fix(screencast): acknowledge frames after callbacks

[Skn0tt]

This PR makes page.screencast.start({ onFrame }) use callback completion as backpressure. Fixes a bug in the OR logic (one sync ack was swallowed by pending clients), and replaced ManualPromise with LongStandingScope so we can use the non-leaking safeRace.

Re microsoft/playwright-python#3145

[github-actions]

Hi, I'm the Playwright bot and I took a look at the CI failures here.

🟢 The PR looks clear — both real failures are pre-existing flakes

This PR only touches screencast frame acknowledgement in packages/playwright-core (server/client screencast, page dispatcher, the cr/ff/wk/bidi hooks, protocol wiring, and tests/library/screencast.spec.ts). Neither failing test goes near that code, and each one already fails on unrelated PRs.

Details

Pre-existing flake / infra

• [playwright-test] › reporter-html.spec.ts:3402 › created › speedboard › clicking on label should not exit speedboard (macos-latest-node22) — pre-existing flake. Same test, same bot, same signature (toMatchAriaSnapshot timing out at 75s on the "Slowest Tests" speedboard) also failed on PR fix(test): increase Intel macOS timeout #41851 (sha fd9b0e9ab6), where a retry rescued it. Across the results DB it's 1 hard failure vs 911 passing runs — a macOS HTML-reporter UI test that has nothing to do with screencast.

• [firefox] › mcp/annotate.spec.ts:173 › user-initiated annotate downloads zip with feedback.md (ubuntu + macos) — pre-existing flake. This one flakes broadly on unrelated work: firefox 12 failures across 12 SHAs / 9 PRs (315 passes), and it also flakes on webkit (12 fails / 7 PRs), chrome, and chromium. The MCP annotate download path doesn't exercise any screencast code.

The 9 "flaky" entries in the tests-1 report (including several library/video.spec.ts screencast/video cases) were all rescued on retry, so there's nothing failing there to triage.

_{Triaged by the Playwright bot}

_{Triaged by the Playwright bot - agent run}

[github-actions]

Test results for "tests 1"

2 flaky

⚠️ [chromium-library] › library/video.spec.ts:699 › screencast › should capture full viewport on hidpi `@chromium-ubuntu-22.04-arm-node20`
⚠️ [firefox-page] › page/page-goto.spec.ts:81 › should work with Cross-Origin-Opener-Policy `@firefox-ubuntu-22.04-node20`

49998 passed, 1189 skipped

---

Merge workflow run.

[github-actions]

Test results for "MCP"

1 failed
❌ [chromium] › mcp/annotate.spec.ts:446 › should switch screencast to -s session on show --annotate @mcp-macos-latest-chromium

7759 passed, 1249 skipped

---

Merge workflow run.

[github-actions]

Hi, I'm the Playwright bot and I took a look at the CI failures here.

🟢 The PR looks clear — the one failure is a pre-existing flake

The latest merged report has a single red: [chromium] › mcp/annotate.spec.ts:446 › should switch screencast to -s session on show --annotate on mcp-macos-latest-chromium. Despite the "screencast" in its name, this is an MCP annotate test that flakes broadly on unrelated work, so it isn't this PR.

Details

Pre-existing flake / infra

• [chromium] › mcp/annotate.spec.ts:446 › should switch screencast to -s session on show --annotate (mcp-macos-latest-chromium) — pre-existing flake. Across the test-results DB it fails on macOS chromium on a string of unrelated SHAs and PRs — chore: fix stack traces in development builds #41672, fix(electron): close pages from the main process to avoid Target.closeTarget hang #41695, chore(recorder): mint page aliases in the language generators #41728, chore(recorder): unify double click handling, drop action timestamps #41732, fix(reporter): expose setup/teardown projects to preprocessSuite #41731, feat(fetch): implement APIResponse.timing() #41800, chore(mcp): remove support for extension protocol v1 #41857, feat(test-runner): support named test locks #41862, feat(expect): add expect.fn() mock functions with awaited assertions #41884, chore(skills): move agent skills under playwright-core, add init-skills #41887, plus push runs — long before this change (chromium: 7 failures / 327 passes). It also flakes on firefox (14 fails), webkit (6), and chrome (4). The signature is a macOS-only timing flake on the MCP annotate download/session-switch flow.

This PR's diff is confined to screencast frame acknowledgement in packages/playwright-core — server/screencast.ts, dispatchers/pageDispatcher.ts, the cr/ff/wk/bidi page hooks, client/screencast.ts, the protocol/validator wiring — plus tests/library/screencast.spec.ts. The MCP annotate CLI path doesn't exercise that code, and the earlier merged reports on this PR (the library screencast/video cases, reporter-html, and annotate.spec.ts:173) were all rescued on retry or are the same broadly-flaking annotate test.

_{Triaged by the Playwright bot - agent run}