothing in that bump reaches this repo.

But this PR is not only a digest bump, and its title says it is.

Two of the three changed files are BuildStream junctions:

elements/freedesktop-sdk.bst    26.08.0-0-gdb97cce  →  26.08.1-0-gb02b59f
elements/gnome-build-meta.bst   50.4-0-g8524013     →  50.4-5-gd78f498

That is a freedesktop-sdk point release and five commits of gnome-build-meta — the entire base of the OS — arriving inside a PR whose title, body and Renovate table describe a GitHub Actions ref change. They came from the second commit on the branch, ed0d7bf "chore: resolve BuildStream junction refs to git-describe format", pushed by github-actions[bot] 49 seconds after Renovate opened the PR.

That commit is build.yml's track-refs job:

if: github.event_name == 'pull_request' && startsWith(github.head_ref, 'renovate/')
...
just bst source track freedesktop-sdk.bst || true
just bst source track gnome-build-meta.bst || true
if ! git diff --exit-code; then ... git push origin "HEAD:${HEAD_REF}"; fi

The message says "resolve to git-describe format", but both refs were already in git-describe format. What actually happens is that bst source track re-resolves the track: globs — freedesktop-sdk-26.08* and gnome-50 — against their remotes and takes whatever is newest. So the job does not normalise a format; it performs an unpinned upstream bump, on every renovate/* PR, whatever that PR was about.

Three consequences, and the combination is why I'm requesting changes rather than commenting:

  1. The label on the tin is wrong. Anyone reviewing "chore(deps): update projectbluefin/actions digest to b0fbcf6" is not reviewing an SDK point release, and the Renovate body gives them no reason to look.
  2. It carries the automerge label and Renovate reports "Automerge: Enabled". So the path exists for an FSDK bump to reach main with no human having read it.
  3. It routes around the automerge policy the org just tightened. projectbluefin/actions 6e4587b gates Renovate automerge to no-new-code update types (pin/pinDigest) specifically so that automerge can't introduce new code. That gate inspects what Renovate proposed. Here the new code is injected afterwards, by CI, into a PR that already passed the gate. The check is being evaluated against a commit that is no longer the head.

What I'd suggest, in rough order of how much I'd want it:

  • Have track-refs open its own PR for junction movement instead of pushing into whatever renovate/* branch happens to be open. That keeps "the SDK moved" a reviewable event with its own title.
  • Failing that, make it refuse to push when the resolved ref crosses a version boundary rather than merely a commit, and strip automerge from any PR it has pushed to.
  • Independently: || true on both bst source track lines means a tracking failure is indistinguishable from "nothing to do". If tracking silently fails, the job reports success and the build proceeds on the old ref. Worth at least logging the outcome.

Also worth a second look, though I'm less sure of it: track-refs grants contents: write and then checks out github.event.pull_request.head.repo.full_name and runs that tree's Justfile. The renovate/* branch-name gate is the only guard, and a branch name is not an identity — a fork PR can be branched renovate/anything. GitHub does force GITHUB_TOKEN read-only for pull_request events from forks, which should make the git push fail rather than succeed, so I don't think this is currently exploitable. I'd still rather the gate checked github.event.pull_request.user.login or head.repo.full_name == github.repository than the branch name, since the workflow's own comment treats the branch gate as the security boundary.

For completeness: mergeable_state: unstable, head ed0d7bf, no commit statuses posted.


Generated by Claude Code

@hanthor

hanthor commented Sep 17, 2026

Copy link
Copy Markdown
Member

Following up on my review with something more useful than a complaint: dakota already solves this, and server is the only repo in the org doing it the other way.

I grepped every repo's workflows for source track. Two repos do BuildStream tracking at all:

  • dakota — .github/workflows/track-bst-sources.yml and track-next-junctions.yml. Triggered by schedule (0 6 * * *), workflow_dispatch and repository_dispatch, never by someone else's pull request. Each element group gets its own dedicated branch (auto/track-brew, auto/track-common, auto/track-gsconnect, … and a core-junctions group for gnome-build-meta.bst + freedesktop-sdk.bst) and its own PR against testing. The workflow_dispatch input even lets you track one group at a time. That is what produced chore(deps): update core junctions (gnome-build-meta + freedesktop-sdk) dakota#1575, "chore(deps): update core junctions (gnome-build-meta + freedesktop-sdk)" — a PR whose title says exactly what it does.
  • server — build.yml's track-refs job, gated on startsWith(github.head_ref, 'renovate/'), pushing into whatever branch triggered it.

So the fix here doesn't need designing, just porting: move the two just bst source track calls out of build.yml into a scheduled workflow that opens its own PR on its own branch, the way dakota's core-junctions group does. That also disposes of the contents: write-on-a-PR-event concern I raised, because the job stops running on pull_request at all.

Worth noting dakota's tracker also carries a comment at track-bst-sources.yml:196 about bst source track normalising ref format without changing the underlying commit — i.e. it distinguishes the two cases that server's commit message ("resolve BuildStream junction refs to git-describe format") conflates. On this PR it was not a format normalisation: freedesktop-sdk moved 26.08.0 → 26.08.1.

Still requesting changes on the PR as it stands, for the reasons in the review.


Generated by Claude Code

@castrojo

Copy link
Copy Markdown
Contributor

Closing. The advertised change is a verified no-op for this repository, so what remains is an unlabelled OS-base bump sitting behind an automerge label — which is net risk for zero benefit.

The advertised payload delivers nothing here. build.yml pins exactly one file from this repo, projectbluefin/actions/.github/workflows/server-installer-test.yml. Across the digest range 6680f74..b0fbcf6 that file is byte-identical; the single commit in the range touches reusable-promote-squash.yml, docs/skills/factory-operations.md and tests/test_promote_gate_order.py. @hanthor established this in his review and I agree with it. The SHA pin itself is well-formed and complies with docs/skills/ci-tooling.md (full 40-char SHA plus # v1).

What the PR actually contains besides that. Two BuildStream junction bumps that appear in no title, body, or Renovate table:

  • elements/freedesktop-sdk.bst: 26.08.0-0-gdb97cce -> 26.08.1-0-gb02b59f
  • elements/gnome-build-meta.bst: 50.4-0-g8524013 -> 50.4-5-gd78f498

That is an FSDK point release and five commits of gnome-build-meta — effectively the base of the OS — inside a PR labelled as an Actions digest update.

Merging it as-is would turn main red. project.conf pins release-version: "26.08.0", and .github/scripts/check-release-version.py is the first thing just validate runs and fails closed on drift. That is precisely why this PR's own build job died in two seconds at "Validate Element Graph" on commit bc225b1.

Why I am not just deleting the two hunks. They are not authored by Renovate. build.yml's track-refs job fires on every pull_request event for renovate/* branches, re-resolves the junction track: globs, and pushes the result back onto the branch as github-actions[bot]. Any push I make re-injects them. That same bot push is also why this PR shows no checks at all: the runs for head ed0d7bf exist but sit at conclusion: action_required with total_count: 0 jobs, never dispatched, so there are no check runs and no statuses to read.

The real fix is the one @hanthor proposed — move junction tracking to a scheduled workflow that opens its own PR on its own branch, as dakota already does with track-bst-sources.yml — and that is well outside the scope of a digest bump. I have not attempted it here.

No content is lost by closing: the pinned workflow file is unchanged across the range, so the current pin is functionally identical. If Renovate treats this closure as "ignore this update" rather than re-raising, the pin can be refreshed deliberately once track-refs no longer injects junction bumps into dependency PRs.

Filing the track-refs behaviour as its own issue is the follow-up worth doing; it currently affects #176 and #49 in exactly the same way.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants