Skip to content

Report a workflow it could not read as unknown, not as pinned - #12

Merged
DenisDrobyshev merged 1 commit into
masterfrom
fix/unpinned-check-cannot-read
Sep 18, 2026
Merged

DenisDrobyshev merged 1 commit into
masterfrom
fix/unpinned-check-cannot-read

Conversation

@DenisDrobyshev

Copy link
Copy Markdown
Member

Follow-up to #11, which shipped the unpinned-actions check an hour ago.

It returned [] when a request failed, so a dropped call reported a repository as fully pinned. That happened on its first real use — a sweep across all nine repositories silently skipped decisionrl, which has 21 loose references; three immediate retries all returned 21, so the failure was one blip.

This file's own docstring states the rule: a repository that could not be read is listed as unread, never silently counted as quiet. I broke it in the check added to enforce another one.

Now: unpinned_actions returns None for unreadable and never [], the report says the pinning is unknown rather than fine, and there are three attempts before giving up — a truthful check that cries wolf on every dropped packet becomes noise, and noise is how a report stops being read.

Verified both paths: normal run reports 21 for decisionrl and 0 elsewhere; with every call stubbed to fail it returns None rather than an empty list.

The check merged an hour ago returned an empty list when a request failed, so a
single dropped call reported a repository with twenty-one loose references as
fully pinned. That happened on its first real use: a sweep across the nine
repositories skipped decisionrl silently, and three immediate retries all
returned twenty-one, so the failure was one blip rather than anything
persistent.

This file's docstring already states the rule I broke -- a repository that could
not be read is listed as unread, never silently counted as quiet -- and I broke
it in the check added to enforce a different one.

unpinned_actions now returns None for unreadable and never an empty list in that
case, and the report says the pinning is unknown rather than fine. Three
attempts before giving up, because a truthful check that cries wolf on every
dropped packet becomes weekly noise, and noise is how a report stops being read.
@DenisDrobyshev
DenisDrobyshev merged commit 4d50911 into master Sep 18, 2026
1 check passed
@DenisDrobyshev
DenisDrobyshev deleted the fix/unpinned-check-cannot-read branch September 18, 2026 20:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant