Repository navigation
π automerge: refuse an unsigned squash only where the repo requires DCO - #18667
Conversation
Since hivecommons#18204 the hive refused to squash any PR whose branch carries no Signed-off-by, without asking whether the repo requires DCO at all. In repos with no DCO gate every Renovate and Dependabot PR, and every unsigned human commit, stalled at 080-awaiting-merge-admission with reason squash-unsigned and nothing a person could do. The refusal now applies only where the repo requires DCO, by the rule the stage router already uses for no-dco: the governor.labels.stage.dco.required override when set for the repo, else a dco/DCO context required on the base branch or reported on the head, read at merge time. A head whose checks cannot be read keeps the refusal. Elsewhere the squash goes out with the message as built, as before hivecommons#18204, and any branch sign-offs that exist are still carried. The sweep lanes (mergeOptions) and the merge relay (MergePR) share the one decision in SquashSignoff. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <6772335+Danathar@users.noreply.github.com>
There was a problem hiding this comment.
Looks correct to me. I read requiresDCO in src/pkg/github/merge_signoff.go against the stage router's dcoRequired (automerge_stage_labels.go:908): both consult dco.required first and fall back to a detected gate, so the "same rule" claim in the body holds. Failing closed on an unreadable head, and on more than 3 pages of statuses or check runs, keeps the refusal wherever evidence is incomplete.
One limit worth knowing, not a defect: detection is by context name (IsDCOContext), so a repo that enforces DCO through a differently named workflow, with no dco/DCO context, will now merge unsigned squashes unless it sets governor.labels.stage.dco.required: true. That matches how the router already treats no-dco, and the override covers it.
Tests cover the gate/no-gate/override/unreadable-head matrix and both callers (sweep mergeOptions, relay MergePR). A few checks (mutation, build-go, golangci-lint) were still running at review time.
β hive: agent=reviewer backend=claude model=claude-sonnet-5-5 claude=2.1.326
|
Hive status β updated in place; edits do not notify. Full history is in the Activity Monitor. Review findings from I read the changed paths in The open point is policy, not code. This turns a hard refusal into "merge unsigned" for any repo where detection finds no Suggestion (not a defect): the CI result cited is from a fork PR, so the base-repo Review findings from HUMAN DECISION NEEDED β Delta re-review (f53bf6bβ¦a37ead98). The delta is a merge of main: 6 commits touching base-red memory, GFI, claims and changelog fragments. None of it touches the PR's own files ( Previous finding "Relaxes DCO merge refusal" (
One residual gap: in a repo with no override, where the base's required checks could not be read and the head has no Review follow-ups: 1 open of 1 (below
|
|
Needs a decision, not a merge:
Head: verdict requires_human @f53bf6b, 64/64 checks green, DCO yes, soak 0s of 10m0s. How to release it:
Clear the cause and the hive re-evaluates on its next pass; the stage moves to Updated 2026-11-18 00:39 UTC. |
clubanderson
left a comment
There was a problem hiding this comment.
Approving head ac459c3c3.
Decision: yes. The hive may squash-merge a branch with no sign-off in a repository that has no DCO gate. The refusal stays wherever DCO is required.
Why:
- The blanket refusal from #18204 left approved, green pull requests in repositories without a DCO gate stalled with nothing a person could do (the Renovate and contributor pull requests cited in the description). That is a customer-facing failure, and
stage-labels.mdalready promises that an unsigned head on an ungated repository is information only. - Nothing is weakened where a gate exists. A repository with a
dco/DCOcheck required on the base or reported on the head still gets the refusal, and that includes this repository. A head whose checks cannot be read is treated as requiring DCO, so the rule fails closed. - A repository that enforces sign-off without a check named
dco/DCOsets the existinggovernor.labels.stage.dco.requiredoverride. No new setting is added. - I read the diff: the override is consulted first, then the base branch's required checks, then the head's statuses and check runs, with bounded paging. The tests cover each branch of that rule.
What happens next: the hold is released and the merge lane takes it from here under the usual gates. This remains the author's pull request.
|
LGTM label has been added. DetailsGit tree hash: 878e797f7f9b4c0458a402237e57e942bb7a88dd |
|
/approve |
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: clubanderson The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:
Approvers can indicate their approval by writing |
|
Status after approval: this pull request is approved and carries |
β¦ame the real blocker (#18772) * π Activity monitor: one rule for the human-gate pill and its verbs; name the real blocker The gate pill lit 'Needs human Β· review' from the cached reviewer verdict for the head while its verbs acted only on a PR the hive had parked on a person, so a PR whose sign-off a person had already approved kept offering Approve / Request changes / Let hive decide and every click answered 'nothing to act on' (#18667). - stage_tracker_gate.go: one predicate for the reasons a person is owed, the verbs that act, and what the item waits on when nothing is owed; carried in the payload as human_gate (rows and linked PRs). - tracker-gate-truth.js: the pill and runRowAction share one mapping; a verb is drawn only where the action has something to act on. The review reason and a bare needs_human stand only while the row is parked. - Required checks queued or in progress past HIVE_TRACKER_CHECK_STUCK_AFTER (2h) are named as stuck, not running. - A stale click says what the item is waiting on and re-reads the tracker. Fixes #18769 Signed-off-by: Andy Anderson <andy@clubanderson.com> * π Gate: judge stuck on check runs only; keep the pinned verbs line; no waiting pill on inert rows Signed-off-by: Andy Anderson <andy@clubanderson.com> --------- Signed-off-by: Andy Anderson <andy@clubanderson.com>
β¦ and fail visibly when it can't (#18796) * π PR Verifier: record PR Type / PR Issue with one completed create The pinned kubebuilder-release-tools v0.4.3 image created each required check run in_progress and then PATCHed it with the verdict. When GitHub answered the PATCH with 404 (#18667, run 44069761061) the runs stayed in_progress for ever, Tide blocked the PR, and the continue-on-error wrapper job still went green. The workflow now computes both verdicts in-repo, reusing the existing port of the upstream rules in integration-governance.cjs (same check names, output titles and summaries), and records each check with ONE create that is already completed: no create-then-update, no lookup by name. A title verdict stays advisory for the wrapper job; a result that cannot be recorded posts a completed failure run with the reason (best effort) and fails the job. No image is pulled, so the job leaves the ARM64 runner and drops packages: read. Refs #18764 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7qFmFJp26r9LPdmi3t Signed-off-by: Danathar <6772335+Danathar@users.noreply.github.com> * π PR Verifier: say which copy pull_request_target runs GitHub runs the default branch's copy of a pull_request_target workflow (verified on the fork: a PR based on a branch carrying this change ran the default branch's old workflow). Correct the checkout comment. Refs #18764 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7qFmFJp26r9LPdmi3t Signed-off-by: Danathar <6772335+Danathar@users.noreply.github.com>
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of
|
β¦eate, and fail visibly when it can't (#18795) * π PR Verifier: record PR Type / PR Issue with one completed create The pinned kubebuilder-release-tools v0.4.3 image created each required check run in_progress and then PATCHed it with the verdict. When GitHub answered the PATCH with 404 (#18667, run 44069761061) the runs stayed in_progress for ever, Tide blocked the PR, and the continue-on-error wrapper job still went green. The workflow now computes both verdicts in-repo, reusing the existing port of the upstream rules in integration-governance.cjs (same check names, output titles and summaries), and records each check with ONE create that is already completed: no create-then-update, no lookup by name. A title verdict stays advisory for the wrapper job; a result that cannot be recorded posts a completed failure run with the reason (best effort) and fails the job. No image is pulled, so the job leaves the ARM64 runner and drops packages: read. Refs #18764 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7qFmFJp26r9LPdmi3t Signed-off-by: Danathar <6772335+Danathar@users.noreply.github.com> (cherry picked from commit 5750c0f) * π PR Verifier: say which copy pull_request_target runs GitHub runs the default branch's copy of a pull_request_target workflow (verified on the fork: a PR based on a branch carrying this change ran the default branch's old workflow). Correct the checkout comment. Refs #18764 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7qFmFJp26r9LPdmi3t Signed-off-by: Danathar <6772335+Danathar@users.noreply.github.com> (cherry picked from commit d12c9ec)
No conflicts. A new head so the PR Verifier from hivecommons#18796 records PR Type and PR Issue: the runs on ac459c3 came from the earlier verifier and have sat in_progress since 11:23 AM. Signed-off-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_01BQYnPGFJx81c8JXR3o1sg1
|
New changes are detected. LGTM label has been removed. |
This branch adds three lines to client.go above RankActionableIssues, so after the v6 merge it starts at 5600; scale-envelope.md now cites 5600-5610 and the docs citation check finds AgeMinutes again. Signed-off-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_01BQYnPGFJx81c8JXR3o1sg1
There was a problem hiding this comment.
HUMAN DECISION NEEDED β @Danathar the earlier security finding (name-based DCO detection; an undetected gate now merges unsigned) is still present and needs a maintainer call.
Docs-currency review of the delta ac459c3..f53bf6b (read at f53bf6b):
- The delta is a 34-commit merge of the base branch. It does not touch
merge_signoff.go, the DCO section ofstage-labels.md, or the changelog fragment. - Docs: no new finding.
stage-labels.mdstill describes the merge-time rule that the PR body states, including thedco.requiredoverride, thedco/DCOcontext on base or head, and that an unreadable head keeps the refusal. This also discloses that detection is name-based. scale-envelope.mdcitesRankActionableIssuesatclient.go:5600-5610. That is correct at this head (client.go:5600).- Previous [security] finding (
merge_signoff.go:100): still present, unchanged by the delta. I did not re-review it in depth from this perspective.
β hive: agent=adjudicator-5 backend=claude model=claude-sonnet-5-5 claude=2.1.326
There was a problem hiding this comment.
HUMAN DECISION NEEDED β @Danathar: confirm the maintainers accept relaxing #18203's unconditional unsigned-squash refusal to "only where DCO is detected or dco.required is set" (a DCO/governance policy call).
Delta re-review, a37ead9 vs f53bf6b (correctness).
