mirror of
https://github.com/electron/electron.git
synced 2026-10-28 11:44:46 +08:00
Samuel Attard
57ab2fc74d
* test: split the Squirrel.Mac updater specs and run them serially on x64 Measured on the Intel runners, running two updater tests at once gained under a minute per suite: each run saturates the three cores, so they took turns, and every run's timing became less predictable. Run them one at a time there, and split the suite into two files, the update lifecycle and the rules for refusing or altering an update, so the shard balancer puts them on different runners. The harness they share moves to spec/lib. A slot minted for a retry is dropped when the retry ends instead of joining the pool, so a retry no longer raises the file's concurrency for good. The abort-while-running test polls for ShipIt in a loop rather than on an interval, so a slow `ps` cannot pile up behind itself. The weights for the two files are estimates from serial runs on the Intel runners; regenerate them from CI once this has run. * test: wait for focus, not show, when focusing an inactive panel On macOS the show event is emitted when the window reports itself unoccluded, and a panel shown while another app is frontmost may never do so; the test then timed out waiting for it on the arm64 MAS shard once api-app-spec ran ahead of it. The panel becoming key is what the test is about, so wait for focus. Also do not assume the app is active before handing activation to Finder, and wait for Finder to be frontmost before showing the panel. * build: regenerate spec weights with the split updater specs From Build run 34149004553 on this branch. The two updater files measured 418s and 379s serially on the Intel runners, against the estimates of 350 and 385; the rest of the weights are refreshed with them.