fix(chromium): only re-fetch replay-safe requests when reading response body

[yury-s]

Summary

• Chromium sometimes evicts response bodies; response.body() fell back to Network.loadNetworkResource, silently issuing a real second network request for any resource, including requests that may have side effects on the server.
• Only re-fetch requests that are safe to replay: GETs of static subresources (side-effect-free Sec-Fetch-Dest values) and prefetch requests (Sec-Purpose: prefetch, served from the http cache). Return an empty body otherwise.

Fixes #42002

[github-actions]

Test results for "tests 1"

8 flaky

⚠️ [chromium-library] › library/video.spec.ts:664 › screencast › should capture full viewport `@frozen-time-library-chromium-linux`
⚠️ [chromium-library] › library/chromium/chromium.spec.ts:299 › should report intercepted service worker requests in HAR `@chromium-ubuntu-22.04-arm-node20`
⚠️ [chromium-library] › library/video.spec.ts:736 › screencast › should work with video+trace `@chromium-ubuntu-22.04-arm-node20`
⚠️ [chromium-library] › library/har-websocket.spec.ts:235 › should attach websocket messages `@chromium-ubuntu-22.04-node24`
⚠️ [firefox-page] › page/page-event-request.spec.ts:181 › should return response body when Cross-Origin-Opener-Policy is set `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-page] › page/page-goto.spec.ts:90 › should work with Cross-Origin-Opener-Policy `@firefox-ubuntu-22.04-node20`
⚠️ [playwright-test] › ui-mode-trace.spec.ts:715 › should indicate current test status `@windows-latest-node22`
⚠️ [playwright-test] › ui-mode-trace.spec.ts:827 › should update state on subsequent run `@windows-latest-node22`

51199 passed, 1188 skipped

---

Merge workflow run.

[github-actions]

Test results for "MCP"

1 failed
❌ [chrome] › mcp/cli-storage.spec.ts:21 › state-save saves storage state to file @mcp-macos-latest-chrome

8049 passed, 1284 skipped

---

Merge workflow run.

[github-actions]

🟡 The one failure is very likely a flaky timeout, but I can't fully clear it

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

The only real failure in the latest reports is [chrome] › mcp/cli-storage.spec.ts:21 › state-save saves storage state to file (mcp-macos-latest-chrome) — a bare Test timeout of 30000ms exceeded. with no stack. This PR only touches Chromium's response.body() re-fetch path (crNetworkManager.ts), which the CLI storage-state test never exercises, so I don't think the diff caused it. But the only time this test has ever gone red is on this PR's own run, so I can't point to an off-PR failure to call it a confirmed pre-existing flake — hence yellow, not green.

Details

Overall: the earlier "cancelled" reports are incomplete (missing shards), so I looked at the latest reruns: tests 1 came back with 8 flaky, 0 failed, and MCP with 1 failed. Only that one failure needs a verdict.

Uncertain (leaning flake / infra)

• [chrome] › mcp/cli-storage.spec.ts:21 › state-save saves storage state to file — a plain 30s test timeout, no error trace. Two reasons it's unlikely to be this PR:
  • The diff doesn't reach it. The change is scoped to Chromium response.body() re-fetch of evicted bodies in crNetworkManager.ts; this test writes storage state (cookies/localStorage) to a file and never reads a response body.
  • The test is rock-solid. In the aggregated results DB it's passed 623 of 624 runs on mcp-macos-latest-chrome, and 0 failures across every other bot. The single red is run 30944790620 — this PR's SHA 410f707a — and it went green again on the very next SHA (30945227797).
  • What keeps it from a clean green: the DB has no instance of this test failing on a SHA the PR can't be responsible for, so per our own rule I can't stamp it a proven pre-existing flake. A re-run of the MCP shard should settle it — I'd expect it to pass.

Flaky (within-run, already retried green)

• The tests 1 report lists 8 flaky (video screencast, HAR websocket, COOP page tests, ui-mode-trace) — all rescued on retry and all in areas untouched by this diff. Nothing to triage there.

_{Triaged by the Playwright bot - agent run}