Repository navigation
fix(electron): enable SwiftShader fallback for WebGL - #43615
Conversation
Move --enable-unsafe-swiftshader into chromiumSwitches so that the Electron loader applies it too. Without it, WebGL is blocklisted on machines without a GPU. Related to microsoft#42885
Test results for "MCP"5189 passed, 614 skipped Merge workflow run. |
Test results for "tests 1"1 failed 2 flaky23553 passed, 701 skipped Merge workflow run. |
🟢 The completed CI failures are pre-existingHi, I'm the Playwright bot. I checked the PR changes, job logs, and refreshed CI test history. DetailsThe failure in the latest merged report, and all failures in the other completed jobs, have matching failures outside this PR. Moving Pre-existing flake — Chromium video/trace
Pre-existing flakes — Windows Edge All three tests time out after 30 seconds. Each has independent evidence on the same project and browser/platform:
Pre-existing flakes — Windows WebKit tracing
Pre-existing failures — Firefox Nightly BiDi All 12 failures have matching failures in their respective
These are existing Firefox BiDi failures, not evidence of a regression from the Chromium flag change. Scope: completed jobs as of 2026-10-24 00:19 UTC, for PR head Triaged by the Playwright bot - agent run |
Test results for "tests 2"1 fatal errors, not part of any test 103 flaky126667 passed, 4066 skipped, 7 did not run Merge workflow run. |
🟡 Most CI failures are pre-existing; two remain uncertainHi, I'm the Playwright bot. I checked the latest CI report, the PR changes, and refreshed CI test history. DetailsFor PR head The change preserves the normal desktop Chromium/Edge switches, apart from their order. It adds the SwiftShader flag to Electron, Android, and non-macOS Chromium BiDi. Firefox launch code is unchanged; however, Firefox recorder tests also start a Chromium UI, so their browser label alone does not exclude a connection. Uncertain — two Windows Firefox recorder tests Both fail all four attempts in the Windows Firefox job:
In the refreshed history, each test passed all 312 executions outside this PR, including 15 on this Windows bot, without a failed retry. The surrounding Firefox failures suggest an existing browser/runner problem, but that is not proof for these two tests. Their recorder UI uses the changed Chromium launcher, although its effective switches are preserved. A comparison of base Pre-existing failures / flakes
Triaged by the Playwright bot - agent run |
91a765d
into
microsoft:main
Summary
--enable-unsafe-swiftshaderintochromiumSwitchesso the Electron loader applies it too, and drop the per-launcher copies in the Chromium and Bidi launchers.SS-095 should capture webgl canvas after its drawing buffer is resizedin the Electron page tests: https://github.com/microsoft/playwright/actions/runs/39828708914/job/119998797057Related to #42885