Docs: the CI failure nobody can read the logs for, and how the library is built #1177

Merged
jaylouisw merged 1 commit from docs/known-ci-issue into main 2026-11-09 02:57:47 +00:00
Owner

Two pages, both about the parts of this project that fail silently.

docs/CI.md → "Known issue: some pushes create no workflow runs at all"

Since ~01:30 today, pull requests I open and merges I push produce no run and no task row — not queued, not failed, nothing — while another writer on the same repository, account and two-runner host dispatches normally. Each wrong hypothesis cost an iteration, so the page records them:

ruled out evidence
runner starvation 0 waiting, 2 running, jobs starting within seconds
the concurrency group other branches dispatch under the same key
a multi-clause if: guard reverted in #1172 — no change
data-only gating an empty-commit probe also produced nothing
PR head format probe opened with and without the owner prefix
only pull requests affected the merge of #1175 produced no push run either

That last row is the useful one: it is every event from this credential, not a pull-request quirk. Two suspects remain, both server-side — token-pushed commits being skipped to avoid recursion, or a repo/user setting suppressing this actor.

The page ends with the one check a person can do in a browser: no run row = the event created nothing; a run row stuck at waiting = runner labels are wrong. Those are different bugs and the API cannot tell them apart — which is exactly why the failure had to be written down somewhere findable.

docs/MAINTAINING.md

The intake pipeline the catalogue work actually follows: request → spec → importer → suite → closing comment; a source-reliability table (rendered spec page > datasheet text layer > wiki table > reseller listing); the five extraction scripts and what each pulls; the rules the importer and suite enforce; and what to do when a vendor is hard. It names this project's own scars — the 25 MB PDF cache that had to be amended out, the PoE suffix that doesn't mean every port, the family-wide range that must stay out of the catalogue rather than be averaged.

Linked from CONTRIBUTING. Suite: 114 suites, 5311 assertions, 0 failed.

Two pages, both about the parts of this project that fail silently. ### `docs/CI.md` → "Known issue: some pushes create no workflow runs at all" Since ~01:30 today, pull requests I open and merges I push produce **no run and no task row** — not queued, not failed, nothing — while another writer on the same repository, account and two-runner host dispatches normally. Each wrong hypothesis cost an iteration, so the page records them: | ruled out | evidence | |---|---| | runner starvation | 0 waiting, 2 running, jobs starting within seconds | | the concurrency group | other branches dispatch under the same key | | a multi-clause `if:` guard | reverted in #1172 — no change | | data-only gating | an empty-commit probe also produced nothing | | PR `head` format | probe opened with and without the owner prefix | | only pull requests affected | **the merge of #1175 produced no push run either** | That last row is the useful one: **it is every event from this credential**, not a pull-request quirk. Two suspects remain, both server-side — token-pushed commits being skipped to avoid recursion, or a repo/user setting suppressing this actor. The page ends with the one check a person can do in a browser: **no run row = the event created nothing; a run row stuck at `waiting` = runner labels are wrong.** Those are different bugs and the API cannot tell them apart — which is exactly why the failure had to be written down somewhere findable. ### `docs/MAINTAINING.md` The intake pipeline the catalogue work actually follows: request → spec → importer → suite → closing comment; a source-reliability table (rendered spec page > datasheet text layer > wiki table > reseller listing); the five extraction scripts and what each pulls; the rules the importer and suite enforce; and what to do when a vendor is hard. It names this project's own scars — the 25 MB PDF cache that had to be amended out, the PoE suffix that doesn't mean every port, the family-wide range that must stay out of the catalogue rather than be averaged. Linked from CONTRIBUTING. Suite: 114 suites, 5311 assertions, 0 failed.
Docs: write down the CI failure nobody can read the logs for, and how the library is built
Some checks failed
CI / Classify the change (pull_request) Successful in 17s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Failing after 20s
CI / Test (Node 18) (pull_request) Failing after 1s
CI / Test (Node 20) (pull_request) Failing after 1s
CI / Test (Node 22) (pull_request) Failing after 0s
CI / Server smoke test (pull_request) Has been skipped
CI / Docker build (pull_request) Has been skipped
f5b25d7def
Two pages, both about the parts of this project that fail silently.

**`docs/CI.md` -> "Known issue: some pushes create no workflow runs at all".** Since around
01:30 today, pull requests I open and merges I push produce **no run and no task row** - not
queued, not failed, nothing - while another writer on the same repository, account and two-runner
host dispatches normally. The document records what was ruled out and how, because each wrong
hypothesis cost an iteration:

| ruled out | evidence |
|---|---|
| runner starvation | 0 waiting, 2 running, jobs starting in seconds |
| the concurrency group | other branches dispatch under the same key |
| a multi-clause `if:` guard | reverted in #1172, no change |
| data-only gating | an empty-commit probe also produced nothing |
| PR `head` format | probe opened with and without the owner prefix |
| only pull requests affected | the merge of #1175 produced no push run either |

That last row is the useful one: **it is every event from this credential**, not a pull-request
quirk. Two suspects remain, both server-side: token-authenticated pushes being skipped to avoid
recursion, or a repository/user setting suppressing this actor. The page ends with the one check a
person can do in a browser - **no run row means the event created nothing; a run row stuck at
`waiting` means the runner labels are wrong** - because those are different bugs and the API
cannot tell them apart.

**`docs/MAINTAINING.md`** is the intake pipeline the catalogue work actually follows: request ->
spec -> importer -> suite -> closing comment, a source-reliability table (rendered spec page >
datasheet text layer > wiki table > reseller listing), the five extraction scripts and what each
pulls, the rules the importer and the suite enforce, and what to do when a vendor is hard. It
names the mistakes this project has already made - the 25 MB PDF cache that had to be amended out,
the PoE suffix that does not mean every port, the family-wide range that must stay out of the
catalogue rather than be averaged.

Linked from CONTRIBUTING. Suite: 114 suites, 5311 assertions, 0 failed.
Author
Owner

The auto-rebase job failed. This is a problem with the automation, not with this change.

the rebase job itself failed before it could report a reason - check the runner.

The run is at the Actions tab. Nothing was pushed, so the branch is exactly as you wrote it.

**The auto-rebase job failed.** This is a problem with the automation, not with this change. ``` the rebase job itself failed before it could report a reason - check the runner. ``` The run is at the Actions tab. Nothing was pushed, so the branch is exactly as you wrote it.
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
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!1177
No description provided.