Gemini CLI PR fixes queued tool calls still executing after scheduler disposal
andreivince · ghdev · 2026-09-21
A community PR (#29432) on google-gemini/gemini-cli fixes a bug where callers stayed pending and queued tool work could still run after scheduler disposal once a blocked batch finished.
- The fix drains the request queue via existing rejection wrappers, removes abort listeners, and checks the disposal signal before accepting new work; repeated disposal stays harmless.
- Regression tests block the active executor behind a promise, verify queued callers settle with listeners removed and tools never run, and cover scheduling after disposal.
- Validation: full macOS preflight with 15,657 tests passed; 180 scheduler-related tests passed on macOS and Linux ARM64 with coverage, plus 30 integration cases. Windows and Docker/Podman/Seatbelt sandbox modes were not tested; no user-facing changes.
More from coding & agent
- rjs: The New UI Gap Is Between the Agent's Context and the Human's Context — _AustinCalvert_ · 2026-09-21
- ModularRSI Paper Names Three Defects in How Agent Harnesses Get Improved — and the Fixes — dair_ai · 2026-09-21
- Dev Reveals His Coding Agent Hid a memcached Bug, Worked Around It and Broke the Service — yacineMTB · 2026-09-21
- The AGI token cannon will eventually discover and codify workflow structure itself — willcb · 2026-09-21
- Model once, code a thousand times: the case for model-plus-code control flow — willcb · 2026-09-21
- Turn any local LLM into a confidence-scored classifier via logprobs, full llama.cpp recipe included — DivideHorror3217 · 2026-09-21