Repository navigation
ci(extensions): automate weekly GNOME extension submodule bumps - #645
Conversation
There was a problem hiding this comment.
Read the workflow at bcff10c against .gitmodules, build.yml and update-bluefin-parity.yml. The bump mechanism and the no-diff/close path look right; two things don't do what the body says they do.
1. The "new" column is always the old SHA — .github/workflows/update-gnome-extensions.yml:67
new=$(git ls-files -s "$p" | awk '{print $2}') reads the superproject index. git submodule update --remote (line 51) only moves each submodule's worktree; it does not stage the new gitlink (that happens later when create-pull-request runs git add on add-paths). Reproduced in a scratch repo: after submodule update --remote, git ls-files -s s and git ls-tree HEAD s both still return the old commit while git -C s rev-parse HEAD returns the new one. Consequence: old == new for every row, so the table prints (unchanged) in the shell-version column for every extension and the metadata.json lookup at lines 72–76 never runs — the body's main deliverable is empty on exactly the weeks it matters. Fix is one line:
new=$(git -C "$p" rev-parse HEAD)(git diff --quiet at line 80 is unaffected; it does see worktree submodule movement — verified exit 1 in the same scratch repo.)
2. verify-gnome-extensions does not gate the bump PR as described
The PR body says "the image build's verify-gnome-extensions gates the merge." On this workflow it cannot run automatically: verify-gnome-extensions executes inside the image build (Containerfile:263), the PR is created with the default github.token (create-pull-request step, line 88), so GitHub does not fire pull_request workflows for it, and the explicit dispatch at line 108 passes contract_only=true, which skips build_main/build_kernel (build.yml:218,243,268). The only checks that land on the bump branch are the contract gates (just check, check-parity, check-repos). Same pattern as update-bluefin-parity.yml, so this is consistent with the repo — but the PR body generated at line 56 ("The image build on this PR runs verify-gnome-extensions") and this PR's description both promise a check that won't appear unless a human pushes or closes/reopens the bump PR. Suggest either rewording both to say the maintainer must trigger the image build, or dropping contract_only (with the publish implications the build.yml:9–19 comment warns about).
Otherwise: the tmp/… submodule paths are covered by add-paths, actions: write is sufficient for gh workflow run, and the empty body-path pre-write on line 41 is a correct workaround for the no-diff case.
Confidence: 3/5 (needs attention) — 2 medium findings
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Pushed fix to line 67: |
There was a problem hiding this comment.
Reviewed at head b9c103a.
correctness
- High — the bump step fails on every run.
.gitmodulespinsappindicatorsupport@rgcjonas.gmail.comtobranch = v65(.gitmodules:4), but upstreamubuntu/gnome-shell-extension-appindicatorhas no branchv65— only the tag.git submodule update --remote --depth 1 -- $paths(update-gnome-extensions.yml:47) therefore exits 128 withfatal: Unable to find refs/remotes/origin/v65 revision in submodule path, andset -euo pipefail(:32) aborts the job before any PR is proposed. Reproduced locally with a fresh submodule +branch v65. Either skip submodules whosebranchresolves to a tag, or change the.gitmodulesentry to a real branch (and decide whether appindicator should track at all). - Low — gsconnect's metadata lives at
data/metadata.json.in(scripts/verify-gnome-extensions.py:19), sofind "$p" -maxdepth 2 -name metadata.json(:66) never matches and the table column is always?for it;| head -n 1can also pick an unrelatedmetadata.json(tests/fixtures) in other trees.
security
- Medium —
python3 -c "... open('$meta') ..."(update-gnome-extensions.yml:68) interpolates a path discovered byfindinside third-party submodule trees into Python source. A directory name containing')in an upstream extension repo becomes code executed in a job holdingcontents: write/pull-requests: write/actions: write(:17-19). Pass the path assys.argv[1]instead. Theshell-versionstrings (also upstream-controlled) are then written unescaped into the PR body table (:79).
intent-alignment
- Low — body says "the image build's verify-gnome-extensions gates the merge", but a PR opened with
GITHUB_TOKENgets nopull_requestrun (docs/skills/ci-workflows.md:61-63) and the explicit dispatch iscontract_only=true(:109-110), which skips every image job (build.yml:218, :243, :268). Only the source-side checkverify-gnome-extensions.py --sourceviajust check(Justfile:84) runs on the bump branch; the image-side verification (Justfile:180-181) does not run until a human pushes or re-triggers a full build. Worth stating accurately in the body and PR template text (:50-52).
style
- Low —
changed=true/false(:83-87) is never consumed (the propose step has noif), and the "All pinned GNOME extensions match" placeholder (:35) is unconditionally overwritten by the table at :81, so the comment at :32-34 describes behavior that no longer exists. Drop the output or gate the dispatch on it.
docs-currency
- Low — docs/skills/ci-workflows.md:39-73 documents
update-bluefin-parity.ymlincluding the GITHUB_TOKEN/dispatch quirk this workflow copies; the new workflow (and itscontract_onlydispatch) is not added there or to theci-workflowsdescription in docs/skills/index.json:17.
Confidence: 2/5 (needs attention) — 1 high finding; 1 medium finding
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Request changes at b9c103a. Reviewed with Codex. |
b9c103a
ℹ️ Sentinel notice — informational; author is trusted (owner), merge is not blockedHive flagged this PR (author @hanthor, head
Hive did not add the |
castrojo
left a comment
There was a problem hiding this comment.
Reviewed: clean, non-disruptive scheduled workflow to automate weekly GNOME extension bumps via create-pull-request and contract checks. Follows repo conventions.
Dismissing self-review: reviewer has commits on this branch and is disqualified per review policy.
|
Review blockers for the next revision: the scheduled updater currently fails before proposing a PR because |
b9c103a to
a2ca5b0
Compare
There was a problem hiding this comment.
Reviewed at a2ca5b0. The new weekly job will fail on its first run: it cannot see the upstream branch for two of the extensions it is supposed to update, so no bump PR gets opened. One fix needed; everything else checks out.
correctness
.github/workflows/update-gnome-extensions.yml:38-40—git submodule update --init --depth 1makes each submodule a shallow, single-branch clone whose fetch refspec covers only the remote's default branch. The followinggit submodule update --remote --depth 1then looks uprefs/remotes/origin/<branch>and dies for any submodule whose.gitmodulesbranch is not the upstream default. Two currently are:BudsLink-Companion(branch = Gnome-Extension, upstream HEAD ismain,.gitmodules:44) andtilingshell(branch = v18.0, upstream HEAD ismain,.gitmodules:40). Reproduced on a fresh shallow clone of this head with git 2.47:Withfatal: Unable to find refs/remotes/origin/Gnome-Extension revision in submodule path '.../BudsLink-Companion@maniacx.github.com' fatal: Unable to find refs/remotes/origin/v18.0 revision in submodule path '.../tmp/tilingshell@ferrarodomenico.com'set -euo pipefailthe step aborts, so the job is red every Monday and never reachescreate-pull-request. Verified fix: before the--remotecall, fetch the configured branch into its tracking ref for each path, e.g.br=$(git config -f .gitmodules submodule."$p".branch); [ -n "$br" ] && git -C "$p" fetch --depth 1 origin "+refs/heads/$br:refs/remotes/origin/$br"— after that the same--remote --depth 1succeeds and moves BudsLink tof76e866. (Thetests/suite does not cover the workflow shell, so this was not caught byjust test.)
No findings from: security, intent-alignment, style, docs-currency.
Confidence: 3/5 (needs attention) — 1 high finding
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
88767b4
ℹ️ Sentinel notice — informational; author is trusted (owner), merge is not blockedHive flagged this PR (author @hanthor, head
Hive did not add the |
|
Updated at The workflow is new in this PR, so its scheduled/dispatch path can only run from the default branch after merge. The updated PR body correctly states that dispatched CI is contract-only; the image-side extension verifier runs during a normal image build. |
|
Addressed the latest review finding in |
There was a problem hiding this comment.
Weekly bot that moves the bundled GNOME extensions to their latest upstream commits and opens a review PR. The automation itself is wired like the existing parity bot and looks safe; the gaps are in the report it produces and in one line of the docs.
correctness
scripts/gnome-extension-shell-versions.py:14only looks at<ext>/metadata.jsonand<ext>/data/metadata.json.in. Nine of the sixteen submodules keep metadata elsewhere (src/,resources/, nested dirs — seescripts/verify-gnome-extensions.py:11-29, e.g. tilingshellresources/metadata.json; confirmed itsv18.0tree has no rootmetadata.json). Those rows in the bump PR table will always read?, so the "declared GNOME Shell versions" column the PR body promises is missing for most extensions. Reuse theSOURCE_EXTENSIONSmap (or accept a metadata path) instead of guessing two locations.
docs-currency
docs/skills/ci-workflows.md:80says the dispatch runs "withoutreview_vm", but noreview_vminput or symbol exists anywhere in the repo (build.ymlworkflow_dispatch has onlycontract_only/sha). Drop the phrase or name the real input.
style
.github/workflows/update-gnome-extensions.yml:86-88—if git diff --quiet; then exit 0; fiis the last statement of the step, so it changes nothing. The parity sibling (update-bluefin-parity.yml:46-48,82) emits achanged=output there; either do the same or delete the block.
No findings from: security, intent-alignment.
Confidence: 3/5 (needs attention) — 1 medium finding; 2 perspectives requested changes
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Addressed the review notes on the current head: |
4e83f8f
ℹ️ Sentinel notice — informational; author is trusted (owner), merge is not blockedHive flagged this PR (author @hanthor, head
Hive did not add the |
There was a problem hiding this comment.
docs-currency (out-of-scope, pre-existing)
docs/skills/desktop-contract.md:245-251still says.gitmodules"pins eight of them" and lists eight names;.gitmodulesat this head has 16 submodules. Not introduced here, but this PR is the one that makes pin movement automatic, so the section describing the pins is the natural place to point atupdate-gnome-extensions.yml. Backlog, not blocking.
No findings from: correctness, security, intent-alignment, style.
Confidence: 5/5 (safe)
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Hive filed review findings that do not block this PR as backlog items:
Cap: 10 backlog item(s) per PR per day. |
There was a problem hiding this comment.
Reviewed at head 4e83f8f. The workflow, the report script and the .gitmodules change all check out: every configured branch is a real remote head (v65 was an annotated tag, master declares GNOME 51), the new tests and the full tests/run_suite.py pass, and the report script produces correct rows against real submodule checkouts.
docs-currency
docs/skills/ci-workflows.md:32-33,:85and the description at:15/docs/skills/index.json:17describe areview_vm=truedispatch input that "builds an image locally and tests an offline ISO in a disposable VM".build.ymlat this head has noreview_vminput (onlycontract_onlyandsha); it comes from the still-open PR #627. If #627 lands first this is fine, but merged on its own this PR documents a flag that does not exist. Suggest dropping thereview_vm/ "offline ISO review VMs" sentences here and letting #627 add them.
No findings from: correctness, security, intent-alignment, style.
Confidence: 3/5 (needs attention) — 1 perspective requested changes
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Also addressed the latest docs-currency note in |
eccfe1b
ℹ️ Sentinel notice — informational; author is trusted (owner), merge is not blockedHive flagged this PR (author @hanthor, head
Hive did not add the |
|
Retry completed successfully after Codeberg recovered: all three kernel image flavors passed on current head |
No content change. The earlier commit message claimed the weekly update-gnome-extensions workflow was documented; it is not on main yet (proposed in projectbluefin#645). desktop-contract.md correctly states pins are bumped manually. PR title/body updated to match and the projectbluefin#682 closing link removed. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Hive-Fix-Attempt: 2/6 Signed-off-by: scanner <scanner@hive.kubestellar.io>
Danathar
left a comment
There was a problem hiding this comment.
Approving at eccfe1b. Both earlier blockers are fixed:
.gitmodulesnow tracks appindicator'smasterbranch, which exists upstream (fbc758a4). The oldbranch = v65was only a tag, so the weekly update would have failed.- The workflow no longer builds Python code from text found in third-party extension trees. It calls
python3 scripts/gnome-extension-shell-versions.py "$p"with the path as a quoted argument, and top-level permissions are{}.
CI is green on this head. Not run: the weekly job itself.
Track appindicator on its real upstream branch, parse metadata without code interpolation, and document that the dispatched run checks contracts only. Assisted-by: GPT-6 via GitHub Copilot Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
A shallow submodule checkout only has the upstream default branch. Fetch the configured branch into its remote-tracking ref before selecting it, so tilingshell and BudsLink updates work too. Assisted-by: GPT-6 via Codex Co-authored-by: OpenAI Codex <noreply@openai.com>
Reuse SOURCE_EXTENSIONS from verify-gnome-extensions.py so the shell-version column resolves nested metadata (src/, resources/, multi-extension repos) instead of rendering '?'. Drop the reference to a non-existent review_vm input from the CI workflow docs. Hive-Fix-Attempt: 2/6 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: scanner <scanner@hive.kubestellar.io>
eccfe1b to
2638af5
Compare
2638af5
ℹ️ Sentinel notice — informational; author is trusted (owner), merge is not blockedHive flagged this PR (author @hanthor, head
Hive did not add the |
There was a problem hiding this comment.
HUMAN DECISION NEEDED — @hanthor decide what to do with the BudsLink extension: its tracked upstream branch is now a README-only stub, so the very first weekly bump PR this workflow opens will fail the extension contract and block every other bump until someone hand-edits it.
This change adds a scheduled job that moves each bundled GNOME extension to the newest commit on its upstream branch and opens a review PR. I ran the bump step from head 2638af5 locally against the live upstreams. The shell and Python logic work as written (tests pass, report renders correctly), but two of the branches .gitmodules points at do not match what is currently shipped, so the first proposal would be wrong.
correctness
- tailscale-gnome-qs would be downgraded.
.gitmodules:57saysbranch = main, but the pinned gitlink7feced3(2026-07-29, "fix: Mullvad Exit nodes filter broken (#62)") is only reachable from upstreamdevelop, notmain. Upstreammainhead is5307ed1(2026-02-27, "Update main (#12)") and declares shell-version 45–49, versus 45–50 at the current pin. Becausezz3-bluefin-unsupported-stuff.gschema.override:2setsdisable-extension-version-validation=true,verify-gnome-extensions.py:77skips the GNOME 51 check, so the contract-only CI this workflow dispatches (update-gnome-extensions.yml:97-108) would pass and the bump PR would read as green. The PR fixed the same class of mismatch for appindicator (.gitmodules:4) but missed this one. Fix:branch = developfor tailscale (or re-pin frommaindeliberately). The report table (update-gnome-extensions.yml:68-82) shows only the new shell-version list, not the old, so a reviewer of the bump PR has no cue that the range shrank — consider printing both. - BudsLink-Companion upstream branch is empty.
.gitmodules:41tracksGnome-Extension; its headf76e866(2026-09-20) contains onlyREADME.md, which says the extension moved tomaniacx/Bluetooth-Battery-Meter. The bump moves the gitlink there,scripts/gnome-extension-shell-versions.py:49prints?, andverify-gnome-extensions.py:69-70failsjust checkwith "missing …/metadata.json". Since all 16 extensions share one branch/PR (update-gnome-extensions.yml:88-95), that red check also holds back appindicator, blur-my-shell, dash-to-dock, gsconnect and fuzzy-application-search, which all have real new heads today. Needs a maintainer call: pin BudsLink out of the bump loop (e.g. dropbranchand skip it, orupdate = none), replace the submodule, or accept a permanently red first PR.
docs-currency
docs/skills/desktop-contract.md:248says.gitmodulespins all 16 trees "by URL and branch", butfuzzy-application-search@mkhl.codeberg.page(.gitmodules:43-45) has nobranchand the workflow has a dedicated remote-HEAD path for that case (update-gnome-extensions.yml:51-53). Say "by URL and, where set, branch" or add the branch.
No findings from: security, intent-alignment, style.
Confidence: 2/5 (needs attention) — 1 high finding; 1 medium finding
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
No content change. The earlier commit message claimed the weekly update-gnome-extensions workflow was documented; it is not on main yet (proposed in projectbluefin#645). desktop-contract.md correctly states pins are bumped manually. PR title/body updated to match and the projectbluefin#682 closing link removed. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Hive-Fix-Attempt: 2/6 Signed-off-by: scanner <scanner@hive.kubestellar.io>
No content change. The earlier commit message claimed the weekly update-gnome-extensions workflow was documented; it is not on main yet (proposed in projectbluefin#645). desktop-contract.md correctly states pins are bumped manually. PR title/body updated to match and the projectbluefin#682 closing link removed. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Hive-Fix-Attempt: 2/6 Signed-off-by: scanner <scanner@hive.kubestellar.io>
No content change. The earlier commit message claimed the weekly update-gnome-extensions workflow was documented; it is not on main yet (proposed in projectbluefin#645). desktop-contract.md correctly states pins are bumped manually. PR title/body updated to match and the projectbluefin#682 closing link removed. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Hive-Fix-Attempt: 2/6 Signed-off-by: scanner <scanner@hive.kubestellar.io>
* docs(skills): update desktop contract GNOME extension submodules currency Update docs/skills/desktop-contract.md to reflect the current 16 GNOME extension submodules declared in .gitmodules (expanded by Dakota extensions) and document the weekly automated update workflow (.github/workflows/update-gnome-extensions.yml). Regenerate docs/skills/index.json to keep skill catalog metadata current. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Hive-Run: #682 Hive-Plan: docs-currency Hive-Spec: desktop-contract#submodules Signed-off-by: mrbobbytables <mrbobbytables@users.noreply.github.com> * docs(skills): correct scope - weekly workflow pending #645 No content change. The earlier commit message claimed the weekly update-gnome-extensions workflow was documented; it is not on main yet (proposed in #645). desktop-contract.md correctly states pins are bumped manually. PR title/body updated to match and the #682 closing link removed. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Hive-Fix-Attempt: 2/6 Signed-off-by: scanner <scanner@hive.kubestellar.io> * docs(skills): clarify extension updater contract Assisted-by: OpenAI Codex via GitHub Copilot Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * docs(skills): clarify Search Light removal Assisted-by: OpenAI Codex via GitHub Copilot Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * docs(skills): split desktop verifier guidance Move build and installed-image verifier details into a reference page, keeping the skill below its hard line cap. Assisted-by: OpenAI Codex via GitHub Copilot Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --------- Signed-off-by: mrbobbytables <mrbobbytables@users.noreply.github.com> Signed-off-by: scanner <scanner@hive.kubestellar.io> Co-authored-by: mrbobbytables <mrbobbytables@users.noreply.github.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: scanner <scanner@hive.kubestellar.io> Co-authored-by: James Reilly <jreilly1821@gmail.com>
Weekly scheduled run moves each GNOME extension submodule to its configured upstream branch and opens or updates a review. The PR body lists old and new commits plus the extension's declared GNOME Shell versions. Tag-only refs are not valid
submodule.*.branchvalues; appindicator tracks its realmasterbranch, which currently declares GNOME 51 support.The workflow dispatches
build.ymlwithcontract_only=trueafter a proposal changes. That checks source contracts only: it does not build an image or run the image-side extension verifier. A trusted full image build must verify the proposed extension heads before merge. The workflow does not auto-merge.