mirror of
https://github.com/openai/codex.git
synced 2026-11-15 21:50:46 +00:00
## Why Switching backends while an overview refresh is pending can lose queued pagination or let a late page consume another backend's refill budget. ## What changed - Track queued refreshes, “Show more” requests, refill counts, and refresh failures per backend. - Preserve pending work across backend switches and use registered live request handles to refresh the owning backend. - Cancel stale reads for the affected backend after mutations, event lag, or disconnects, and hide “Show more” when its backend is disconnected. ## Testing Add regression coverage for pagination across focus changes, backend-specific refills, pin completion, stale reads after event lag, and connection registration during reconnect. GitOrigin-RevId: 3429b2b3ac93156b4dc4234ad042431d26c84525