[waleedlatif1]

Summary

The three macOS Electron smoke shards currently split files by test count, leaving one job idle while long files finish. This change assigns complete files by observed duration across the same three runners and keeps one worker per runner. Playwright still discovers the complete suite; unknown files receive a fallback weight, and weights never determine which tests exist.

The wrapper uploads the complete discovery report and shard plan alongside the normal results. Local runs, the weekly canary, packaging smoke, Linux integration providers, and test assertions remain unchanged.

Testing

Three paired hosted samples on identical app/test source, runners, and setup:

Pair	Existing slowest job	Balanced slowest job	Slowest test step, existing → balanced
1	8m38s	7m20s	5m01s → 4m02s
2	9m12s	7m13s	5m34s → 4m08s
3	7m50s	7m21s	5m01s → 4m08s

Median slowest job: 8m38s → 7m20s, a 78-second reduction. Job times include setup/cleanup and exclude queue time. These are three pairs, not a long-run percentile estimate. All 18 jobs passed on the first attempt with zero retry events. Each arm in every pair preserved the exact 111-case/status multiset: 96 passed and 15 existing skips. Same three free macOS runners per suite; no provider or runner-count change.

The final wrapper additionally selects Playwright's outer file suites, so tests declared by imported factories cannot be omitted. A real browser-free Playwright regression failed before that guard and passes after it, including a nested filename containing regex characters and an unmeasured .test.ts file. The final file-suite selection matches all nine benchmark discovery reports.

Ten focused scheduling and CLI checks pass (3.3 seconds locally). Negative controls confirm that removing longest-first ordering, unknown-file inclusion, or the zero-duration floor causes failures. Actionlint, targeted Biome checks, and git diff --check pass. Heavy validation uses hosted runners.

Final head d718ac3e: full CI passed in 7m46s and macOS CI completed in 7m55s, both on the first attempt. Final smoke artifacts preserve all 111 cases with zero retries. Cubic and Greptile each returned 5/5 with no actionable findings; all review threads are resolved.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
docs	Ready	Preview	Oct 22, 2026 8:41pm UTC

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The PR appears safe to merge; no actionable defects were found.

Summary

The macOS Electron smoke suite now assigns whole files to three runners using measured durations.

• The macOS smoke runners balance whole test files by measured duration.

No actionable issues were found. Tests were not run during this review.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Playwright discovery] --> B[Discovered files]
  C[Measured durations] --> D[Assign longest files first]
  B --> D
  D --> E[Select this runner's files]
  E --> F[Run Playwright with exact file filters]
  F --> G[Save results, discovery, and shard plan]

Loading

_{Reviews (1) · Last reviewed commit: "test: prove desktop discovery preserves ..." · Reviewed by Greptile}

[cubic-dev-ai]

No issues found across 7 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

<sub>Heads up: you’re close to your flex budget. Increase your flex budget so reviews don’t pause.

Turn on auto-fix | Re-trigger cubic</sub>