Repository navigation
fix(renovate): match the freedesktop-sdk series the junction actually tracks - #84
6 commits merged into
Conversation
… tracks
renovate.json filters freedesktop-sdk tags with
"extractVersion": "^freedesktop-sdk-(?<version>25\\.08\\.[0-9]+)$"
while elements/freedesktop-sdk.bst has tracked 26.08 since the bump to
freedesktop-sdk-26.08.0-0-gdb97cce:
track: freedesktop-sdk-26.08*
ref: freedesktop-sdk-26.08.0-0-gdb97cce32cecadc7a3e98f06d557ebfa6ba9ad46
No tag upstream publishes can match that rule any more, so Renovate reports
no updates rather than an error. The junction silently stops receiving bumps
and the track-refs job in build.yml never has a ref to resolve. Nothing in
any log says so — the failure looks exactly like "already up to date".
Bumps the rule to 26.08, and adds check-renovate-series.py so the pair cannot
drift apart again: it reads the series out of each junction's `track:` glob
and asserts the matching extractVersion is anchored to the same one, failing
closed. Same shape as check-release-version.py and check-k0s-version.py —
wired into .pre-commit-config.yaml, and unit-tests.yml now also fires on
renovate.json and elements/*.bst so the guard runs when either side moves.
tests/unit/test_check_renovate_series.py covers the passing case, the exact
regression above, a junction with no packageRule, a junction with no
`track:`, the two series helpers, and the checked-in tree itself.
$ python3 -m pytest tests/unit -q
207 passed, 1 xfailed in 0.64s
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VwEniffvKGH7gHfqKneuaJ
Independent verification — the bug is real, the guard is non-vacuousI can't approve this (same account authored it), so recording the verification instead. Everything below I ran myself rather than taking the description's word for it. The mismatch is real, confirmed against the actual element: # elements/freedesktop-sdk.bst
track: freedesktop-sdk-26.08*
ref: freedesktop-sdk-26.08.0-0-gdb97cce32cecadc7a3e98f06d557ebfa6ba9ad46against The description's framing of the failure mode is the important part and it's correct: an The guard actually catches it. I reintroduced the Clean tree: And it's wired to fire. The
One thing to flag before mergingThe checker is reachable from pre-commit and from the test suite, but not from @castrojo @mrbobbytables — this is clean and needs no approval to merge under current protection, but it's Claude-authored end to end and nobody has reviewed it. Worth a human glance before it lands rather than merging on the strength of its own test suite. The Generated by Claude Code |
`check-renovate-series.py` was wired into pre-commit and the unit suite but not into `just validate`, unlike its siblings `check-release-version.py` and `check-k0s-version.py`. Since AGENTS.md and docs/skills/ci-tooling.md point contributors at `just validate` as the merge-contract check, the new invariant was effectively unenforced there. Assisted-by: Claude Opus 4.6 via GitHub Copilot Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
`bump-fsdk-version.md` credited Renovate with opening FSDK point-release PRs and treated the scheduled tracker as a secondary path. That is backwards: after moving junction tracking out of the `pull_request` path, `track-junctions.yml` is the mechanism that delivers point-release bumps as reviewable PRs. It is also unsupported. A repo-wide PR search for `freedesktop-sdk in:title` returns exactly one result — #84, which fixes the Renovate config rather than bumping anything. Describe the tracker as the primary path and stop claiming Renovate creates FSDK PRs. `renovate.json` is untouched; #84 owns it. Assisted-by: Claude Opus 4.6 via GitHub Copilot Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
castrojo
left a comment
There was a problem hiding this comment.
Approving, after pushing a one-line fix. Explaining it since I am not the author, and flagging one caveat that outlives this PR.
The bug is real and I verified it against main rather than the description. renovate.json on main carries "extractVersion": "^freedesktop-sdk-(?<version>25\\.08\\.[0-9]+)$" while elements/freedesktop-sdk.bst declares track: freedesktop-sdk-26.08* with ref: freedesktop-sdk-26.08.0-0-gdb97cce. The filter targets a series the junction no longer tracks, so it matches nothing.
The guard is non-vacuous and correctly scoped: extractVersion applies to tag names returned by the git-refs datasource, not to the ref: value, and the $ anchor deliberately excludes the git-describe -0-g<sha> suffix that project.conf's ref-format: git-describe forces onto the pin. I read the checker line by line — digits() strips \. and [0-9]+ before extracting groups, so freedesktop-sdk-26.08* -> ['26','08'] matches the corrected regex's ['26','08'], and gnome-50 -> ['50'] matches ^(?<version>50\.[0-9]+)$. Comparing digits only means it would not catch a same-digits wrong-project regex, but the depName-to-junction mapping is explicit, so that is an acceptable bound and consistent with the sibling checkers.
What I changed (commit 312e11d, 2fd5283..312e11d, one line in Justfile): added python3 .github/scripts/check-renovate-series.py to the validate recipe, after its two siblings.
The new invariant was enforced only by pre-commit and pytest, while check-release-version.py and check-k0s-version.py both run under just validate. Since AGENTS.md and docs/skills/ci-tooling.md point contributors at just validate as the merge-contract check, a guard outside it is a guard people will not run. You flagged this yourself ("the unit test is the only enforcement in CI"), so I took it as an in-scope gap rather than a design change.
Proof it actually guards, which matters more than that it runs: temporarily reverting extractVersion to the 25.08 series makes just validate exit 1 with
ERROR: elements/freedesktop-sdk.bst tracks `freedesktop-sdk-26.08*` (series 26.08) but renovate.json filters ... extractVersion '^freedesktop-sdk-(?<version>25\.08\.[0-9]+)$' (series 25.08). Renovate matches no tag, so this junction gets no update PRs.
Restored afterwards; git status --porcelain showed only Justfile. Clean run reports OK: 2 Renovate version filter(s) match their junction track series alongside the other two checkers.
Verified at the new head: build pass (44m33s), build-kernel pass, unit pass. installer-test is the pre-existing main failure and unrelated.
One caveat worth recording, because this PR does not close it and should not be expected to. Fixing the series is necessary but may not be sufficient for Renovate to actually open FSDK PRs. The custom manager's matchStrings capture currentValue from the ref: line — i.e. the git-describe string freedesktop-sdk-26.08.0-0-gdb97cce... — which cannot satisfy the anchored extractVersion either. Consistent with that, a repo-wide search for freedesktop-sdk in:title returns exactly one PR: this one. There is no evidence Renovate has ever opened an FSDK bump here. That deserves its own issue; it is out of scope for a series fix.
Related: I have opened #180, which moves junction tracking out of the pull_request path onto a scheduled workflow with its own PR. That is what currently delivers FSDK bumps in practice — via track-refs injecting them into unrelated Renovate PRs, which is the behaviour #180 removes.
Also correcting one line in your description, no action needed: "the track-refs job in build.yml never gets a ref to resolve" is not right — track-refs re-resolves the track: glob independently of Renovate and has been bumping FSDK all along. That is precisely how #49 and #176 acquired their undisclosed junction hunks.
Approving. You are the author so you cannot self-approve; this needs one more independent approval.
ConfirmDelivered keyed undelivered kicks by their first perspective. A combined kick writes one pending entry per perspective, so when the reviewer was busy and the kick was not delivered only the first entry was released and re-sent alone; the rest stayed pending forever — never re-dispatched, never judged, holding unanimity open. Observed on projectbluefin/server#84: security re-dispatched solo ("no findings from any of the 1 perspectives") while intent-alignment, style and docs-currency sat pending from the missed kick. Recent entries are now recorded per perspective too, so the relay can bind a late verdict for any perspective of a combined kick. The all-clean comment line names the perspectives instead of counting them. Signed-off-by: Andrew Anderson <andy@clubanderson.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
Reviewed and approved for merge — blocked only on an OAuth token scope, not on content.
The maintainer token currently carries Merge order matters here, because these four collide in
After that sequence the merged tree passes: A direct push of that merge is refused by the |
c3bd1fe
The bug
renovate.jsonfilters freedesktop-sdk tags with a rule anchored to the 25.08 series:elements/freedesktop-sdk.bsthas been on 26.08 since the bump tofreedesktop-sdk-26.08.0-0-gdb97cce:No tag upstream publishes can satisfy that rule any more. Renovate does not error on an
extractVersionthat matches nothing — it reports no updates. So the junction has silently stopped receiving bump PRs, and thetrack-refsjob inbuild.ymlnever gets a ref to resolve. The failure mode is indistinguishable from "already up to date", which is why it is easy to miss.The fix
Bump the rule to 26.08, and stop the two from drifting again.
.github/scripts/check-renovate-series.pyreads the series out of each junction'strack:glob and asserts the matchingextractVersionis anchored to the same series, failing closed:It is deliberately built in the same shape as
check-release-version.pyandcheck-k0s-version.py— a single invariant, a docstring explaining the silent failure it prevents, fail-closed, no dependencies beyond the stdlib. Wired into.pre-commit-config.yamlnext to the other two, andunit-tests.ymlnow also triggers onrenovate.jsonandelements/*.bstso the guard actually runs when either side of the invariant moves.The gnome-build-meta rule already agrees with its junction (
track: gnome-50/^(?<version>50\.[0-9]+)$) and is unchanged; the check now holds it there.Tests
tests/unit/test_check_renovate_series.pycovers the passing case, the exact regression above, a junction with nopackageRule, a junction with notrack:, both series helpers, and the checked-in tree itself so drift fails CI rather than only pre-commit.Note on scope
I came at this from a sibling BuildStream image repo and only touched the Renovate invariant. One thing I noticed but deliberately left alone:
just show-me-the-futureandjust install-vmgive you a real QEMU install-and-reboot test, and no workflow calls either, somainhas no automated boot gate. Happy to wire that up as a separate PR if it would be welcome.🤖 Generated with Claude Code
https://claude.ai/code/session_01VwEniffvKGH7gHfqKneuaJ