Skip to content

[!] MERGE QUEUE [!] *JULY 1 PINCER* #733

Open

Description

@loganfinney27

"Require linear history" requirement appears to apply to all merges in a commit, not just the branch itself's squash or rebase. Testing hypothesis.

Activity

  1. added
    bugSomething isn't working
    documentationImprovements or additions to documentation
    on Jul 2, 2026
  2. added theissue type on Jul 2, 2026
  3. changed the title [-][!] MERGE QUEUE [!] *JULY 1 PINCER* RESOLVED[/-] [+][!] MERGE QUEUE [!] *JULY 1 PINCER* RESOLVED?[/+] on Jul 2, 2026
  4. loganfinney27 commented on Jul 2, 2026

    @loganfinney27
    MemberAuthor

    AUTOMATION still blocked by bot auth

  5. loganfinney27 commented on Jul 2, 2026

    @loganfinney27
    MemberAuthor

    re-enabled linear history requirement and switched merge queue to rebase

  6. changed the title [-][!] MERGE QUEUE [!] *JULY 1 PINCER* RESOLVED?[/-] [+][!] MERGE QUEUE [!] *JULY 1 PINCER*[/+] on Jul 2, 2026
  7. loganfinney27 commented on Jul 2, 2026

    @loganfinney27
    MemberAuthor

    Hypothesis tested — the second jaw of the pincer is real, with one refinement

    Refinement to the hypothesis: required_linear_history itself only constrains what lands on main (no merge commits may be pushed there); it never inspects a PR branch's internal history. The bite comes from its interaction with the queue's merge_method:

    1. Linear history forbids the MERGE method → the queue must REBASE (current setting) or SQUASH.
    2. REBASE replays the PR's non-merge commits onto main one by one and drops its merge commits — and with them every conflict resolution those merges carried.

    Live test (2026-07-02 ~21:00 UTC)

    I rebased origin/logan/obsidian (a30c58b, #563's head) onto origin/main (5f55e3b) in a throwaway worktree — the exact linearization a REBASE queue would attempt:

    Rebasing (1/74)
    CONFLICT (add/add): Merge conflict in 2026-06-18.md
    error: could not apply 6e28cebf... Create 2026-06-18.md
    

    Failure at commit 1 of 74. #563's branch carries 15 merge commits among 89 (eight Merge branch 'logan/obsidian' self-merges, plus repeated Merge branch 'main' into logan/obsidian); the resolutions live in those merges, and flattening orphans them. A REBASE-method queue can never land #563 as-is — independent of the actor-starvation jaw (#731).

    Chronology anchors

    • **logan/obsidian** #563 has never reached the queue at all: zero added_to_merge_queue events since it opened Jun 18 (auto-merge armed 2026-06-18T18:49Z, mergeable_state: blocked ever since — open review threads gate it upstream). The linear-history wall is the next jaw after threads clear; it hasn't even been reached yet.
    • Fix guard read-root bug: secret/large-file gates scanned the base tree, not the PR #727 merged 2026-07-02T18:43Z during the MERGE-method / linear-history-off window → merge commit 68ab72ae now permanently on main (re-enabling the rule doesn't retroactively object; it gates new pushes only).
    • Fix Agent Git Guardrails wrapper: install as git, fix remote-URL detection #732 merged this evening under the restored REBASE + linear-history rules as three linear commits (46d468ff, d5581be9, 5f55e3b4) — proving the queue handles linearizable branches fine under the current config.
    • API note: **logan/obsidian** #563's mergeable: True is GitHub's merge-commit test (a MERGE would apply cleanly); rebase-ability is not exposed by the API, hence the local probe.

    Options for landing #563 (in order of my recommendation)

    1. Clear the open review threads first — nothing merges before that under any method.
    2. Then open a brief SQUASH window: set the queue's merge_method to SQUASH, human-enqueue **logan/obsidian** #563 with a clear runway (no bot entries ahead, per CI Health Report — 2026-07-02 (24h review, nothing blocking main) #731 Addendum 2), revert after. Squash is linear-history-compatible and collapses the 9,006-file reorganization into one blameable commit; other entries landing during the window also just squash, which is benign.
    3. Alternative: repeat today's MERGE-method + rule-off window (proven by Fix guard read-root bug: secret/large-file gates scanned the base tree, not the PR #727) — but it reopens the window where main accretes merge commits.
    4. Not recommended: hand-linearizing the branch — my probe says conflict resolution starts at commit 1 of 74.

    Claude-Session: https://claude.ai/code/session_01SfreowpdMionR3SiGEjRBw


    Generated by Claude Code

  8. loganfinney27 commented on Jul 2, 2026

    @loganfinney27
    MemberAuthor

    queue method changed to squash

  9. loganfinney27 commented on Jul 2, 2026

    @loganfinney27
    MemberAuthor

    limiting ref "update" blocked #736 - now added to queue

  10. loganfinney27 commented on Jul 3, 2026

    @loganfinney27
    MemberAuthor

    CodeQL analysis REVERTED TO Default setup

  11. loganfinney27 commented on Jul 28, 2026

    @loganfinney27
    MemberAuthor

    Current swarm directive — #563 is a special divergence case

    As stated by Logan in Codex on 2026-07-28:

    #563 is a special case of divergence: main and logan/obsidian are not yet reconciled.

    Operational routing:

    • Do not treat **logan/obsidian** #563 as the template for every open PR.
    • Do not re-litigate the main-vs-logan/obsidian reconciliation fight across unrelated PRs.
    • Land the other open PRs first where they can be handled on their own merits.
    • Return to **logan/obsidian** #563 as its own reconciliation problem after the surrounding PR queue is less congested.

    This is a dated directive, not a permanent live-state claim.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions