mirror of
https://github.com/openai/codex.git
synced 2026-11-18 22:20:48 +00:00
## Why SSE-only mocks return 404 on WebSocket upgrades by default, causing retries before the tests reach the HTTP stream. In parallel runs, each CLI subprocess can also start a Tokio worker per host CPU. ## What changed - Share an SSE mock that returns HTTP 426 on WebSocket upgrades to trigger immediate HTTP fallback. Reuse it in the fallback test. - Set `TOKIO_WORKER_THREADS=2` for CLI subprocesses launched by the affected test helpers. Run streaming and resume cases that don't need the repository checkout in temporary working directories. - Test promptless `codex exec fork` option validation without creating a source thread. In the successful promptless fork case, keep stdin open and rely on the test runner's timeout to catch a blocked read. GitOrigin-RevId: 94c20d4b2fd7cc65bc0f2a1cd0318ac188f59f22