Skip to content

test(e2e): stabilize mobile layout readiness before assertions - #232

Merged
alirezasafaei-dev merged 2 commits into
mainfrom
codex/home-e2e-readiness
Oct 11, 2026
Merged

alirezasafaei-dev merged 2 commits into
mainfrom
codex/home-e2e-readiness

Conversation

@alirezasafaeigfx

@alirezasafaeigfx alirezasafaeigfx commented Oct 10, 2026 •

Copy link
Copy Markdown
Collaborator

When the homepage DOM arrives before visible layout, the mobile E2E test measures zero-sized task/proof geometry and fails despite correct DOM order. Main CI run 38137683400 reproduced this at 360px and 390px after PR #233 merged.

This tests-only change waits for the heading, first task link and first proof article to be visible before measuring. The keyboard test also verifies the starting hero anchor is visible and focused. Existing geometry, DOM order, overflow, keyboard order and 44px target assertions remain unchanged; no retries or thresholds are relaxed.

Validation: updated targeted Chromium tests repeated four times against a local production standalone build, 8/8 passed with retries=0 at both widths. Prettier, licensing, repository secret checks, Gitleaks and production dependency audit passed. Independent review requested on updated visibility precondition. Hosted CI must pass on the exact updated head before merge.

Closes #231 for the homepage readiness race only; OCR/other baseline intermittent test failures remain separately observable. No deployment in this PR.

Signed-off-by: alirezasafaei-dev <asdevelooper@gmail.com>
Signed-off-by: alirezasafaei-dev <asdevelooper@gmail.com>
@alirezasafaei-dev
alirezasafaei-dev marked this pull request as ready for review October 11, 2026 12:17
@alirezasafaei-dev
alirezasafaei-dev merged commit db1d7ba into main Oct 11, 2026
14 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[quality] Investigate /tools CLS regression and existing flaky E2E before merging image PR

2 participants