diff --git a/src/docs/operator-reference.md b/src/docs/operator-reference.md index 4a031d9e1b..5c73e6be4b 100644 --- a/src/docs/operator-reference.md +++ b/src/docs/operator-reference.md @@ -768,12 +768,18 @@ into its branch, pinned to the evaluated head; it never rebases or force-pushes, and never rewrites a contributor's commits. A fork pull request that already contains the current tip with passing required checks needs no update and merges normally. Fork CI waiting for maintainer approval -(`action_required`) counts as not passed. When Hive cannot update a fork's -branch (GitHub refuses the update, for example because "Allow edits by -maintainers" is off), the pull request does not merge from the front; it -leaves the front with the recorded reason once the front timeout runs out, -the next pull request starts, and it stays eligible to merge at a later turn -once its author has brought it up to date. +(`action_required`) counts as not passed. The lane updates a behind fork +branch only when GitHub reports that Hive can update it +(`viewerCanUpdateBranch`, which needs "Allow edits by maintainers") **and** +`review.contributor_prs.base_sync` is on. Otherwise the pull request leaves +the front at once, without waiting for the front timeout, with the reason +`fork pull request is behind the target branch and Hive is not allowed to +update it: …` naming which condition failed. Hive posts one comment asking the +author to merge the target tip, and the next pull request starts. Once the +author has merged the tip and the required checks pass, the pull request +rejoins the lane and merges at a later turn. A pull request merged through a +`/hive queue` approval must be queued again after any new head, including a +merge of the tip: the approval covers exactly the approved commit. ### Branches with GitHub's merge queue diff --git a/src/docs/v6-readiness.md b/src/docs/v6-readiness.md index ecd0a1f693..1f084ed38b 100644 --- a/src/docs/v6-readiness.md +++ b/src/docs/v6-readiness.md @@ -119,7 +119,7 @@ not a unit test: | Runs via chat | One live run reaches a human gate and is approved from the chat spine; the next runs snapshot shows it advancing. | ✅ Live exercise passed with `just runs-e2e-v6` against `clubanderson/hive-runs-e2e#11`; evidence on [#8466](https://github.com/hivecommons/hive/issues/8466#issuecomment-5805851875) and tracker row checked in the [#7563 v6 readiness bar section](https://github.com/hivecommons/hive/issues/7563). | | Email | One HUMAN DECISION NEEDED escalation delivered; one allowlisted inbound reply acted on (or reply-to-act explicitly deferred here). | ⬜ | | Push / on-call | One `requires_human` verdict pages a real device via at least one provider. | ⬜ | -| Serialized merge lane (`merge_strategy: hive-serialized`, [#10884](https://github.com/hivecommons/hive/issues/10884)) | On a personal-account repo with no GitHub merge queue, following the [runbook](v6-readiness-runbook.md#evidence-template) evidence template: (1) strategy switch and ACMM Eval credit ("satisfied by Hive serialized merge lane"); (2) a single merge; (3) two PRs at once (ordering, and the behind PR is updated); (4) a hold refused at the front; (5) a failed required check at the front; (6) the front timeout; (7) a fork PR. | 🟨 Partial. A self-hosted v6 hive at `f55b5fe66` merged four PRs through the lane, which produced [#10984](https://github.com/hivecommons/hive/issues/10984) (fixed by [#10985](https://github.com/hivecommons/hive/pull/10985)) and [#11023](https://github.com/hivecommons/hive/issues/11023). **Done:** single merge; update-then-merge on the App sweep path; front timeout. **Not yet done:** hold at the front; red-at-front leave (#11023 fixed by [#11031](https://github.com/hivecommons/hive/pull/11031)); fork PR. Row stays unchecked until every step has linked evidence. | +| Serialized merge lane (`merge_strategy: hive-serialized`, [#10884](https://github.com/hivecommons/hive/issues/10884)) | On a personal-account repo with no GitHub merge queue, following the [runbook](v6-readiness-runbook.md#evidence-template) evidence template: (1) strategy switch and ACMM Eval credit ("satisfied by Hive serialized merge lane"); (2) a single merge; (3) two PRs at once (ordering, and the behind PR is updated); (4) a hold refused at the front; (5) a failed required check at the front; (6) the front timeout; (7) a fork PR. | 🟨 Partial. A self-hosted v6 hive at `dd406b3` re-ran all seven steps against one build, which includes #11031 ([evidence](https://github.com/hivecommons/hive/issues/7563#issuecomment-6470375751)). **Done:** (1) ACMM credit; (2) single merge; (3) two at once, with the behind PR updated by a merge of the tip; (5) red at the front left on the first sweep after the check failed, and the next PR was promoted in the same transaction; (6) front timeout; (7) up-to-date fork merges, and a behind fork with `base_sync` off leaves the front at once with one comment, then rejoins and merges after the author merges the tip. **Failed:** (4) a `hold` on the front PR never released the front, which waited out the 60m timeout ([#14934](https://github.com/hivecommons/hive/issues/14934)). **Also found:** contributor-lane merges bypass the lane ([#14935](https://github.com/hivecommons/hive/issues/14935)); a 4xx on the lane's merge call wedges the front until the timeout ([#14936](https://github.com/hivecommons/hive/issues/14936)). **Not yet done:** `front-extend` at a short `merge_front_timeout`; the `base_sync`-on fork leg. Row stays unchecked until step 4 passes. An earlier partial run at `f55b5fe66` produced [#10984](https://github.com/hivecommons/hive/issues/10984) (fixed by [#10985](https://github.com/hivecommons/hive/pull/10985)) and [#11023](https://github.com/hivecommons/hive/issues/11023) (fixed by [#11031](https://github.com/hivecommons/hive/pull/11031)). | ## Admin MCP readiness evidence