Compare commits
1
Commits
| Author | SHA1 | Date | |
|---|---|---|---|
|
|
7b9f733c97 |
@@ -144,18 +144,15 @@ missing from its test. `make -C ci test` goes from **FAILED to all suites passed
|
||||
|
||||
- **The live Gitea rule still says `required_approvals: 1`. I did not change it.** Mutating live
|
||||
platform settings is an operator action; `infra/gitea/allow-admin-self-merge.sh` is written,
|
||||
syntax-checked, and **unrun**. The declaration and the live rule must move together.
|
||||
*Now measured, not predicted:* PR #38's CI run (run 202, job 0) fails at **1m44s** with
|
||||
`required_approvals: DECLARED 0 but LIVE is 1` → `gitea-settings: FAIL`. Correct behaviour, and
|
||||
the PR stays red until the script is run.
|
||||
- **The override has never been ACCEPTED by a real run** — only offered. Measured in the same CI
|
||||
log: the stage reached `approval-policy: FAIL` on the sensitive paths and printed the *new*
|
||||
code's guidance, `Or, if you hold admin/owner on this repository, comment on the PR:
|
||||
/approve-override <reason> ... NEWER than head commit 7b9f733c`. So the new stage **is** what
|
||||
runs on this PR (self-bootstrapping confirmed by measurement, not by reasoning), and the
|
||||
authorising branch — comment parse → live permission lookup → exit 0 — is still **unexercised**.
|
||||
Posting the comment on PR #38 is its first real test. If it does not work, the recovery is to
|
||||
PATCH `required_approvals` and read run 202's successor log.
|
||||
syntax-checked, and **unrun**. Until it runs, `ci/gitea-settings.sh` will report **DRIFT** on
|
||||
this branch (declared 0, live 1) and the PR will be **red** — correctly. The declaration and
|
||||
the live rule must move together.
|
||||
- **I have not seen the override accepted by a real CI run.** The code path has never executed
|
||||
against live Gitea. The first proof will be this PR itself: it touches `ci/**`,
|
||||
`infra/gitea/**` and `CLAUDE.md`, so it needs an override to pass its own new stage.
|
||||
Because CI checks out the PR's merged tree, the *new* `approval-policy.sh` is what runs — it is
|
||||
self-bootstrapping — but that is reasoning, not a measurement. **If the override does not work,
|
||||
this PR cannot merge**, and the recovery is to PATCH `required_approvals` and read the job log.
|
||||
- **I have not seen the gate still REFUSE a merge** now that approvals are 0. The claim that
|
||||
`block_admin_merge_override: true` keeps a red gate unmergeable rests on Gitea's documented
|
||||
behaviour and on the 2026-09-29 session, **not** on a provoked refusal in this one. That probe
|
||||
|
||||
Reference in new issue
Block a user