Repository navigation
π fix(blocked): an unmarked block becomes a needs-human ask; markers in comments name the blocker - #16832
Conversation
β¦in comments name the blocker A contributor relay's blocked verdict that named no issue or PR left `blocked` with no hive:blocked-on marker: lifted, it looped (projectbluefin/utah#771, 19 cycles); kept, it sat in no inbox ("labelled blocked β the label records no reason"). - blocked sweep: a hive-applied marker-less block (not review backlog, no open native link) gets needs-human, loses blocked and help wanted, and one note naming the two ways out; a person's block is left alone. - hub: a blocked verdict whose reason is empty or only the relay footer is booked idle and its label released (projectbluefin/utah#1002). - a blocked-on marker the sweep reads in a comment names the row's reason and the band signal (projectbluefin/finpilot#624, hivecommons#390); the band reads the reason instead of saying "records no reason". Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com>
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: 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 |
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of
|
|
Hive status β updated in place; edits do not notify. Full history is in the Activity Monitor. Hive filed review findings that do not block this PR as backlog items:
Cap: 10 backlog item(s) per PR per day. Review findings from Docs currency (read at head 53b26fc) The
Suggested patch (sentence-level, both lines are single long lines): - **Lifting the label stays human**: the relay never removes it, so the admission gate holds the issue past the cooldown until someone who knows the dependency has cleared says so.
+ **The relay never lifts the label.** A verdict whose reason names an issue or PR carries a `<!-- hive:blocked-on β¦ -->` marker and is lifted by the blocked sweep when that item merges or closes; a verdict that names nothing is turned by the same sweep into a `needs-human` ask (`needs-human` added, `blocked` removed, one `<!-- hive:unblocked -->` note), so the issue reaches an inbox instead of sitting unlifted.and, appended to the + A `blocked` verdict whose reason is empty, or only HTML comments and the relay's `---` / π footer lines (`isBlockedVerdictReasonless`), is demoted the same way: booked idle, label released, one `<!-- hive:blocked-released kind=no-reason -->` comment saying the agent gave no reason (projectbluefin/utah#1002).Review findings from Style / maintainability review (head 1. +// hiveAppliedBlocked reports the actor that last applied `blocked` and
+// whether it was this hive; errors are counted and a missing hive identity
+// is warned once so the stuck block stays visible.
+func hiveAppliedBlocked(ctx context.Context, client *Client, t blockedIssueTarget, res *BlockedIssueResult, logger *slog.Logger) (string, bool) {
+ actor, found, err := client.blockedLabelActor(ctx, t.repo, t.number)
+ if err != nil {
+ res.Errors++
+ logger.Warn("blocked sweep: timeline lookup failed; marker-less block left alone", "repo", t.repo, "issue", t.number, "error", err)
+ return "", false
+ }
+ if !found {
+ return "", false
+ }
+ if !client.isOwnBotLogin(actor) && !client.IsHiveAuthorLogin(actor) {
+ if len(client.hiveBotLogins()) == 0 {
+ logger.Warn("blocked sweep: marker-less blocked label with no hive identity to compare against; left alone", "repo", t.repo, "issue", t.number, "actor", actor)
+ }
+ return actor, false
+ }
+ return actor, true
+}and both functions become 2. 3. Misattached doc comment β Minor, same files: |
There was a problem hiding this comment.
Correctness review (head 53b26fc) β no in-scope blocker; one advisory on the writeback gate.
Advisory (low): the conversion's gate refusals are silent and pre-recorded β src/pkg/github/issue_blocked.go:666
askHumanBlockedMarkerless asks WritebackAllowed(β¦, KindGateComment, blockedMarkerlessAskNote) with a constant preview. KindGateComment is in ContentDedupeKinds (src/pkg/writeback/dedupe.go:84), and contentCheckLocked records the write at the moment it allows it (dedupe.go:271-273), before AddLabels/unblockIssue run. Two consequences:
- Burst hold (
DefaultBurstItems = 2,DefaultBurstWindow = 60s,dedupe.go:30-33): on a repo with several parked relay blocks (the body lists four on projectbluefin/testsuite), only the first converts per sweep; the others returnReasonBurstandaskHumanBlockedMarkerlessreturnsfalsewith no log and no counter, falling through tostripBlockedHelpWanted. They do convert on later sweeps, so the body's "become asks on the first sweep after deploy" is one-per-repo-per-sweep in practice. - If
AddLabelsorunblockIssuefails after the gate allowed, the retry withinsame_item_ttl(24h) is refused asReasonDuplicate, again silently.
Neither loses data β both self-heal β but the silence hides the delay from operators. Suggested: log the refusal like the sibling callers do.
if d := client.WritebackAllowed(owner+"/"+name, item, writeback.KindGateComment, blockedMarkerlessAskNote); !d.Allowed {
logger.Info("blocked sweep: needs-human ask held by writeback gate", "repo", t.repo, "issue", t.number, "reason", d.Reason, "detail", d.Detail)
return false
}
Verified against the tree: needs-human does reach the inbox as your_turn via parkedOwnership (src/pkg/dashboard/ownership.go:417-419); the open-native-link and person-applied guards hold (issue_blocked.go:647-657); blockedOnSeen is mutex-guarded and cleared in unblockIssue (item_refresh_hook.go:53); test ordering assertion (needs-human before blocked removal) matches the code.
β hive: agent=adjudicator-3 backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
@Danathar 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. Queued in your relay fix queue (6 waiting): fixes are handed out before new work as soon as you have capacity. |
Signed-off-by: Danathar <Danathar@users.noreply.github.com>
Adjudicator advisory on hivecommons#16832: a burst hold (2 items / 60s) or a duplicate after a failed write within same_item_ttl made askHumanBlockedMarkerless fall through silently. Log repo, issue, reason and detail; the conversion retries on a later sweep. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com>
|
Thanks, the advisory was right. Fixed in the latest commit (v6 merged in first):
|
There was a problem hiding this comment.
Read the sweep branch and the dashboard reclassifier against the tree at merge-base. The sweep change looks correct to me; one low-severity note on the reason-less detector.
Verified:
- The new ask sits inside the existing
scanned && t.recheckAfter.IsZero() && !t.opsTrackingguard and thelookups < blockedSweepCommentLookupsbudget (src/pkg/github/issue_blocked.go:449-481), andaskHumanBlockedMarkerless(:646) takes!t.reviewBacklogwhileliftBlockedMarkerless(:598) takest.reviewBacklog, so the two are disjoint and the review-backlog lift is unchanged. - Partial-failure ordering converges:
needs-humanis added beforeunblockIssue; if the unblock fails the next sweep sees the label already present (issueHasLabel(t.labels, issueNeedsHumanLabel),:671) and retries only the removal.needs-humanis already a filter case inclient.go:5284, so the converted issue leaves the agent queue as the comment promises. unblockIssuenow clears theblockedOnSeenentry (item_refresh_hook.go:53), so a lifted issue does not keep reporting a stale "on owner/repo#N" reason from the process-global map.go build ./pkg/github/ ./pkg/dashboard/passes on the head.
Low β footer regex can swallow a real reason. blockedVerdictFooterLineRE at src/pkg/dashboard/contribute_blocked_tooling.go:77 treats any line containing π or **Hive Agent** as footer. A one-line verdict such as **Hive Agent** blocked: waits on owner/repo#12 to merge is stripped to nothing, isBlockedVerdictReasonless returns true, and the verdict is demoted to idle with a "gave no reason" release comment even though it named its blocker. Anchoring the pattern to the footer's actual shape (line starts with π / the --- separator) would avoid it. Not blocking β the failure mode is "stays actionable", not a stuck block β but it is the inverse of the bug the PR fixes, so worth tightening.
β hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
Executive summary
When a contributor's agent reports an issue "blocked" without naming what it waits on, the hive leaves the
blockedlabel with nohive:blocked-onmarker. Nothing can ever lift that.your_turn=false(feat(gnome): prove a genuine GNOME OS guest lane for extension validationΒ projectbluefin/testsuite#908, fix(security): critical CVE detected in testing buildΒ projectbluefin/utah#1002).This PR:
blockedthe hive applied becomes aneeds-humanask in the inbox.What
blockedstuck, no inbox, "records no reason"needs-humanadded,blockedremoved, one note with the two ways outblockedstuckblockedpr=projectbluefin/actions#697)blocked; band "records no reason"blocked on projectbluefin/actions#697; band names itWhy
blockedat 04:17:25Z with no marker.blockedFilterReason) reads only the body, and the band signal (IssueBand) read only native "blocked by" links. A marker in a comment, or even in the body, never reached the band text.How
issue_blocked.go,askHumanBlockedMarkerless: a hive-applied unmarked block that is not a review-backlog item and has no open native link is gated byWritebackAllowed(KindGateComment). The steps are:needs-human(before removingblocked, so a failed add leaves the issue parked);blocked;help wanted;<!-- hive:unblocked -->note.A person's label is never converted. Converted issues are counted in
BlockedIssueResult.AskedHuman.issue_blocked.go, comment markers: the blocked sweep remembers the newest comment marker it reads (blockedOnSeen).blockedFilterReasonForgives it precedence over the body, the same precedence the sweep lifts by.unblockIssueforgets it.contribute_blocked_tooling.go:isBlockedVerdictReasonlessstrips HTML comments,---and π footer lines, and finds no letter or digit.reclassifyToolingBlockedVerdictthen books the verdict idle, and the release comment says "gave no reason" under its own marker.bands.go/overview_ownership.go:blockedReasonSignalturns the row reason intoblocked on owner/repo#N β lifted when it merges or closesorblocked until <t> β re-checked then, replacing the no-reason signal.Files
src/pkg/github/issue_blocked.goblockedFilterReasonForsrc/pkg/github/client.go,webhook_item_patch.gosrc/pkg/github/item_refresh_hook.gounblockIssueforgets itsrc/pkg/dashboard/contribute_blocked_tooling.gosrc/pkg/dashboard/contribute_ws.gosrc/pkg/dashboard/bands.go,overview_ownership.gosrc/docs/labels-and-control-signals/reference.md, changelogblockeddocsTests
TestSweepBlockedIssuesAsksHumanOnRelayMarkerlessBlockTestReconcileBlockedIssues(updated)needs-humanprecedes the removalTestEnumerateActionableRecordsBlockedRecheckAfter(extended)TestUnblockForgetsCommentMarker,TestBlockedMarkerlessAskNoteTestBlockedVerdictReasonlessIsReleased#450and real text stay blockedTestBlockedRowSignalNamesMarker,TestBlockedReasonSignalgo testgithub/..., writeback, dashboard, cmd/hivecheck-docs-citations.py src/docsNotes
origin/v6..origin/v5has nothing for these files.π€ Generated with Claude Code
https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1