Public: anyone can read this; the crew can change it.

Each look records whether its page scrolls sideways and how many screens tall it is, and the pull's looks name any phone page that scrolls sideways #9264

Parked opened this
Closed without merging

Each look now records whether its page scrolls sideways and how many screens tall it is, and a phone look that scrolls sideways is named in one line. One checklist item is not done: the comment on #9240 with the held and bets screens.

Notes

stopped parked by the crew deploy 7236e93

opened this

Each look now records whether its page scrolls sideways and how many screens tall it is, and a phone look that scrolls sideways is named in one line. One checklist item is not done: the comment on #9240 with the held and bets screens. That needs this pull's preview to deploy and take its looks, and my shell can't wait for that.

tsc and scripts/gate.ts are clean, as is bun scripts/design-lint.ts. In looks.test.ts and the other check, vitals and studio tests that touch these modules, 219 pass and 3 fail. The three failures are the real-Chromium tests, which fail the same way without my change because the sandbox won't let Chromium bind a socket. The forge's check runs them. scripts/fast.ts named 1425 of 2105 test files for this change, so I picked the covering tests myself.

What changed

  • src/looks/capture.ts: the existing read-back after load now also reads scrollWidth, scrollHeight, innerWidth and innerHeight in the same DevTools evaluation. It adds no browser start and no second navigation. parseRecorded keeps each as a whole non-negative number; a missing or malformed one is absent, never a failure.
  • src/looks/service.ts: pageReach works out sideways (scrollWidth > innerWidth) and screens (scrollHeight / innerHeight, one decimal). Both are stored beside LCP and layout shift in the commit's vitals file, not in the look table.
  • src/checks/summary.ts, src/checks/looks.ts: the stored look carries optional sideways and screens, and sidewaysNotices makes the notice line. Phone looks only.
  • src/web/studio/vitals.ts, src/checks/progress.ts: the run page and the check's progress line print /path scrolls sideways at phone width when a phone look scrolls sideways, and say nothing otherwise. There is no banner and no red.
  • Tests (test/looks.test.ts): they cover parseRecorded with the fields present, missing and malformed, and the rounding in pageReach. They also cover the stored look through the fake capture (sideways: true, screens: 2.4 for phone) and the notice for a sideways phone look and for a page that fits.

I did not touch src/http/api.ts.

To try it: open the pull's preview and its check run page. The held and bets phone screens for the #9240 comment will be in the vitals data once the preview has taken its looks.

Closes #9249

2 earlier comments
commented

Each look now records whether its page scrolls sideways and how many screens tall it is, and a phone look that scrolls sideways is named in one line. One checklist item is not done: the comment on #9240 with the held and bets screens. That needs this pull's preview to deploy and take its looks, and my shell can't wait for that.

tsc and scripts/gate.ts are clean, as is bun scripts/design-lint.ts. In looks.test.ts and the other check, vitals and studio tests that touch these modules, 219 pass and 3 fail. The three failures are the real-Chromium tests, which fail the same way without my change because the sandbox won't let Chromium bind a socket. The forge's check runs them. scripts/fast.ts named 1425 of 2105 test files for this change, so I picked the covering tests myself.

What changed

  • src/looks/capture.ts: the existing read-back after load now also reads scrollWidth, scrollHeight, innerWidth and innerHeight in the same DevTools evaluation. It adds no browser start and no second navigation. parseRecorded keeps each as a whole non-negative number; a missing or malformed one is absent, never a failure.
  • src/looks/service.ts: pageReach works out sideways (scrollWidth > innerWidth) and screens (scrollHeight / innerHeight, one decimal). Both are stored beside LCP and layout shift in the commit's vitals file, not in the look table.
  • src/checks/summary.ts, src/checks/looks.ts: the stored look carries optional sideways and screens, and sidewaysNotices makes the notice line. Phone looks only.
  • src/web/studio/vitals.ts, src/checks/progress.ts: the run page and the check's progress line print /path scrolls sideways at phone width when a phone look scrolls sideways, and say nothing otherwise. There is no banner and no red.
  • Tests (test/looks.test.ts): they cover parseRecorded with the fields present, missing and malformed, and the rounding in pageReach. They also cover the stored look through the fake capture (sideways: true, screens: 2.4 for phone) and the notice for a sideways phone look and for a page that fits.

I did not touch src/http/api.ts.

To try it: open the pull's preview and its check run page. The held and bets phone screens for the #9240 comment will be in the vitals data once the preview has taken its looks.

commented

I changed nothing for this failure, because it doesn't come from my code. The branch is already pushed, so there was nothing new to push.

  • What the log shows: the check stopped for memory (8775 MiB used against the 8192 MiB limit in forge.json) while running the fast tier's shards. Every shard it had finished passed, including the 278 tests in the last block shown, with 0 failures.
  • Why it's not my change: the gate rule from the earlier failure, the missing "Try:" line, is satisfied. It was fixed by commit 7236e938, and the check got past it. My change adds a few small pure functions and one extra field read in an existing DevTools call, which can't plausibly account for about 600 MiB. The tier is large: scripts/fast.ts named 1425 of 2105 test files for this diff, since the diff touches shared modules (checks/summary.ts, checks/progress.ts).
  • What I couldn't do: my sandbox can't measure the check's memory use, and running that many test files here would take about two hours. I didn't change limits.memory or the sharding, since the pull shouldn't alter the forge's resource settings.

A retry of the check on the same head should show whether the memory spike is a one-off. If it repeats, the cause is the shard layout or an existing memory-hungry test, and a person should decide whether to raise the limit or split the shards.

commented

Parked by the crew: the check stayed red through 2 automatic tries: check failed: stopped: used 8775 MiB of memory, over its 8192 MiB limit (forge.json limits.memory)

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs