mirror of
https://github.com/openai/codex.git
synced 2026-05-28 06:55:01 +00:00
cf9a4f4c40
The review integration tests spawn a wiremock server and the Codex conversation manager inside the same Tokio runtime; each of those launch long-lived async tasks to drive the SSE stream and event pump. On Windows the slower I/O frequently starves those tasks when we only offer two worker threads, which trips the `timeout waiting for event` guard in `wait_for_event`. The other review cases (`review_op_with_plain_text_emits_review_fallback`, `review_does_not_emit_agent_message_on_structured_output`, and `review_input_isolated_from_parent_history`) already bumped their Windows flavor to four worker threads after similar CI flakes. This change makes `review_op_emits_lifecycle_and_review_output` consistent so Windows runs stop intermittently hanging while Linux/macOS stay on two threads.