The delta is a merge of v6 into the branch (v6.44.7 release, #18839, #18834, #18842). Nothing the PR itself owns changed: merge_signoff.go, main_sweeps.go, client.go (PR hunks), the tests and changelog.d/fixed-squash-unsigned-dco-gate.md are not in the delta. I did not review the merged-in v6 code, which landed through its own PRs.
Previous finding [security] merge_signoff.go Relaxes DCO merge refusal: still present. It is the PR's stated intent, not a defect. I re-read merge_signoff.go at head:
SquashSignoff(lines 55-82) returnsErrSquashUnsignedonly whenrequiresDCOis true.requiresDCO(92-123) checks thedco.requiredoverride first, then the base's required checks, then the head's contexts.- It fails closed when
headis empty or the head's checks can't be read (108-116).
Correctness read of that code found no in-scope defect. A pre-existing gap, not caused by this PR: a DCO check that is not yet reported on the head when the merge runs would let an unsigned squash through. The PR body acknowledges this case.
Verdict stays requires_human only because of the policy change.
β hive: agent=adjudicator backend=claude model=claude-sonnet-5-5 claude=2.1.326
Summary
Since #18204, the hive refuses to squash-merge any PR whose branch has no
Signed-off-byline, and it never checks whether the repo requires DCO at all. In repos without a DCO gate, every Renovate and Dependabot PR, and every human commit made without-s, now sits at080-awaiting-merge-admission. It then goesstage/stalledwith reasonsquash-unsigned, and there is nothing a person can do about it.This PR keeps the refusal only where the repo requires DCO. It uses the same rule the stage router already applies to
no-dco:governor.labels.stage.dco.requiredsets a value for the repo, that value decides.dco/DCOcheck, either required on the base branch or reported on the head.Repos with a DCO gate keep the refusal from #18203, and that includes hivecommons/hive, whose PR heads report both
dcoandDCO. Everywhere else the squash goes out with the message as built, as it did before #18204. Any sign-offs that do exist on the branch are still carried into the merge commit.There is no new setting. The existing
dco.requiredoverride now also applies at merge time.Evidence (knuckle, build 3b50113, times Eastern)
Tests Passed. Nodco/DCOcontext appears on the head, and the hive's config sets nodco.required.automerge skip β¦ reason=squash-unsigned tier=trusted-botevery few minutes, most recently at 5:24 AM on 11-17.approve.Canary,Unit testsandRuntime contract, with no DCO check.stage-labels.mdalready says that on an ungated repo an unsigned head "keeps the lane's stage, carriesno-dcoas information only".no-dcoand sit at080.Nothing already covers this:
What changed
MergeCreditor.SquashSignoff(src/pkg/github/merge_signoff.go)requiresDCO, which checks, in order:dco.requiredoverride for the repo;RequiredStatusCheckContexts);IsDCOContext.dco.requiredoverride reaches the merge pathClient.SetDCORequiredis wired inmain_sweeps.gotocfg.Governor.Labels.Stage.DCO.RequiredFor, the same function the stage router reads.mergeOptions) and the merge relay (MergePR), so both follow the same rule.stage-labels.md(DCO routing section) describes the merge-time rule.changelog.d/fixed-squash-unsigned-dco-gate.md.DCO repos already ask for the sign-off
In a repo that requires DCO, the stage router already sends an unsigned PR to
071/072(070when self-authored) with reasondco-missing. The072nudge includes the exactgit rebase --signoffcommand (automerge_stage_dco.go,dcoAsk). That is the actionable note a person sees, so I didn't add a second one.With this change,
squash-unsignedonly fires in a DCO repo when the merge-time check finds a gate that the stage pass had not yet seen, for example adcostatus that arrived after the lane's CI walk.Tests
CI on a draft test PR in my fork (base: a copy of v6 at
3b221dde0):v2 Tests(the full Go suite) andv2 CIpassed, along with 60 other checks: Danathar#1. The only failure there,verify PR contents, is the fork being unable to pull the private verifier image (403). It is not a code result.New and updated unit tests:
TestSquashSignoffDCOGate(merge_signoff_test.go) covers:dco.required: true: refused without reading the repo;dco.required: falsewith a detected gate: the override wins and it merges;DCOrequired on the base: refused;dcostatus on the head: refused;DCOcheck run on the head: refused;TestMergePRViaMergesUnsignedSquashWithoutDCOGate: the relay squashes a Renovate-shaped unsigned PR in a repo without a gate and sends GitHub's default message.TestMergePRViaRefusesUnsignedSquashnow models a head with Prow'sdcostatus. Without it the PR would now merge.TestUnsignedSquashMergesWithoutDCOGate(automerge) runs the lane-levelmergeOptionswith no gate, with adcostatus, and with thedco.required: falseoverride.TestUnsignedSquashRefusalIsNotSelfReportednow setsdco.required: trueexplicitly.Locally I ran
gofmt -lonly, as AGENTS.md requires.Out of scope
mergelane/gate.go), which merges with an empty message and never calledSquashSignoff.π€ Generated with Claude Code
https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1