Release candidate for 1.3.1, with reliability fixes since 1.3.1-rc.1.
What's changed
- Stop issuing reads once a client starts closing, improving cleanup during unload and reload. (#1065)
- Report capability retry attempts and exhausted retries in logs and diagnostics, making missing entities easier to investigate. (#1066)
- Share lighting recovery storage across a recorder's channels and serialize lighting changes per device, preventing one channel from overwriting or discarding another channel's recovery state. (#1067)
- Retry unanswered alarm-output, disarming-linkage and event-notification probes so supported switches can appear when the device responds later. Gate each disarming switch on its own configuration table. (#1068)
- Make request-queue regression tests less sensitive to machine load while retaining their timeout and queue-limit checks. (#1069)
- Retry unanswered coaxial-status probes so siren, security-light and flood-light states can recover after a setup timeout. (#1070)
- Retry unanswered picture-adjustment, lighting-profile and cloud-upgrade probes, and create the corresponding number, sensor and update entities when supported capabilities are discovered later. (#1071)
Compatibility note
Devices that do not provide the disarming-linkage or event-notification configuration table no longer receive the corresponding switch that previously always read off. Automations referencing those unsupported switches may need updating.
Validation
All 4,592 tests passed against the combined changes on Python 3.14. Black passed across 316 Python files, and all merged PRs passed GitHub CI.
Changes since the previous candidate: 1.3.1-rc.1...1.3.1-rc.2
Full changelog since stable: 1.3.0...1.3.1-rc.2