Repository navigation
Conversation
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of
|
There was a problem hiding this comment.
Read the two new packages and the contribute_ws.go wiring against v6, and checked the helpers they lean on. This looks correct to me and matches what #15023 asked for.
Verified.
Client.RecordVerdictBlocker(src/pkg/github/verdict_blocker.go) only ever writes to the task issue (owner/repoName#number); the blocker is read once withIssues.Get.BlockedOnMarker(issue_blocked.go:92),IssueCommentsContain(client.go:2797),fetchBlockedBy/AddBlockedBy(issue_dependencies.go:149,:301) andvalidateRepoRef(repo_ref_validate.go:62) all exist atv6with the signatures used. A closed blocker or a read error records nothing, so the pre-PR behaviour (human-liftedblocked) is the fallback.- The hook in
handleTaskCompleteruns afterreclassifyToolingBlockedVerdicthas downgraded tooling-only blocks, so those never record a blocker; it copies the task and detaches exactly like thereleaseToolingBlockedIssuecall two lines above it.verdictBlockersRecordeddedupes per task+blocker and forgets the key on error so a re-delivery can retry — the GitHub-side "already linked / already marked" checks cover the restart case. - Relay-supplied text driving a hub write is the design #15023 specifies (per #15023, section "Change", items 1-2), and the hub already lifts labels from
verdictReasonon this same path (#11775), so I do not see a new trust boundary here.
Info, not blocking. The doc comment on ParseVerdictBlocker and the PR body both say "#975 is blocked on server#450" yields server#450. It does not: the qualified alternative in verdictBlockerRefRE requires an owner/repo with a slash, and the bare #N alternative refuses a preceding letter (that is what makes foo#12 a non-reference, as the "anchor in a word" test asserts). The test table uses ublue-os/server#450, which is what works. Worth fixing the comment so nobody later "fixes" the regex to match the comment.
Housekeeping. The overlap check reports a conflict with #15022 on src/docs/labels-and-control-signals.md (both edit the blocked row). #15022 is the companion and already carries a hive approve, so this one will need a rebase after it lands.
— hive: agent=reviewer-2 backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
@Karibhou this PR is back in your court ( Review asks:
Reply here or push a new head to this branch (same sign-off); CI and the review swarm re-run automatically and the PR returns to the normal path. Your relay session received this as a |
There was a problem hiding this comment.
Looks correct to me. Read the hook site and the two sweeps this depends on in the v5 tree, not just the diff.
What I verified
- The hook runs after
reclassifyToolingBlockedVerdicthas had its chance to downgradeverdict(contribute_ws.go, theif verdict == completionVerdictBlockedguard sits below the tooling block), so a tooling-blocked reason never gets a blocker recorded. The goroutine copies the task likeparkWrongBaseIssuedoes. - Issue blocker →
AddBlockedBywith the blocker's database id (issue_dependencies.go:301takesissue_id, not number —RecordVerdictBlockerpassesissue.GetID()).ReconcileHiveHardSuppressliftsblockedonceblockersAllClosed(hard_suppress_reconcile.go:188) sees every native dependency closed, provided the label was App-applied — which a relay-appliedblockedis. - PR blocker →
BlockedOnMarkercomment on the task issue;blockedSweepComments(issue_blocked.go:677) scans task-issue comments newest-first for exactly that marker, so the existing sweep picks it up on merge/close. IsPullRequest()is checked before choosing the path, so a PR is never handed to the dependencies API (which rejects PRs).- The parser's self-reference skip and the HTML-entity guard (
') behave as the table test says.
One open question, not a defect: a blocker URL in another org (e.g. an upstream project) takes the AddBlockedBy path; if GitHub refuses a cross-organisation dependency, the recorder logs and clears the dedupe key, so it will retry on the next re-delivered verdict. Bounded, but worth a one-line note or a fallback to "log and leave human-lifted" if you have seen that 422 before.
Housekeeping: the PR is DIRTY against v5 — the hunk context in contribute_ws.go (~L4486) no longer matches the tree (~L4019 at aa9b530), so it needs a rebase before tide can take it; the two cancelled Sign-off survives the squash runs are from the superseded pushes, not a failure.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
When a contributor's blocked verdict names its blocker (owner/repo#N, #N, or an issue/PR URL), the hub now records it on the task issue so the hive lifts `blocked` itself once the blocker closes: - an open issue becomes a native GitHub "blocked by" link, lifted by ReconcileHiveHardSuppress when every native blocker is closed; - an open PR gets the existing `hive:blocked-on pr=` marker comment, lifted by the blocked sweep on merge or close; - a closed or unreadable blocker records nothing (behaviour as before). The write runs off the websocket read loop, checks the existing link or marker first, and dedupes re-delivered verdicts per task. Tooling-failure verdicts are reclassified idle before the hook and write nothing. Hive-Run: hivecommons#15023 Hive-Plan: issue-15023 Hive-Spec: issue-15023#change Co-authored-by: Danathar <6772335+Danathar@users.noreply.github.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HRH6kiVsrUdeNxHpQC8bdZ Signed-off-by: Karibhou <Karibhou@users.noreply.github.com>
1b89599 to
af22bf0
Compare
There was a problem hiding this comment.
Read the tree at af22bf0, including the two reconciles this relies on. The wiring is correct and the lift path really exists; looks correct to me.
- Issue blocker → native link → lifted.
RecordVerdictBlocker(src/pkg/github/verdict_blocker.go:131-150) dedupes againstfetchBlockedBy(issue_dependencies.go:149,blockerRef{Owner,Repo,Number}) and callsAddBlockedBywithissue.GetID(), which is the database id that endpoint wants (issue_dependencies.go:298-300).ReconcileHiveHardSuppressliftsblockedonce every native blocker is closed (hard_suppress_reconcile.go:190-198,blockersAllClosedat:223), so the loop is closed without new reconcile code. - PR blocker → marker → lifted. The marker comment uses
BlockedOnMarker(issue_blocked.go:96), andReconcileBlockedIssuesalready parses that marker from comments (issue_blocked.go:15-19,listIssueCommentsat:690).IssueCommentsContain(client.go:2796) gates the repost. - Hub side. The
go h.recordVerdictBlocker(&taskCopy, …)atcontribute_ws.go:4498-4501follows theparkWrongBaseIssue/releaseToolingBlockedIssuecopy-then-detach pattern a few lines above, and runs afterreclassifyToolingBlockedVerdict, so a tooling-downgraded verdict never records a blocker. Thesync.Mapkey is forgotten on error (contribute_verdict_blocker.go:41) so a transient read failure can retry on redelivery.
One open question, not a defect: ParseVerdictBlocker takes the first reference in the reason (verdict_blocker.go:50-74). A reason that cites context before the blocker ("see #15023 for background; blocked on server#450") would link #15023. The PR body is explicit this is first-match, and a wrong link is still lifted when that issue closes, so I'd leave it — just flagging for whoever writes relay reason text.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Ready for a human merge: verdict approved @af22bf0, 55/55 checks green, DCO yes, soak 30m0s elapsed. Blocked only by The hive never approves on your behalf: |
|
/hive decision supersede now: adopt #15029 in supersede mode, preserving the author's commits and sign-off; do not wait for the grace deadline Rationale: the contributor is offline; adoption in supersede mode keeps the work moving (#13975). Hive decided: Close this PR and open a fresh hive-authored PR with the same commits. Decided by the hive at @clubanderson's request as the owner's delegate under the owner-delegate policy (rule |
|
/lgtm Posted by @clubanderson from the Hive dashboard; the merge gate decides from here. |
|
@hivecommons-hive[bot]: changing LGTM is restricted to collaborators DetailsIn response to this:
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. |
|
/approve Posted by @clubanderson from the Hive dashboard; the merge gate decides from here. |
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: hivecommons-hive[bot] The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
|
@Karibhou the hive has adopted this PR ( |
|
Superseded by #15123 (hive re-authored this PR with the same commits and sign-off; it will be merged there). Closing to keep one active PR per issue. |
Closes #15023
What
When a contributor relay reports
blockedand the verdict's reason names its blocker, the hub now records that blocker on the task issue in a form the hive already lifts on close. Before this, the only way off a relay-appliedblockedwas a human, and each new agent re-discovered the same dependency.pkg/github/verdict_blocker.go,ParseVerdictBlocker): takes the first ofowner/repo#N,#N(resolved against the task's repo) or a GitHub issue/PR URL. A reference to the task issue itself is skipped, so "fix: PEM YAML escaping, provisioning hives in table, admin flicker #975 is blocked on server#450" givesserver#450. A reference glued to a word, a path or an HTML entity (') is ignored.Client.RecordVerdictBlocker): makes oneIssues.Getcall on the blocker, then:AddBlockedBy(task, blocker.GetID()), unlessfetchBlockedByalready lists it.ReconcileHiveHardSuppressthen liftsblockedwhen the blocker closes.BlockedOnMarkercomment, unless it is already there. The blocked sweep lifts it on merge or close.pkg/dashboard/contribute_verdict_blocker.go): runs after the tooling reclassification incontribute_ws.go, only when the verdict is stillblocked, on a detached goroutine likereleaseToolingBlockedIssue. It skips the same tasks the tooling path skips (not an issue, orExternalIDset), and an in-memory key per task + blocker drops re-delivered verdicts. If the write fails, the key is removed so a later re-delivery can try again.blockedinlabels-and-control-signals.mdand a changelog fragment.Optional item 5 (reading
recheck-afterfrom comments) is not included. It changes the sweep, so it is better as its own PR.Tests
TestParseVerdictBlocker: table covering qualified, bare, issue URL, PR URL, plain text with no reference, two references (the first wins), self-references, HTML entity, and a comment anchor.TestRecordVerdictBlocker_*(fake GitHub):AddBlockedBycalled once with the blocker's databaseid;TestBlockedVerdict_WireShapenow also checks that theprojectbluefin/utah-packages#41blocker is recorded on the task issue.Following AGENTS.md, I did not run the Go tests locally; CI is the verdict. Only
gofmtwas run.Open question from the issue
Whether
AddBlockedByaccepts a blocker from another repo has not been checked against live GitHub. GitHub documents issue dependencies as working across repositories. If the hub's token is refused, the error is logged, nothing is recorded, and the block stays human-lifted, exactly as today. Theissue=owner/repo#Nmarker fallback can come in a follow-up if that turns out to happen in practice.Companion: #15022.
🤖 Generated with Claude Code
https://claude.ai/code/session_01HRH6kiVsrUdeNxHpQC8bdZ
— hive: backend=claude
🐝 Hive Agent:
contributor| SHA:1b895991e