Skip to content

v6 top-up workflow fails on every v5 push after a squash-merged top-up PR (v5 ancestry lost) #18931

Closed

Description

@clubanderson

What happened

v6-topup.yml failed on every push to v5 from 2026-11-17 until the manual top-up (for example run 44244210058: 176 commits behind, 109 conflicted files, 17 of them add/add). The autoresolver only knows CHANGELOG, static/index.html, the TRUSTED_SCRIPTS_TREE pin and discord/, so the job exits with "needs a human" and nothing is pushed. v6 therefore missed the fixes merged to v5 on 2026-11-18 (for example src/crux/proxy/sandbox_read_relay.go, #18868).

Root cause

The previous manual top-up, #18689, was squash-merged into v6 (c7d563daf has one parent). Its content reached v6, but v5's history did not, so git merge-base origin/v5 origin/v6 stayed at b2a1c149f (the end of #18620, which was merged with a merge commit). Every later forward-merge replays the whole range since that base against v6's already-adapted copies of the same files, so everything #18689 carried conflicts again (including add/add on files both sides created in that range, such as the workflow-dispatch and meetings packages).

What is needed

  1. Guard the merge method for top-up PRs. Merge sync/v5-to-v6* and topup/* PRs with a merge commit only. The Tide/merge path used for hand-carried top-ups should refuse a squash or rebase for a PR whose head is a v5 forward-merge, or the job should fail fast when git merge-base --is-ancestor <last-carried v5 commit> origin/v6 is false, and say that the ancestry was lost.
  2. Self-heal lost ancestry. When origin/v5 is not an ancestor of v6 but the workflow's own previous top-up PR is known, record the carried v5 head in the PR body or a tag/note and, on the next run, start with git merge -s ours --no-ff --signoff <carried head> before merging origin/v5. This is what the manual top-up (🌱 chore: top up v6 from v5 (2026-11-18) #18929) did with 2a56dc578.
  3. DCO. The top-up PR carries v5 squash commits whose author and sign-off identities differ (Andy Anderson authored, hive-release[bot] / noreply sign-offs; bot-authored ones). The workflow should append a remediation commit (.github/dco.yml already allows individual remediation) or the top-up PR will sit red on DCO. On 🌱 chore: top up v6 from v5 (2026-11-18) #18929 the app listed 13 such commits.
  4. The workflow's own failure message says to run git checkout -B sync/v5-to-v6 origin/v6 && git merge --no-ff --signoff origin/v5, which hits the same 100+ conflicts when ancestry is lost; it should point at the ancestry check first.

Not fixed in the manual top-up because it is a workflow change, not a one-liner.

Activity

  1. added
    help wantedDenotes an issue that needs help from a contributor. Must meet "help wanted" guidelines.
    on Nov 18, 2026
  2. hivecommons-hive commented on Nov 18, 2026

    @hivecommons-hive
    Contributor

    🔓 Claim by @Karibhou released (lease revoked).

    Claim history
    • 2026-11-18 07:18 UTC — claimed by Karibhou (contributor on hive hosted-available-oke-11-placeholder-r05x)
    • 2026-11-18 07:22 UTC — released by Karibhou (contributor; lease revoked)
  3. Karibhou commented on Nov 18, 2026

    @Karibhou
    Contributor

    The requested automatic DCO remediation needs a maintainer policy decision before this can be implemented safely. .github/dco.yml permits individual remediation but explicitly sets thirdParty: false, and CONTRIBUTING.md says maintainers must not add sign-offs on another author’s behalf. A workflow-bot remediation cannot certify the inherited v5 commits for their different authors. The existing native-queue-policy.cjs documents a narrowly scoped prow-DCO path for automated top-ups (decision B on #17030), but that is not an individual remediation of the probot DCO check.

    Please decide how the inherited commits are to receive valid author-owned remediation or which explicitly authorized DCO handling applies to these top-ups; this should not be solved by forging author sign-offs or weakening the gate. The ancestry repair and merge-method protection do not resolve that DCO requirement.

    — hive: backend=omp model=openai-codex/gpt-6.1-terra

  4. added
    stage/045-needs-decisionIssue parked on a person: needs-human or a maintainer decision is owed before the hive picks it up
    and removed
    stage/020-claimedClaimed by the scanner or a relay contributor
    on Nov 18, 2026
  5. hivecommons-hive commented on Nov 18, 2026

    @hivecommons-hive
    Contributor

    What to reply

    • /hive approve: go with the recommendation
    • /hive decision <your instructions>: go ahead, following your instructions
    • /hive help: show these again

    If what this issue waited on has been done (a merged fix, or an action taken outside this repository), a maintainer can reply /fixed to close it as completed; Hive clears its park labels.

    Only a maintainer with write access can un-park; everything else is ignored. Full command reference: https://github.com/hivecommons/hive/blob/v5/src/docs/maintainer-commands.md.

  6. clubanderson commented on Nov 18, 2026

    @clubanderson
    MemberAuthor

    /hive decision proceed with items 1 and 2 now: (1) guard the merge method: top-up PRs (sync/v5-to-v6*, topup/*) must be merge-commit only, and the workflow fails fast with a clear message when v5 ancestry was lost; (2) self-heal lost ancestry with a signed-off merge -s ours of the previously carried v5 head recorded in the top-up PR body, as the manual top-up (#18929) did. DCO (item 3): no forged or bot-added sign-offs on other authors' commits and no weakening of the gate. Use only the narrowly scoped, already documented prow-DCO path for automated top-ups (decision B on #17030) if it covers the inherited commits, with the bot signing off only its own merge commit. If #17030 B does not cover the probot DCO check on those inherited commits, stop and report exactly which commits fail and why, and leave the human-remediation path (the owner of each commit remediates under .github/dco.yml) as the fallback. Item 4 of the issue: proceed as written.

  7. added
    stage/010-actionableTriaged; eligible for the hive to pick up
    and removed
    stage/045-needs-decisionIssue parked on a person: needs-human or a maintainer decision is owed before the hive picks it up
    on Nov 18, 2026
  8. hivecommons-hive commented on Nov 18, 2026

    @hivecommons-hive
    Contributor

    Un-parked by @clubanderson. Decision: proceed with items 1 and 2 now: (1) guard the merge method: top-up PRs (sync/v5-to-v6*, topup/*) must be merge-commit only, and the workflow fails fast with a clear message when v5 ancestry was lost; (2) self-heal lost ancestry with a signed-off merge -s ours of the previously carried v5 head recorded in the top-up PR body, as the manual top-up (#18929) did. DCO (item 3): no forged or bot-added sign-offs on other authors' commits and no weakening of the gate. Use only the narrowly scoped, already documented prow-DCO path for automated top-ups (decision B on #17030) if it covers the inherited commits, with the bot signing off only its own merge commit. If #17030 B does not cover the probot DCO check on those inherited commits, stop and report exactly which commits fail and why, and leave the human-remediation path (the owner of each commit remediates under .github/dco.yml) as the fallback. Item 4 of the issue: proceed as written.

    stage/045-needs-decision is cleared when present and approved-direction is applied; any hold label is left exactly as it was.

  9. clubanderson commented on Nov 18, 2026

    @clubanderson
    MemberAuthor

    Update after the manual top-up (#18929, merged) and the ancestry repair (#18956, merged):

    • 🌱 chore: top up v6 from v5 (2026-11-18) #18929 was squash-merged by Tide, because it did not carry the tide/merge-method-merge label. That repeated the root cause described above (v5 stopped being an ancestor of v6), so 🌱 chore: restore v5 ancestry on v6 after the squash-merged top-up #18956 restored it with an ours-merge of c57fc02, merged with the merge-method label. git merge-base origin/v5 origin/v6 is now c57fc02. Top-up PRs need tide/merge-method-merge before lgtm/approved; this is the guard item 1 above should enforce.
    • With ancestry fixed, a dry merge of the 31 v5 commits that landed since (c57fc02..v5) still conflicts in 16 files: CHANGELOG.md, static/index.html, integration-governance.cjs pin (known patterns), plus dashboard/openapi.json, docs/api-reference/{health-status-events,packs-and-acmm}.md, docs/dashboard-sections/activity-monitor.md, docs/scale-envelope.md, pkg/dashboard/{api_agent_control.go,import_count_test.go,sidebar_version_panel_test.go,static-src/css/topbar.css,static-src/js/{tracker-row-layout,tracker-row-render,version}.js}, pkg/delegatedecision/decision_glossary.go. So v6-topup.yml will keep failing on the next push until a human resolves them or the resolver learns more patterns.
    • Resolver candidates that were mechanical in the manual top-up: a three-way JSON merge for dashboard/openapi.json (parse base/ours/theirs, apply v5's path/tag additions, round-trip check), normalizing pkg/<file>:<line> citations before git merge-file on docs/api-reference/*.md and then check-api-reference-citations.sh --fix, and building CHANGELOG as v6's file plus v5's new sections (the stock merge-file --union interleaved headings and bodies across releases and produced a wrong file). Everything else needed judgment (the config.go split, duplicate implementations such as the Claude first-run notice, signature drift in tests).
    • Semantic duplicates between the lines (v6 and v5 each implemented the Claude first-run screen, the sandbox preflight/Job config split, workflow dispatch) only show up as compile or test failures in CI, so the top-up PR cannot be auto-merged blind.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    approved-directionhelp wantedDenotes an issue that needs help from a contributor. Must meet "help wanted" guidelines.

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions