ci: a request closed by its own disclaimer - check the sentence before pushing #2020

Merged
jaylouisw merged 1 commit from catalogue/1118-1906 into main 2026-11-18 21:19:08 +00:00
Owner

The forge closed two requests on their own disclaimers

Two merges today closed a [Device]: request using a sentence that said the opposite, two
seconds after the merge, with no comment on the issue:

pull request the words in its body the request what happened
#2000 "Refs #803 - does not close it", and further down "Why it cannot close it:" issue 803 closed 16:04:48Z, reopened 18:27:58Z
#2005 "This PR does not close it - UDW-PRO still has no Ubiquiti source" issue 130 closed 18:00:05Z, reopened 18:27:58Z

Neither of those was scripts/close-merged-issues.js: that sweep posts a comment carrying its
marker before it changes state, and neither issue has one. Both were the forge's own merge-time
parser reading a keyword next to a number - the mechanism that legitimately closed one issue from
#1635, doing the same thing to a sentence that disclaimed it.

The identical shape in #1867 ("Does not close something") and #1873 ("it does not close
the issue") closed nothing at all. So the forge's rule is not learnable from out here, and the
conclusion is not "write a better negation" - it is never put a closing keyword in a sentence
about a request you are not answering
:

  • Refs #N - a reference with no closing keyword anywhere in the sentence;
  • issue N - no #, so there is nothing for either parser to read.

This pull request is written that way, which is why it names issue 803 and issue 130 rather than
quoting the two bodies faithfully.

What this adds

  • scripts/check-closing-keywords.js - reports every keyword-plus-number the forge could read as
    a close while this repository's rule would not, with the reason (negated, mid-sentence,
    hypothetical), the line and the snippet. --pr N reads a pull request body and its commits
    from the forge; --strict exits 1. Run against #2005 and #2000 it reproduces both closes, and
    against #1635 it reports one claim and no risk.
  • test/closing-keywords.test.js - 25 checks, including that the new rule and
    parseClosingKeywords() agree on every real body quoted in docs/CI.md, so the two cannot
    drift apart.
  • docs/CI.md - a section recording both measurements and the two safe shapes; the paragraph that
    used to say Forgejo had agreed about #742/#743 now says what it can: those two merges closed
    nothing, which is not a rule.
  • scripts/close-merged-issues.js - the comment block that made the same claim, corrected. No
    behaviour change; test/close-merged-issues.test.js still passes 81/81.

Gates

node scripts/syntax-check.js      -> 423 files, all OK
node scripts/audit-port-types.js  -> No port contradicts its declared type
node test/closing-keywords.test.js      -> 25 passed, 0 failed
node test/close-merged-issues.test.js   -> 81 passed, 0 failed
node test/hardware-invariants.test.js   -> 26 passed, 0 failed
node test/request-queue.test.js         -> 32 passed, 0 failed

No spec, no catalogue entry and no engine behaviour is touched: this is the queue machinery that
keeps the two open requests where a reader can find them.

The two requests this protects, for the record: #803 and #130.

## The forge closed two requests on their own disclaimers Two merges today closed a `[Device]:` request using a sentence that said the opposite, two seconds after the merge, with no comment on the issue: | pull request | the words in its body | the request | what happened | |---|---|---|---| | #2000 | "Refs #803 - does not close it", and further down "Why it cannot close it:" | issue 803 | closed 16:04:48Z, reopened 18:27:58Z | | #2005 | "This PR does not close it - UDW-PRO still has no Ubiquiti source" | issue 130 | closed 18:00:05Z, reopened 18:27:58Z | Neither of those was `scripts/close-merged-issues.js`: that sweep posts a comment carrying its marker *before* it changes state, and neither issue has one. Both were the forge's own merge-time parser reading a keyword next to a number - the mechanism that legitimately closed one issue from #1635, doing the same thing to a sentence that disclaimed it. The identical shape in #1867 ("Does **not** close something") and #1873 ("it does **not** close the issue") closed nothing at all. So the forge's rule is not learnable from out here, and the conclusion is not "write a better negation" - it is **never put a closing keyword in a sentence about a request you are not answering**: - **`Refs #N`** - a reference with no closing keyword anywhere in the sentence; - **`issue N`** - no `#`, so there is nothing for either parser to read. This pull request is written that way, which is why it names issue 803 and issue 130 rather than quoting the two bodies faithfully. ## What this adds - `scripts/check-closing-keywords.js` - reports every keyword-plus-number the forge could read as a close while this repository's rule would not, with the reason (`negated`, `mid-sentence`, `hypothetical`), the line and the snippet. `--pr N` reads a pull request body and its commits from the forge; `--strict` exits 1. Run against #2005 and #2000 it reproduces both closes, and against #1635 it reports one claim and no risk. - `test/closing-keywords.test.js` - 25 checks, including that the new rule and `parseClosingKeywords()` agree on every real body quoted in `docs/CI.md`, so the two cannot drift apart. - `docs/CI.md` - a section recording both measurements and the two safe shapes; the paragraph that used to say Forgejo had agreed about #742/#743 now says what it can: those two merges closed nothing, which is not a rule. - `scripts/close-merged-issues.js` - the comment block that made the same claim, corrected. No behaviour change; `test/close-merged-issues.test.js` still passes 81/81. ## Gates ``` node scripts/syntax-check.js -> 423 files, all OK node scripts/audit-port-types.js -> No port contradicts its declared type node test/closing-keywords.test.js -> 25 passed, 0 failed node test/close-merged-issues.test.js -> 81 passed, 0 failed node test/hardware-invariants.test.js -> 26 passed, 0 failed node test/request-queue.test.js -> 32 passed, 0 failed ``` No spec, no catalogue entry and no engine behaviour is touched: this is the queue machinery that keeps the two open requests where a reader can find them. The two requests this protects, for the record: #803 and #130.
ci: a body that disclaims a request still closes it - check the sentence before pushing
All checks were successful
CI / Classify the change (pull_request) Successful in 6s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 8s
CI / Server smoke test (pull_request) Successful in 9s
CI / Docker build (pull_request) Successful in 22s
CI / Test (Node 20) (pull_request) Successful in 51s
CI / Test (Node 22) (pull_request) Successful in 54s
CI / Test (Node 18) (pull_request) Successful in 58s
CI / Arm auto-merge (pull_request) Successful in 18s
baa7d488ab
The forge closed two device requests on their own disclaimers on 2026-11-18: issue 803 from
"Why it cannot close it" at 16:04:48Z, and issue 130 from "This PR does not close it" at
18:00:05Z - both two seconds after a merge, neither carrying a comment, so neither was
close-merged-issues.js, which comments before it changes state.

The identical shape closed nothing in two earlier pulls, so the forge rule is not knowable from
here. scripts/check-closing-keywords.js reports every keyword-plus-number that the forge could
read as a close and this repository's rule would not, with the reason, and
test/closing-keywords.test.js holds it in step with parseClosingKeywords. docs/CI.md records both
measurements, the reopenings, and the two shapes that cannot be misread.
forgejo-actions force-pushed catalogue/1118-1906 from baa7d488ab
All checks were successful
CI / Classify the change (pull_request) Successful in 6s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 8s
CI / Server smoke test (pull_request) Successful in 9s
CI / Docker build (pull_request) Successful in 22s
CI / Test (Node 20) (pull_request) Successful in 51s
CI / Test (Node 22) (pull_request) Successful in 54s
CI / Test (Node 18) (pull_request) Successful in 58s
CI / Arm auto-merge (pull_request) Successful in 18s
to 90260db88e 2026-11-18 19:03:36 +00:00
Compare
forgejo-actions force-pushed catalogue/1118-1906 from 90260db88e to b252bb816c 2026-11-18 19:38:51 +00:00 Compare
forgejo-actions force-pushed catalogue/1118-1906 from b252bb816c to 6ac7e85f78 2026-11-18 20:08:57 +00:00 Compare
Author
Owner

Still no CI on this PR — and the cause is now visible.

  • Head was b252bb81 when I surveyed at 20:07Z; at 20:08:57Z the auto-rebase workflow pushed 6ac7e85f (behind=0, ahead=1).
  • Neither SHA carries any commit status: /commits/<sha>/status returns total_count: 0, empty state — not a failure, no run at all.
  • Two pull_request runs exist under this title (6987, 6988, both success), so the workflow can fire for this branch; they predate the rebase push.
  • Runner queue is congested at the same time: runs 7019 running and 7015/7023 waiting on sibling branches.

Net effect: auto-rebase keeps minting a fresh head SHA which never receives a pull_request run, so the PR can never go green and can never auto-merge — a livelock. Not touching the branch myself (nothing to rebase; it is already 0 behind). Needs -ci/Jay: why is no run queued for the post-rebase SHA.

Still no CI on this PR — and the cause is now visible. - Head was `b252bb81` when I surveyed at 20:07Z; at 20:08:57Z the auto-rebase workflow pushed **`6ac7e85f`** (behind=0, ahead=1). - Neither SHA carries any commit status: `/commits/<sha>/status` returns `total_count: 0`, empty `state` — not a failure, no run at all. - Two `pull_request` runs exist under this title (6987, 6988, both `success`), so the workflow *can* fire for this branch; they predate the rebase push. - Runner queue is congested at the same time: runs 7019 `running` and 7015/7023 `waiting` on sibling branches. Net effect: auto-rebase keeps minting a fresh head SHA which never receives a `pull_request` run, so the PR can never go green and can never auto-merge — a livelock. Not touching the branch myself (nothing to rebase; it is already 0 behind). **Needs `-ci`/Jay: why is no run queued for the post-rebase SHA.**
forgejo-actions force-pushed catalogue/1118-1906 from 6ac7e85f78 to 2a0f890d3b
All checks were successful
CI / Classify the change (pull_request) Successful in 8s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 4s
CI / Server smoke test (pull_request) Successful in 10s
CI / Docker build (pull_request) Successful in 21s
CI / Test (Node 18) (pull_request) Successful in 1m0s
CI / Test (Node 20) (pull_request) Successful in 57s
CI / Test (Node 22) (pull_request) Successful in 58s
CI / Arm auto-merge (pull_request) Successful in 1m1s
2026-11-18 20:38:59 +00:00
Compare
Author
Owner

Same shape as #2019, different ending: the rebase minted 6ac7e85f with no run, but a close+reopen produced pull_request runs 7055/7056 against head 2a0f890d (main c6053d2e, behind=0). It is queued behind a single runner, so it should go green on its own; no dispatch needed here.

Root cause is in the rebase workflow, not this PR: pushes made with the auto-rebase action token do not re-trigger ci.yml, so any SHA the bot mints is unbuildable. Noted on #2019 with the suggested fixes — needs Jay's decision on .forgejo/workflows/auto-rebase.yml.

Same shape as #2019, different ending: the rebase minted 6ac7e85f with no run, but a close+reopen produced `pull_request` runs 7055/7056 against head **2a0f890d** (main c6053d2e, behind=0). It is queued behind a single runner, so it should go green on its own; no dispatch needed here. Root cause is in the rebase workflow, not this PR: pushes made with the auto-rebase action token do not re-trigger `ci.yml`, so any SHA the bot mints is unbuildable. Noted on #2019 with the suggested fixes — **needs Jay's decision** on `.forgejo/workflows/auto-rebase.yml`.
forgejo-actions scheduled this pull request to auto merge when all checks succeed 2026-11-18 20:50:18 +00:00
forgejo-actions force-pushed catalogue/1118-1906 from 2a0f890d3b
All checks were successful
CI / Classify the change (pull_request) Successful in 8s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 4s
CI / Server smoke test (pull_request) Successful in 10s
CI / Docker build (pull_request) Successful in 21s
CI / Test (Node 18) (pull_request) Successful in 1m0s
CI / Test (Node 20) (pull_request) Successful in 57s
CI / Test (Node 22) (pull_request) Successful in 58s
CI / Arm auto-merge (pull_request) Successful in 1m1s
to cc311e001a 2026-11-18 20:54:27 +00:00
Compare
forgejo-actions force-pushed catalogue/1118-1906 from cc311e001a to 741ae00f9c
All checks were successful
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 4s
CI / Classify the change (pull_request) Successful in 6s
CI / Server smoke test (pull_request) Successful in 8s
CI / Test (Node 18) (pull_request) Successful in 54s
CI / Docker build (pull_request) Successful in 22s
CI / Test (Node 20) (pull_request) Successful in 42s
CI / Test (Node 22) (pull_request) Successful in 47s
CI / Arm auto-merge (pull_request) Successful in 3s
2026-11-18 21:12:26 +00:00
Compare
jaylouisw deleted branch catalogue/1118-1906 2026-11-18 21:19:09 +00:00
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
2 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
jaylouisw/packetforge!2020
No description provided.