Repository navigation
🐛 fix: stop inferring idle external PR owners from missing Hive actors - #16037
Conversation
|
[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
|
There was a problem hiding this comment.
The core change reads correctly: stageTrackerExternalPRWork (src/pkg/dashboard/static-src/js/tracker-row.js:227-236) resolves the visible covering PR through the same stageTrackerLinkedDrawnHere / stageTrackerLinkedSupersededGone predicates the row layout uses, so an inline row never falls back to the issue filer, and kind === 'hive' || 'bot' keeps the existing routing words. The new inline/nested regressions in stage_tracker_actor_icon_test.go cover that.
One medium finding — the tooltip now discards the server's why for every none-actor state on an external PR, not just the author-unknown one.
stageTrackerActorTip (tracker-row.js:242-246) returns the "current work activity is unknown" sentence before it ever looks at sa.why. The server sets stage_actor.kind = "none" with a specific reason in several states that are not "the hive has no evidence":
src/pkg/dashboard/stage_tracker_actor.go:151—on hold — the owner paused it on purpose; nothing moves until the hold is liftedstage_tracker_actor.go:170—blocked on PR #N — nothing moves until the dependency closesstage_tracker_actor.go:173-174—draft — the author has not marked it ready for reviewsrc/pkg/dashboard/stage_tracker_fixer.go:43-47—nobody — <fix exclusion>; the hive will not re-kick it — a person decides(fixer exhausted)
For a human- or clanker-authored PR in any of those states the hover now reads Author-owned PR @alice — current work activity is unknown and the hold / block / exhausted-fixer explanation is gone. The diff's own test shows the same loss on the CI-running path: the PR 506 assertion was rewritten from STAGE_TRACKER_NOBODY_LEAD + ' — CI is still running' to the activity-unknown sentence (stage_tracker_actor_icon_test.go:81), so a PR whose CI is simply running is described as having unknown ownership activity. The Fixing word at tracker-row.js:295-296 has the same shape: a held external PR at 070 now reads CI failing · author-owned with no trace of the hold.
Suggested shape: keep sa.why as the trailing clause (' · Author-owned PR @' + author + ' — ' + (sa.why || 'current work activity is unknown')), or only substitute the activity-unknown text when sa.why is the author-unknown reason. Either keeps the fix and restores the PR 506 / hold / blocked wording.
Not blocking on it, but noting: this branch and #16036 both edit tracker-row.js and the generated static/index.html (hence conflicts-with-open-pr); whichever lands second needs a regen of index.html.
— hive: agent=reviewer-2 backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Hive status — updated in place; edits do not notify. Full history is in the Activity Monitor. @clubanderson this PR is held at Why:
Next: the hive does not push to your branch. Push a new head (or reply here); CI and the review swarm re-run automatically and the PR returns to the normal path. hive/awaiting-ci: rebased onto v5 (63cefc0 → e6e1db8) to clear the conflict in src/docs/dashboard-sections/activity-monitor.md; kept both the PR's change and the v5 side (both paragraphs retained). The interim merge-from-v5 commit was linearized away; the net PR diff is unchanged. Authorship and sign-off unchanged. CI pending — sweep will merge when green. 🐝 Hive Agent: — hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88 Review findings from Read Finding — the server's
Verified, no action needed:
Minor doc note: |
|
Actual terminal blocker on e9f7fb4: gosec replacement job124543579309/run41202371411 attempt2 FAILED13:10:59Z. GitHub again annotates self-hosted runner lost communication; Scan still marked in progress, no reported scan finding, logs404. This is the automatic github-actions[bot] retry, not my manual retry. All other current source/Node/Go/lint/docs/image checks including ARM passed. Source merge is authorized after genuine exact-head green, but I will not bypass this real security CI gate or retry until green. Independent review/code corrections remain complete; no deployment. |
|
Bounded infrastructure recovery retry rationale: a NEW worker553844 (hive-runners-lke-qh55w-runner-skb4w) completed gosec SUCCESS13:12:41Z in current961 scan job124545716533, after the second loss13:10:59Z. Production .go files on e9f7 are identical to independently green f784; only *_test.go files differ. The repository runner listing does not expose the LKE organization pool, so failed-worker removal is NOT independently established and is not a retry premise. I am rerunning the failed security job ONCE against unchanged e9f7 using the actual successful replacement-worker scan as recovery evidence, not changing gates or looping retries. If the replacement fails again, report blocked. No source change/deployment. |
|
FINAL exact-head CI receipt e9f7fb4: all20 latest applicable workflow/event groups terminal SUCCESS/SKIPPED, all current substantive checks pass/skip (four setup-cancelled DCO/policy checks superseded by later same-head successful runs). Recovery-backed single retry completed gosec SUCCESS13:24:22Z, job124548367833/run41202371411 attempt3, new worker554172. Prior runner-loss failures remain documented; no scan finding was suppressed. All Go/Node/lint/docs/images/ARM green. No real human hold; proceeding under explicit guarded source-merge authorization, no deployment. |
|
Guarded source merge on tested e9f7 correctly refused a real merge conflict; no auto/queue flag used. Resolved ONLY the Activity Monitor docs paragraph conflict by preserving both landed terminal semantics and author-owned wording, and regenerated the combined SPA from sources. New signed head9334d44d3d4ea50dde344999dd4addbc35984b1d integrates current v5; PR diff remains exactly6 standalone files (fragment/docs/actor fixtures/prelude/row JS/generated index), no inherited workflow or scheduler changes. Prior20-group green e9f7 verdict is not transferred. Owning full fresh CI on9334 before authorized source merge; no local suites or deployment. |
|
Current signed head63cefc00512897597cf23c228226b6fdb7638065 also covers one tightly coupled truth case: an authorless PR with an existing typed claimant now says owner recorded (activity still unknown), not owner unknown. Uses the existing claimant seam; inline ownership stays the PR holder, never the issue holder. Narrow independent review found no significant issue; root/inline regressions added. PR diff still exactly6 standalone files. Fresh CI is required on63cefc, no carried-forward e9f7/9334 green. |
Signed-off-by: Andy Anderson <andy@clubanderson.com>
Signed-off-by: Andy Anderson <andy@clubanderson.com>
63cefc0 to
8570191
Compare
|
Correction to the comment above: the rebased head is 8570191 (63cefc0 → 8570191), not e6e1db8. The resolution is the same and CI is pending. 🐝 Hive Agent: — hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88 |
|
@clubanderson this PR is back in your court (
Once the new head lands, CI and the review swarm re-run automatically and the PR returns to the normal path. |
There was a problem hiding this comment.
Read tracker-row.js, tracker-row-layout.js and tracker-words.js on head 63cefc0 against v5. Looks correct to me; one informational note.
What I checked:
static-src/js/tracker-row.js:70—stageTrackerFixingWordis only applied whencur === 70, so the new "CI failing · author-owned / owner unknown / owner recorded" word cannot appear on a CI-running (stage 50) row; those only pick up the new tooltip wording.static-src/js/tracker-row.js:227-239— for avia_prissue row the code readsstageTrackerClaimant(item), which looked like it would pick up the issue's holder. It does not:tracker-row-layout.js:491builds thevia_prshownitem withclaimant: lp.claimant || null, so the holder is the PR's, which is whatTestStageTrackerActorIcon's@pr-holder/ not-@issue-holderassertion relies on.tracker-words.js:469-476—stageTrackerReturnedAuthoralready filters linked PRs bylp.author, so an author-less inline PR falls through to{}and yields "owner unknown" rather than the issue filer, matching the human/hive filer regressions.static/index.htmlhunks are byte-identical to thestatic-srcones.
Info (not a defect): for an external PR with stage_actor.kind === 'none', stageTrackerActorTip (tracker-row.js:244-250) now returns before using sa.why, so a server-supplied reason such as "CI is still running" is no longer shown in the tooltip — the updated test at stage_tracker_actor_icon_test.go:85 asserts exactly that. If you want to keep the server's reason, appending — + sa.why to the author-owned text would do it; fine to leave as is if the loss is intended.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Fresh-head attribution: GitHub HeadRefForcePushedEvent records hivecommons-hive force-pushing63cefc→857019196190b3836c15d2859f879412e625247d at13:40:58Z. This is a rebase, not my push: author Andy preserved, committer scanner. Direct tree comparison63cefc vs857019 is EMPTY; PR scope remains6 standalone files. Watching full fresh CI on857019; no old-head green or review proof transferred. |
Signed-off-by: Andy Anderson <andy@clubanderson.com>
|
Actual CI finding fixed, not retried: TestStageTrackerActorIconKindsJS detour extractor required the exact class ham-detour and immediate closing span, whereas existing recorded-claimant markup legitimately adds ham-has-holder and nested ham-holder. The narrow extractor now reads the leading word with that optional exact modifier. Expected owner-recorded text and exact PR-owner/never-issue-holder tooltip assertions remain unchanged. Independent read-only code-review found no significant issue. DCO-signed correction pushed; fresh entire exact-head CI required. No local suite/build/lint/docs guards run. |
|
@clubanderson this PR is back in your court (
Once the new head lands, CI and the review swarm re-run automatically and the PR returns to the normal path. |
|
Thank you for your contribution! Your PR has been merged. We'd love to hear how your experience was: share feedback |
|
FINAL source receipt: explicit authorized admin squash guarded with --match-head-commit0cee1be3d05b22929216692221d901edd4550f24 merged2026-10-31T14:10:48Z byclubanderson as6fc5147b85930d4839e68a16f618e98fb8f7d175. Before merge: all77 exact-head CheckRuns pass/skip, all16 latest workflow/event groups terminal success/skipped, actual no human hold verified. v2 Tests41207327418/v2 CI41207327401 include current ownership fixtures; security41207327452 and ARM/image41207327423 passed. Independent reviews fixed inline unknown-author/issue-filer leakage and recorded-owner completeness; actual new-head CI exposed exact-class fixture extractor, fixed in signed0cee1 rather than retrying. DCO trailer and fresh origin/v5 ancestry verified. Canonical generation on landed squash unchanged, preserving informational sentinel and repo-scoped merge banner/verified terminal semantics. Owned worktree and both old/current local branches cleaned; remote branch auto-deletion verified. No deployment/live configuration/policy activation. |
Change
An absent Hive stage actor is not evidence that an externally authored PR is abandoned. Failing external PRs now read CI failing · author-owned, or owner unknown when the PR author is unavailable. Tooltips explicitly leave current work activity unknown. Conflicts, actual CI failure and explicit Hive/bot routing remain intact; no production knowledge of local CLI agents is invented.
Identity and review
Root, inline and nested rows resolve the actual visible PR author, never the issue filer. Independent review caught the unknown-inline-author fallback to the issue filer; fixed by using the existing canonical visible-child predicates and added human/Hive issue-filer regressions. Real unclaimed Hive and bot controls remain covered.
Validation
DCO commit e9f7fb4. Only changed-file gofmt and required SPA generation locally, per AGENTS. CI is the sole verdict; exact-head terminal results will be recorded. No deployment, policy activation or manual merge.