mxschmitt / playwright-go

80%

main: 78%

LAST BUILD ON BRANCH fix/webkit-folder-upload-eats-package-timeout

branch: fix/webkit-folder-upload-eats-package-timeout

CHANGE BRANCH

x

Reset

• fix/webkit-folder-upload-eats-package-timeout
• chore/migrate-to-mxschmitt
• chore/modernize-docs-site
• ci/rerun-flaky-tests-gotestsum
• codex/fix-page-frames-race
• cursor/cdp-download-guid-test-8e4e
• feat/allow-cli-path-env
• feature/websocket
• fix/assertions-mixed-string-regexp-lists
• fix/ci-concurrency-serializes-main
• fix/coveralls-badge-rendering
• fix/expect-request-response-object-predicate
• fix/getby-invalid-argument-panic
• fix/lost-wakeup-in-wait-for-load-state
• fix/page-frames-data-race
• fix/readme-badges
• fix/transport-stdlib-json
• fix/waiter-predicate-panics
• main
• master
• roll/v1.62.1
• skill/release-playwright
• skill/roll-pr-footer
• upstream/cdp-event-unknown-guid-passthrough

Build # 31989952511

Build Type

Pull #632

github

Committed by mxschmitt

Commit Message

test: stop TestShouldUploadAFolderRemote from eating the package timeout

The WebKit-on-macOS folder-upload flake was mitigated in #616 with an
in-test retry loop (3 attempts) plus a 180s action timeout. Since #621
CI also re-runs failed tests via gotestsum --rerun-fails=2, so the flake
now had two retry layers -- and the in-test one is the expensive one: a
wedged attempt burns its full 180s inside the same `go test -timeout=15m`
budget as the other 723 tests.

On run 31647267347 two attempts wedged and the third passed, so the test
reported PASS after 366.99s (baseline: 1.37s). That pushed the ~490s
suite past 900s, and the resulting timeout panic made gotestsum bail out
with "rerun aborted because previous run had a suspected panic" --
disabling the retry safety net for the whole job.

Drop the in-test retry and the inflated timeout, and let gotestsum supply
the retries instead: it re-runs in a fresh process with its own -timeout
budget, and records the failed attempt so the flake stays visible on
flakiness.io rather than being swallowed by a silent in-test retry.

Observed durations are bimodal -- ~1s when it works, "never" when it
wedges -- so the shorter timeout does not reduce the odds of passing, it
only bounds what a wedge costs. Verified locally on macOS/arm64: 22 runs
under WebKit (incl. --race and heavy CPU contention) plus chromium and
firefox all complete in <=1.5s.

withRetry/attemptT had no other callers and are removed with it.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RbHRZ366YgsmofGFq21Kxy

Pull Request Pull Request #632: test: stop TestShouldUploadAFolderRemote from eating the package timeout

Coverage Stats

9121 of 11453 relevant lines covered (79.64%)

15158.72 hits per line