Repository navigation
B45: port upstream #24846, #24864 and #24854 - #223
Merged
Merged
Conversation
Most of upstream's grab bag was already fixed on the fork by B36 (backend) and UI155 (web), each in its own way. The conflicts were resolved toward the fork, and only what the fork lacked was taken. Ported: - preview.gif and preview.mp4 measure the clip duration (-t) from the clamped start, so a preview that starts after start_ts no longer reads past the requested end - the exporter discards the row and partial files when an export fails before or during ffmpeg (an exception, or DoesNotExist while building the command), instead of leaving it in progress until a restart, and skips an empty thumbnail path when discarding - the password strength meter is removed (it scored the old 8 character rules); editing the password revalidates the confirm field, and the set password dialog shows a live match or mismatch indicator - push registration clears the unsaved registration message when it fails - the `t` shortcut unmutes camera audio when it starts two-way talk, like the button - live tiles default to volume 1 when the group has no saved setting - dependabot holds pandas below 3.1 until numpy 2 - upstream's tests: in-progress export delete, recordings/unavailable scale=0, and the first minute of the time-of-day filter (its event times are now on the hour in UTC; a fork comment in that test is updated) Kept the fork's version (already fixed): the unknown-user login hash (B36's verify_password against a fixed hash; upstream's new unknown-user warning log is 0.19-only and left out), the recordings/unavailable scale, vod/event padding (B36 also caps it at 3600), preview offsets and earliest preview, the in-progress bulk delete, the events time-of-day filter (B36's DST-aware time_of_day_clause), the recording share link base path, Last N hours export math, the detail stream camera switch, per-group live audio (groupAudioStates), per-trigger highlight timers (use-timed-highlights), the media sync 409 toast, the known plates unnamed entry, and the refused subscription toast. The set password dialog keeps the fork's dropped autoFocus (an accessibility fix). Left out: removing the severity order_by in /review. The fork's B37 sorts by severity then start_time in one order_by call so alerts stay ahead of the limit; upstream's change keeps only the start_time sort. (cherry picked from commit 96ce75c) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…r#24864) Timers that measure elapsed time move to time.monotonic(), so a wall clock step (an NTP sync at boot, a manual change) can no longer hold every push notification in cooldown, end a VLM watch job early, pause state classification, freeze Birdseye, make the detector watchdog restart a healthy detector, or make the camera watchdog restart ffmpeg or never restart it. The camera watchdog's no frames check now tracks when the capture thread's frame time last changed, since that value is a wall clock frame time. Stats convert the detector's monotonic detection_start back to wall clock time for the API. Conflicts, resolved toward the fork: - frigate/video/ffmpeg.py: the fork split upstream's inline watchdog loop into helpers (_check_detect_process, _check_record_processes, _check_shared_sub_stream) and sends status through the B32 status cache. The loop kept the fork's structure; the no frames check moved into _check_detect_process, and the disable path uses monotonic time with the fork's _send_*_status helpers. - frigate/stats/util.py: kept the fork's _runner_device next to the new _monotonic_to_wall. - frigate/comms/webpush.py: imports only; the fork does not use Iterator. Fork follow-ups in the same file, so the watchdog never mixes clocks: the D30 hwaccel retry restart now records last_restart_time on the monotonic clock (a wall clock value there would have blocked every later restart), and a stalled recorder restart sends its offline status with a monotonic resend time like the rest of the loop. test_fork_cov_comms_webpush.py drives the cooldown clock through webpush's time module and expects -inf as the unset cooldown time. Unchanged: the fork's SV12 watchdog heartbeat and SV6 outage tracker keep the wall clock. The heartbeat is published as an age in stats, and the outage tracker compares against frame times. (cherry picked from commit 2b4397b) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review summary prompts state the length of the scene (no more than five sentences) and shortSummary (one sentence) fields, since grammar constrained providers cut text off at the schema cap, and every user facing field repeats the rule to use a named subject's name. The observations guidance asks for chronological order from the first frame, and the frame annotation guidance says a lost track can mean the object went out of view, not only that it left the frame. Tracker notes title-case sub labels the way the Objects in Scene list does, and an object already tracked before the first sampled frame gets one "already in view" note instead of its earlier history. Conflict: frigate/genai/prompts.py, where the fork had already replaced the em dashes in the base observations and scene guidance and the detailed scene preset. Upstream's new text has none, so it was taken as is. (cherry picked from commit 47ea942) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- test_fork_cov_record_export.py: an export whose source rows are missing now drops its row (the fork test asserted the old in-progress row), and so does one whose run raises; discarding skips an empty thumbnail path (blakeblackshear#24846) - test_genai_review_tester.py: follows the tester's new load_frames and followup_loop signatures, expects "default" to restore the current field guidance, and covers frame captions being loaded and sent (blakeblackshear#24854) - test_fork_cov_jobs_vlm_watch.py: the fake clock answers monotonic(), which the job's deadlines now use (blakeblackshear#24864) - test_http_fork_cov_media.py: preview gif and mp4 from a preview that starts after start_ts read only up to end_ts (blakeblackshear#24846) - SetPasswordDialog.test.tsx (new): no strength meter, a live match line, and editing the password clears the confirm field's stale mismatch error (blakeblackshear#24846) - test_record_watchdog.py: the watchdog harness drives ffmpeg's time module from the same hand-moved clock as datetime and binds _no_frames_received, since the loop now paces restarts on the monotonic clock (blakeblackshear#24864); a new test fails if the D30 hwaccel retry records its restart on the wall clock again - fork/ledger/B45.md Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars.
 |
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Ports three upstream 0.19 fixes onto
nextas reviewedcherry-pick -xcommits, under ledger ID B45 (fork/ledger/B45.md).Upstream PRs
What changes
-tis measured from the clipped start).tshortcut and the mic button unmute audio for two-way talk; tiles with no saved volume play at full volume.Conflicts and how they were resolved
auth.py,event.py,api/export.py,review.py, the recordings query params,frigate-api.yaml): already on the fork through B36 with the fork's own shape. Kept the fork's version.test_http_eventtakes upstream's timestamps; one fork comment was updated.ExportDialog,DetailStream,recordingReviewUrl,MediaSync,TriggerView,KnownPlatesField): already covered by UI155. Kept the fork's version.media.pypreview handlers: took upstream's clip math and kept the fork'sasyncio.to_threadquery.SetPasswordDialogandCreateUserDialog: dropped the meter, kept the fork'sSessionsEndNoticeand its removal ofautoFocus; the revalidation call isvoid form.trigger(...)for the floating-promise lint.NotificationsSettingsExtrasandLiveCameraView: took upstream's behavior and kept the fork comments,talkState === "available", mic title and disabled logic.frigate/video/ffmpeg.py: upstream's watchdog loop does not exist on the fork, which splits it into helpers with a B32 status cache. Ported_no_frames_receivedinto_check_detect_process, and made everylast_restart_timemonotonic, including the fork's own D30 hwaccel retry and record stall restarts, so the comparisons stay on one clock._send_roles_offlinegetstime.monotonic().stats/util.py: kept the fork's_runner_devicenext to upstream's_monotonic_to_wall.genai/prompts.py: took upstream's text (the fork had only removed em dashes there, and the new text has none).Left out
/review. Not ported: B37 keeps alerts ahead of the limit.--checkpasses).Tests
test_fork_cov_record_export(missing rows and unexpected errors discard the export),test_http_fork_cov_media(gif and mp4 durations),test_record_watchdog(monotonic fake clock, plus a hwaccel retry that does not block later restarts),test_fork_cov_comms_webpushandtest_fork_cov_jobs_vlm_watch(monotonic fakes),test_genai_review_tester(new signatures and captions), vitestoverlay/fork/SetPasswordDialog.test.tsx(no meter, and a fixed mismatch clears; it fails without the fix).make check-fast: passes (lint, typecheck, ratchet, vitest, i18n, ruff, mypy, gitleaks, scripts, and the full Python suite in Docker). One run hit a load-related timing flake intest_fork_api_responsiveness, which passed 3 of 3 on rerun; the final run was green.🤖 Generated with Claude Code