Repository navigation
0.4.3c
Pre-release
Pre-release
0.4.3c (pre-release)
Third pre-release of the 0.4.3 cycle. Everything from 0.4.3b, plus a fix so pre-release baselines (e.g. 0.4.3b) pass the version gate, and a corrected note about what a tgz install does.
New features
- Firmware install from the web UI (#329) - a Firmware Package section on the Service page accepts a
yi_dome_x.x.x.tgzupload, validates it, and runs the standard install (config backup, two-stage extract, reboot). The Service page is now the single place for all three update methods (latest, pre-release, from file); the Update page keeps the legacy home+rootfs flash and is linked from Service. The SD card is still required - the hack and the upgrade both run from it - but it no longer needs to be touched from a PC. Verified end-to-end on a yi_dome: a 12.8 MB package uploaded via the web UI installed cleanly with configuration preserved. - Stable ONVIF / WS-Discovery UID (#107) - the camera previously advertised a random UUID on every boot, so ONVIF clients could not re-identify it. The UID is now configurable (
ONVIF_UIDon the Configurations page), falling back to the camera serial number when unset. - "Delete all" on the Events page (#302) - wipes every recorded event directory behind a confirm dialog; per-directory delete is unchanged.
- Status page links follow the host you use (#255) - snapshot and RTSP links now resolve against the hostname/port the browser used to reach the camera, so they work through reverse proxies and tunnels instead of pointing at a LAN IP.
Bug fixes
- Pre-release field showed the branch name, not the tag - GitHub changed the layout of its releases API response, which misaligned the line-offset parsing, so the web UI displayed the default branch (
test-0.4.3) instead of the pre-release tag. The release list is now parsed properly withjq(bundled in the package, built viasrc/jq/compile.jq) after downloading to a temp file, with up to three retries - the old stdout pipe truncated the large API response intermittently. The same fix covers the matching JS field typo that kept the pre-release status line empty. - Firmware status went blank after normal use - the status check hit the unauthenticated GitHub API twice per page load, and that quota (60 requests/hour) is per IP, shared with the rest of the LAN. After a few hours of use, both the Latest and Pre-Release fields rendered empty until the hourly window reset. The status check now serves from a 10-minute local cache; the cache is only written when the API actually answered, so a rate-limited response is never pinned.
- Empty MQTT motion files list (#388) -
yicam/motion_filescarried correct timestamps but an emptyfilesarray. Two root causes in the mqttv4 component: a GMT/local time mismatch against the recorder's directory naming, and a directory-walk that aborted on the first missing minute. Verified with a 1:1 port of the lookup logic against a realistic recording tree. - PTZ "Go To Preset" never worked (#472) - the select's display text was sent instead of its value, so every request was rejected as invalid.
- Spurious "baseline is incorrect" boot warning - the baseline gates were frozen as exact
0.4.1equality checks, so every correctly flashed 0.4.2+ camera logged a boot error and the GitHub-download upgrade path refused to run. The checks now correctly mean "baseline 0.4.1 or newer". - Upgrade refused on pre-release baselines - once the check above was relaxed, a pre-release baseline such as
0.4.3bstill failed: the version gate parses each dotted field withprintf %03d, and busybox reads the non-numeric suffix (3b) as 0, so0.4.3bcomputed as0.4.0and was rejected with "Wrong baseline version" on every upgrade attempt (plus the spurious boot warning). The letter suffix is now stripped before the numeric comparison in both the upgrade gate and the boot check, so0.4.3bis treated as0.4.3. - RTSP watchdog efficiency (#496) - the MQTT guard spawned four subprocesses every 10 s (even with MQTT off) and misfired when the setting was empty; it now uses the shared
read_confhelper. - RTSP key-frame detection (#495) - restored the STAP-A and H.265 fragment offsets in
rRTSPServer's key-frame check. No effect on current H.264 v5 streams; a correctness fix for future stream types.
Documentation & release plumbing
- README: documents the web-UI package install and clarifies that the SD card remains mandatory.
- Service-page upload note corrected: on a FAT32 card, a tgz install also updates the baseline (rootfs and home), not just the yi-hack tree.
- Version bumped to 0.4.3 (package name, in-tree version file, README links).
For testers
- Upgrading from 0.4.3c through the web UI is itself a good end-to-end test, and exercises the baseline fix: a pre-release baseline should no longer be refused with "Wrong baseline version".
- On the Service page, confirm the Pre-Release field shows a tag (0.4.3c or newer), not a branch name, and that "Upgrade to Pre-Release" is disabled when the installed version already matches.
- #388 is verified by logic-port tests, not yet against a live MQTT broker - worth an end-to-end check on a camera with MQTT enabled.
- The install flow is the main new surface: try an upload + install with config present (now on the Service page), and a reinstall with a different package version.
Thanks to @grzesiczek1 for the RTSP fixes (#495, #496).
Full Changelog: 0.4.3a...0.4.3c