Freegle / Iznik / 23654

71%

master: 72%

Pull #910

circleci

fix(test): make Playwright coverage deterministic (SpinButton timeout line)

Coveralls "Coverage decreased" failures were appearing on PRs that don't
touch the frontend at all (e.g. backend-only #908/#907, config-only #894).
Root cause, confirmed by diffing the per-file Coveralls data across three
master runs (22.861% / 22.886% / 22.91%): the ONLY file whose coverage
varies run-to-run is components/SpinButton.vue, by exactly one line.

That line is the `done.value = false` reset inside
`setTimeout(() => { ... }, props.timeout)` (5s) in finishSpinner. Playwright
e2e covers it only when a test happens to remain on the page for the full
timeout, so it flips between 10/12 and 11/12 covered depending on timing.
When a master build lands on the 11/12 sample (e.g. current HEAD at 22.91%),
every subsequent PR that lands on the normal 10/12 is reported by Coveralls
as "-0.05% decreased" and the playwright check goes red — pure measurement
noise, not a real regression.

Exclude that single timing-only line from V8/Playwright coverage with a
`/* v8 ignore next */` comment (monocart honours it; v8Ignore defaults on).
vitest uses istanbul, which ignores v8 comments, so the line is still
counted in unit coverage. This matches the file's existing approach of
excluding e2e-undeterministic paths from Playwright coverage only
(useSuppressException, useUppyRetryCoalesce, ChatMobileNavbar).

After this lands, per-commit Playwright coverage is deterministic, so a PR
that doesn't change frontend code reports no delta and the gate stops
firing spuriously.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012NjYtwJBGymZyZgZbc59oS

11303 of 15029 branches covered (75.21%)

Branch coverage included in aggregate %.

123929 of 174405 relevant lines covered (71.06%)

37.27 hits per line