[JamelHammoud]

Failure

Check on main, b0fc0e7

Cause

Class: broke at a commit. The LiDAR test came in with 1fe08d9 (batch merge) and failed both full runs since: f5a6da2 (360 s, timed out twice) and b0fc0e7 (180 s timeout, then the retry's Play click never became stable on a stalled page). The last green full run, 03d9170, didn't include it. #530 (864eaa6) tried to fix it forward: it switched to the RPLIDAR S2 and ended the try with endTryAt. It merged untested because tools/e2e-changed.mjs matched a test only when its name sat on the call's line. Prettier wraps long names (itUnlessWindows(\n 'scans a corridor…'), so changed lines were credited to the previous single-line test. This change fixes that gap. The LiDAR failure itself is not fixed: current main fails it in the Linux container.

Fix

A pull request's e2e-changed job now runs every e2e test it edits, including tests whose name Prettier wraps onto the line after it( or itUnlessWindows(. Before, the picker credited those lines to the test above, so #530's LiDAR repair merged without its own Check ever running the LiDAR test. This does not make the nightly pass: in CI-like Linux, current main still fails drive.e2e's LiDAR test and two others in 3 of 3 runs.

Evidence

• npm run check:changed -- tools/e2e-changed.mjs tools/e2e-changed.test.mjs passes: Prettier, ESLint, and 4 of 4 tests.
• The new unit test covers a wrapped itUnlessWindows( name; the old regex returns [] for it.
• Old vs new picker on the LiDAR test's lines: old → "makes Pick Up a Block…", new → "scans a corridor…".
• The new picker finds 9 more tests across 5 e2e files.
• CI logs: Repair: End the LiDAR e2e try at a fixed clock time and carry a light-mesh RPLIDAR S2 #530's run 43506851871 and Feature: Multiply a passed try into up to 50 varied passing tries for the dataset #533's run 43513984093 never ran the LiDAR test.
• node tools/loop/linux-e2e.mjs e2e/drive.e2e.test.ts --runs 3 --cpus 2 --ref HEAD on main 0f8bc7c passed 0 of 3. Every run failed LiDAR, Cases > Touch the Dots, and Multiply. Touch the Dots passes on CI in 20–100 s, so the container is harsher than CI and 0 of 3 doesn't prove a CI failure.

[JamelHammoud]

Review: pass (1470b45)

The picker now matches the test regex across the whole source with the m flag, and \s* after ( lets a name sit on the next line. So a test whose name Prettier wraps is credited with its own changed lines instead of the test above it. The unit test covers the wrapped itUnlessWindows( case with the right line numbers, all 4 tests pass, the loop guide's description of e2e-changed stays true, and no protected path, copy or test weakening is involved.

Copy: 0 added, 0 changed, 0 removed, net 0