RenderJobsThatCannotBeReadNeverStopTheEngine timed out on the shared runner
(PostgreSQL 17) while the upload's background jobs ran first: the 15 s wait was
the one job wait not scaled by CHRONOS_TEST_TIME_FACTOR.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The scrolling page list could shrink to its padding, hiding every link below about
255 px of height; it now keeps one link's height and the sidebar scrolls past that,
without a horizontal scrollbar. The panel test pins the About rows that fill in late.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A whole folder imported again to the server sent its folder name as the
sequence title on every run, undoing a rename made on the server. The
sink now sends a typed title as before, but the folder's default name
only to a sequence that has no title yet (it reads the sequence first).
The archive already applied a title only when it created the sequence.
A session that ends on another day, as a whole folder often does,
showed only its first date ("28/09/2026 · 18:02 → 18:05"), which read
as a few minutes on one day. Its plan row and Show plan heading now
show both dates, from two new invariant resources (with the
right-to-left arrow).
The Gap and Minimum boxes now show their tooltip while disabled, so the
reason Whole folder turns them off reaches them, not only the Sessions
check box. Doc comments that said a merge needs the same camera now
mention the whole-folder rule.
Tests: a server whole-folder import run twice keeps the server's rename
and still applies a typed title; multi-day time text in both reading
directions; and the Import view model's Group rules (folder only, cards
planned by sessions, switching drops splits, tooltips say why).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The release step decided :latest and the releases repository's content
after the Gitea release was already public, and could then fail or decide
wrongly (owner review 2, phase 1 review):
- Re-publishing v0.1.0-rc.1 or rc.2 builds Chronos.Tools from the tag,
which has no release-channels: the step failed after publishing and the
image job never ran. Any failure to decide (a tag older than the
command, an unreadable page, a body that is not a list) now warns and
moves no channel, so the version's image is still pushed.
- Only the first page (50, newest first) of releases was read, and
pre-releases now share it with stable ones: 50 pre-releases after a
stable release hid it and moved :latest to a pre-release. The workflow
reads every page up to the first empty one (at most 100) and
release-channels takes one --releases file per page (Args.GetAll).
- The Version step accepted tags release-channels refuses (v01.2.0), which
then failed only after publishing. It now uses release-channels' SemVer
2.0 pattern without +BUILD, and a test keeps the two in step.
BRIEF §19.3, DECISIONS, CHANGELOG and README step 5 say so.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Unraid share tooltip told users only to map the share's array view to
/space. With CHRONOS_SPACE_PATH unset (the shipped docker-compose file, or a
template whose Space path was cleared) the server never looks at /space, so
following the tip changed nothing. The tip (en, it) and BRIEF 14.8 now also
say to set CHRONOS_SPACE_PATH=/space; the English text was shortened to stay
within the 200-character tooltip budget.
A render node completing its result on a full archive volume answers 507
before the upload moves to Committing, but no test covered that path. The new
VolumeSpaceTests case checks that the upload keeps receiving with its chunks
and the node keeps the job, then that the same call succeeds once there is
room.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The About entry of the Settings search always listed "Translation" among its
keywords, and its tooltip named the translation, but that row shows only
outside English. In English a search for "Translation" left About alone in the
list and selected a page without such a row. The page's search words are now
read at each search, so the keyword follows the language while Settings is
open, and the tooltip no longer names a row that may be hidden.
Also drop About from the doc comments of the update notice and the
diagnostics bundle: since owner review 2, R5 only Settings › Updates and
Settings › Advanced offer them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The owner asked to fold the About section into Settings to save room in the
side rail (owner review 2, R5). The rail now docks Settings alone at the
bottom; About is the last Settings page with this copy's identity: version,
build, runtime, system, licence link, third-party notice and, outside English,
the machine-translation notice with its review link.
Nothing is offered twice inside Settings: the update check, its notice and the
releases link live on Updates, the data folder and the diagnostics bundle on
Advanced, so About's duplicate buttons and its own diagnostics code are gone.
AboutViewModel stays as a plain helper owned by the shell (Chronos.App.About),
which keeps every Shell.About binding and the update check on start.
SectionKey.About, AboutView and the rail strings are removed (orphans pruned
from every culture, which also re-sorts two Setup tooltip keys); the capture,
the smoke tests, DESIGN, ARCHITECTURE, the desktop mockups and the docs that
pointed at About follow.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Owner review 2 (R4, R5): the Settings icon, a circle with rays, read as a sun,
and the web panel's sidebar spent an entry on About.
- branding/ui-icons/icons.py draws Settings as a toothed gear (8 teeth, centre
hole) with a new gear() helper and now also writes the web admin's icons.js,
so both programs and the panel share one path; it writes UTF-8 with LF on
every OS (on Windows it wrote Icons.xaml in the ANSI code page).
- The panel's menu has a top group (Dashboard, Users, Archive, Jobs, Logs) and
Settings docked at its foot above the account; collapsed, narrow and RTL
layouts keep working.
- About (version, API, database, uptime) is the last, read-only group of
Settings; pages/about.js is gone and #/about opens Settings on that group.
The server name and the fingerprint rows are dropped there: the Server and
Network groups already show them on the same page.
- Admin.Nav.About(.Tip), Admin.About.Name and Admin.About.Fingerprint are
removed (pruned in every culture); Admin.Settings.About is new (en, it).
- DESIGN.md and the admin mockups follow; the tests read both menu groups and
check that About is folded into Settings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Owner review 2, R9: Import › Folder… gains a "Group" choice for picked
folders: Sessions (as before) or Whole folder. A whole folder, with its
subfolders when Subfolders is on, becomes one sequence per file format;
camera bodies and pauses are ignored, while RAW+JPEG of the same shots
stays two sequences so no frame is ever doubled. Only the user's splits
break it, and Merge joins a split part back.
The planner implements it (ImportOptions.Grouping, Core, Linux-tested):
the session key is "folder-" plus a hash of this PC's name, the full
path and the format, so an interrupted or repeated import continues the
same sequence; the default title is the folder's name. The plan shows
one row per format with the format and the number of cameras. Cards and
drives always import by sessions; the choice is not part of a card
profile and a folder starts on Sessions.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On Unraid the archive is a user share (FUSE shfs), which reports the share's
primary storage, often the cache, while the mover sends the photos to the
array: the admin panel showed the cache's free space (owner review 2, R2).
- New option SpacePath (server.json) / CHRONOS_SPACE_PATH: the admin panel
shows that folder's volume; empty, or a folder that cannot be measured, falls
back to the archive's volume (one warning, at start, not per request).
- VolumeSpace measures folders (DriveInfo; the path root on Windows) and reads
the filesystem type and mount from /proc/self/mountinfo on Linux (longest
mount point on segment boundaries, octal escapes, optional fields, the top of
stacked mounts), since DriveFormat calls an shfs share only "fuse". It is an
injectable service, so tests simulate a full volume; the /data checks use it.
- ArchiveStatsDto keeps FreeBytes/CapacityBytes for the shown volume and gains
optional trailing members: the write volume's figures (when a space path is
in use and they differ), the filesystem type and UsesSpacePath.
- Admin panel (dashboard and archive): a tooltip on the storage figures names
the volume and its filesystem and, on an shfs share without a space path,
says they are the primary storage and to map the array view to /space; a
"Write volume" line shows the archive's own volume when it differs.
- Completing an upload, a render upload or a render node's result checks the
archive volume for its size plus the 64 MiB reserve before the staging copy
and answers 507 insufficient_storage, the upload still receiving.
- Diagnostics list the archive and space volumes with filesystem and mount.
- Unraid template: optional read-only "Array view" /space (default
/mnt/user0/Timelapse) and "Space path" CHRONOS_SPACE_PATH=/space; compose and
the releases README describe the same as an option.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Only pre-releases (v0.1.0-rc.1, rc.2) are published, and :latest followed the highest stable release only, so the
Unraid template and the compose file, which pull :latest, found no image while :nightly was pushed every night.
The new Chronos.Tools command release-channels reads the published releases (stable and pre-releases), ignores drafts
and tags that are not SemVer, counts the tag itself as published and compares by SemVer 2.0 precedence. It prints
latest (the public releases repository: highest stable release only) and image_latest (:latest: the highest stable
release, or the highest pre-release while none is stable). Re-publishing an older tag never rolls either back.
release.yml uses it in place of curl|grep|sort -V; the image job pushes :latest on image_latest. README, the public
README, the compose file and ARCHITECTURE explain :latest, :<version> and :nightly.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Gitea writes a release's html_url and browser_download_url with its ROOT_URL,
which on the owner's server is the plain-http LAN address
http://192.168.30.23:3000. The update check accepts only https links, so it
ignored v0.1.0-rc.1 entirely. The page and download links now come from
ChronosReleasesUrl, the address the build names for its users; only the
tags, file names and sizes are read from Gitea's answer.
The lease renewal test's timer periods follow TestTiming too: on the shared
CI host a 100 ms renewal timer came 700 ms late once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Alberto Balbo <alberto.balbo.96@gmail.com>
The owner renamed the product and moved the repository to the self-hosted
Gitea (https://gitea.encke-hake.ts.net/BalboSoft/Chronos) for more control,
and asked for Actions and a first automatic release with the server package
and the Windows exe (docs/requests/2026-10-23-chronos-gitea.md).
- Chronos everywhere: a byte-exact rename of contents and paths, then a
per-language review of the forms the old name's ending had dictated.
- Gitea instead of GitHub: the update check, licence, issue and support
links read ChronosReleasesUrl; the update check takes the highest stable
version, offers a pre-release build its final release and reports a
repository it cannot see as a failed check. EULA, security policy and
Gitea issue forms live at the repository root.
- Gitea Actions for the Linux act_runner (.gitea/workflows): CI, release,
nightly image and performance gates in the .NET SDK container with git
checkout and the system SQLite. The release cross-compiles the Windows
programs, checks their icon, version and DPI manifest, packs installers
and zips, publishes the Gitea release and pushes the server image
(REGISTRY_TOKEN). Windows-only tests run in an opt-in job on a Windows
runner (WINDOWS_RUNNER).
- New Chronos.Tools commands: zip, gitea-release, postgres-ci (replaces the
docker-exec script, which Gitea's service containers do not allow; the
throw-away server runs without durability) and pe-check.
- Timed tests take their budget from TestTiming (CHRONOS_TEST_TIME_FACTOR,
3 on the shared runner; the weekly performance gates keep theirs).
BREAKING CHANGE: programs, settings, data folders, file types, environment
variables and the Docker image use the new name; nothing is migrated from
Titano, which never had a release.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Alberto Balbo <alberto.balbo.96@gmail.com>
- Implement tests for picker behavior, including loading sequences, searching, and handling empty libraries.
- Validate sequence opening via Enter key and double-click actions.
- Ensure proper handling of unsaved edits and sequence deletion scenarios.
- Verify accessibility features for picker rows across different languages.
- Introduce a dialog counting mechanism to manage unsaved edit confirmations.
Owner review 1, phase 3a (docs/requests/2026-10-15-owner-review-1.md, R1).
The desktop rail is 200 px wide, so the section names fit in every
language, and collapses to 48 px of icons with its ≡ button or Ctrl+B.
Collapsed, each icon's tooltip names its section before saying what it is
for, while screen readers keep the name and the description. The choice is
written to the settings file the moment it changes (nothing else restarts)
and survives Settings › Reset all and Import; Ctrl+B waits behind the
first-run dialog like Ctrl+1…6, and a held key toggles once.
The admin panel's sidebar is 256 px and collapses to icons the same way,
with its own ≡ button or Ctrl+B (on the physical key where the layout's B
types no Latin letter, so Cyrillic, Greek, Hebrew and Arabic layouts work).
The choice is kept in the browser, for the page when the browser refuses
storage, and the tables fit their columns again after each toggle. At
900 px and below it stays the top bar and Ctrl+B stays the browser's.
The wider rail leaves 49-72 px less for every screen at the 1200 px
minimum. Six adversarial review rounds found what that cut, and in every
shipped language no action is cut off now: the Jobs columns share their
width (Target and Progress give way first, Kind/Where/State capped on wide
windows), the editor's preview tools take a second line, a server job's
Cancel has its own header column, the Archive tree keeps a year's label,
the Import rows and result, the certificates and cache rows and the frame
table fit. Headers, titles and cells that may be shortened end in an
ellipsis and show their text in their tooltip; state tooltips start with
the state, and their explanations were reworded not to repeat it (a queued
job "waits for its turn", jobs running one at a time). The crop-result
tool sits beside Crop mask, disabled with its reason while the mask is
off. Tests check these rows in all 38 languages at 1200 px; the App test
program now shuts WPF down in order, so its exit no longer crashes after
every test passed.
Phase 3 lands as two commits: R7 (the Edit section) edits the same shell
files and follows. Also updates BRIEF (12.1, 14.8, 24), DECISIONS,
CHANGELOG, STATUS, the request file, DESIGN and the mockups; every new or
reworded string is in all 38 languages.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Owner review 1, phase 2 (docs/requests/2026-10-15-owner-review-1.md, R2).
Jobs now run one at a time on the desktop and on the server. Parallelism
(Settings > Rendering, TITANO_PARALLELISM) is how many photos the running
job processes at once: thumbnail and proxy passes draw that many photos
together and still write the pack and checkpoints in frame order, decode
tiers are probed that many at a time, the deep analysis measures that many
frames and a render decodes that many ahead. Wherever originals are
developed whole (TIFF, PNG, RAW without a usable preview), fewer run at
once within the memory budget, which the running job now has whole; the
server's frame window is sized to the host instead of 1 GiB.
A new job waits its turn and pausing the running job starts the next one.
An optional priority queue gives each job kind a level (renders, imports
and downloads high, verifies and proxies low by default): a queued job of
a higher level pauses the running one without asking, and that job goes
back to the queue with its checkpoint and resumes by itself afterwards.
Jobs of the same level, and a job's own follow-ups, never interrupt it.
The Jobs screen and the admin panel show the parallelism and the queue;
Settings > Rendering and the admin panel's Settings set them, at once.
TITANO_WORKERS and the Workers key of server.json still set the
parallelism (1 to 64, with a warning at start for the variable).
Titano.Tools gains i18n prune, which removed the renamed keys from the 36
machine-translated languages; the new strings are translated in all 38.
Found while testing: uploads and imports failed when a pass read the
sidecar while they wrote it (atomic writes now retry the rename while a
reader holds the file, and never for a read-only one), and taking a lease
just released while another process read it failed on Windows (the create
now waits for the reader). Five adversarial review rounds checked the
phase before this commit.
Also updates BRIEF (12.8, 14.6, 14.8, 14.9, 24), DECISIONS, CHANGELOG,
ARCHITECTURE, STATUS, the request file and the compose/Unraid templates.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Owner review 1, phase 1 (docs/requests/2026-10-15-owner-review-1.md).
Thumbnail and proxy jobs over several sequences counted sequences while
they checkpointed inside one: the rate decayed until the time left
overflowed TimeSpan, the job failed while publishing progress and again
while saving it, and stayed Running. The throughput is now seeded by the
first window with progress, the time left is bounded (JobEta), saving
progress never throws, image passes count frames, and job speeds use the
right unit (bytes for verify and import).
The debugger stops in WicDecoderProvider.cs when presets changed were
cancelled preview decodes leaving synchronous thread-pool delegates; the
editor's sources, LatestRunner and the render node's proxy task now keep
them in their own tasks, and a cancellation is never an unreadable frame.
GoPro DNGs converted by Adobe software were classified GPR and read only
through their 8-bit preview, and Windows' preview-only DNG decoder was
reported as a native decode. A DNG is GPR only when its raw image is
VC-5, WIC decodes RAW only through the Raw Image Extension and never DNG,
analyses and previews take an adequate embedded preview by choice while
renders develop natively, and frames already indexed as GPR are
reclassified by the next light analysis.
Also records the owner's request (docs/requests), the BRIEF amendments
(§8.1, §8.2, new §24), DECISIONS, CHANGELOG, STATUS and ARCHITECTURE.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The orphans fix and scans treated a folder whose only new photo could not be
read yet (held open, no permission, an offline placeholder) as changed: the
sidecar was rewritten, the covering flags cleared and analyses queued again at
every click. SequenceAdopter.Reconcile now returns null when it indexed nothing
and changed no flag, as its contract says.
A deep pass counted as current by frames version alone, so a pass whose verdict
a proxy made stale overwrote the report a later pass made from the proxies,
then dropped the deep status. It now counts as current only when its report
describes the frames as they are, and otherwise replaces only a saved report
that does not describe them either.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A desktop scan queues a background deep analysis of the sequences whose
frames it changed when Settings > Analysis > After a scan is on; photos the
Archive screen's orphans fix indexed got only the light analysis, which then
dropped the deep status. ArchiveService.ReconcileAsync takes that setting
and queues the deep analysis too, at background priority. A follow-up is
left out when a queued job of its kind already covers the sequence, whatever
else that job targets, as the scan's follow-ups check.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A deep pass overtaken by a frames change that keeps the frame count, with a
light analysis queued, leaves an analysis.json of that count in place rather
than save its own. It then still recorded its own report: the deep status,
summary, warning count and Auto values described a report that is not on
disk, and the queued light analysis kept that summary until the next deep
pass. SaveReport now says whether it saved; when it did not, the pass reads
the saved report as a light analysis would and records that one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Each index of a shared archive bumps a sidecar put back behind its own row
to one version past that row, which can be the version another index holds
for other frames (a desktop rebuilding a damaged sidecar while the server is
one upload ahead). That index then took the sidecar's frames as unchanged:
it kept the project's positions and what covered the frames, and clients and
editors holding that version saw nothing. When the versions match, a scan
now also compares the row's frame ids and order with the sidecar's; other
frames are read again under the lease and written one version past the row.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The Archive screen's orphans fix indexed photos copied in by hand but left
AnalyzedLight, ThumbsBuilt and ProxiesBuilt set and recorded the folder's
stamps, so no scan came back to it and the light analysis was never redone:
Auto values, warnings and preview-only tiers stayed those of the old frames.
It now clears the three flags in the sidecar and the row as a scan's
reconcile does, and queues the light analysis on the host's engine (on a
server, the thumbnails and proxies too).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A deep pass saved analysis.json as soon as it finished, before it checked
that its frames were still the sequence's. A pass overtaken by an upload, or
by another Titano sharing the archive, replaced the report a later pass had
made of the frames as they are: the report endpoint answered 404 and the
next light analysis dropped that pass's deep status. The report now waits in
memory and is saved under the sequence lock, when its frames are current or
the saved report describes the frames as they are no better; the checkpoint
is deleted only then.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Each tag upsert locks its row until commit, and tags were written in the
order the caller listed them. A PATCH listing them in another order than
an upload, a scan or another edit of a sequence sharing two of them could
lock the rows in opposite orders, and PostgreSQL aborted one writer with a
deadlock that nothing retries. Tags are now written in the order of their
keys, and a set the sequence already carries is not written again.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
ApplyLocallyAsync recorded the folder's scan stamps unconditionally after
writing the server's title, tags, rating or notes into the sidecar. Photos
another process (a desktop sharing this archive) had listed there since
this index last synced the folder then stayed out of the index until the
folder changed again, and the row's frames version stayed behind the
sidecar's, so analyses here recorded nothing for the sequence meanwhile.
It now asks ArchiveIndexer.InStepAsync before its sidecar write, as the
other partial writers do, and records the stamps only when that held.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
An analysis compared only this host's index row with the frames version it
read at its start. On a desktop sharing a server's archive, the server's
upload bumps only the shared sidecar until the desktop's next scan, and a
desktop scan clears nothing. A desktop pass that recorded in that window
wrote Auto values and statuses for the old frame count over the server's
analysis in the shared project.json and sidecar, and nothing analysed the
photos again.
Under the sequence lock the analyser now reads the sidecar first. When its
frames version is not the row's, the photos changed in another process and
this index does not hold them yet: the pass records nothing, logs
Advisor.Log.FramesChangedElsewhere and queues nothing, since a light
analysis here would read the same index. The process that changed them
analyses them, and this host's next scan indexes them (a server's scan
then queues their analysis). A moved row still queues a light analysis
unless one is queued, in which case the pass records as before.
Tests cover the shared-archive case with a second index and engine, and
the queued-light escape on a one-worker engine.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
FrameRepository.UpdateLightStatsAsync updated every frame of a sequence
in batches without locking the sequence row. On PostgreSQL a scan's or
rebuild's sync holds the row and rewrites every frame in its own order,
so a deep analysis writing its statistics at the same time could hold
some frames while waiting for others the sync held, and PostgreSQL
aborted one of the two jobs with a deadlock.
On PostgreSQL the method now locks the rows of its frames' sequences
first, in the same transaction, like every other writer of a sequence's
frames. SQLite's write lock already serialises writers.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
An upload batch that found the folder out of step resynced the sequence
from its sidecar and recorded the photo directory's current stamp. When
the photo directory had changed since the last scan (a photo copied into
originals/ by hand or by another process), that stamp hid the change from
every incremental scan, so only a full scan or the orphans check indexed
the photo.
The batch now asks ArchiveIndexer.PhotosChangedAsync under the lease
before it files its photos. When the photo directory had changed, the
resync records the photo stamp the index held before (none when none was
recorded), so the next incremental scan still reconciles the folder.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A proxy download marked the cache filled whenever the server's sequence had
ProxiesBuilt, even when frames added since the server's last proxy pass got
a thumbnail or nothing (the flag offers their proxies, the endpoints answer
404 until the pass draws them), and frames without any cached file were not
counted at all. The agent never compared such a cache again, so the editor
kept thumbnails or blank frames until the frames changed.
The cache now counts as filled only when every frame, missing or damaged
ones aside, is cached as a proxy. On a server with ProxiesBuilt, a download
that left stand-ins is queued again once the server's update time moves, so
a frame the server cannot serve yet is not fetched at every refresh.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A light or deep analysis read its frames at the start and, under the
sequence lock, recorded AnalyzedLight, the warning count and the advisor's
Auto values for them whatever happened meanwhile. Frames added by an upload,
an import or a scan's reconcile then kept advice and warnings counted on the
old frame count, over a newer light analysis, and scans never queued another
one because AnalyzedLight was set again.
Under the lock the analysis now compares the row's frames version with the
one it read. When it moved and no light analysis is queued for the
sequence, it records nothing (a deep pass keeps only analysis.json and its
per-frame measurements), and the job logs it and queues a light analysis of
the frames as they are. A queued light analysis replaces what the pass
records, as for image passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
ArchiveIngest appended the new frame rows and recorded the folder's stamps
whenever the batch appended at the end and the frame count matched, even
when a desktop sharing the archive had changed the sidecar alone (a
verify's hashes and Missing/Corrupt flags, decode tiers, duplicates, a
title). The server's frames then never got those changes, and the stamps
hid them from every incremental scan until a full scan.
A batch now asks ArchiveIndexer.InStepAsync under the lease, before it
changes the folder, and appends only when the index held the folder as it
was; otherwise it syncs the whole row and every frame from the sidecar.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The scan's plain branch indexed a sidecar read without the lease. When an
upload or a partial writer (decode tiers, a verify) committed between that
read and the scan's transaction, the scan wrote its older copy over the
newer row and frames (dropping the uploaded frame rows), then recorded
the folder's current stamps, so incremental scans skipped the regression.
Under the row lock, SyncAsync now refuses an unleased read when the
sidecar no longer has the stamp the scan listed or the row holds another
frames version, and the scan reads the sidecar again under the lease. An
unleased sync records the stamps it was checked against, and the index
snapshot files each sidecar under the stamp it was read at. Under the
lease, a sidecar behind the locked row (an older copy put back) is
written again one frames version past the row, so the version never goes
back. Statuses cleared in this host's row only (Uncovered) now pass as a
parameter, so that rewrite never clears them in the shared sidecar.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A verify's flush replaced every flag of a pending frame, in the sidecar
and the index, with the flags it computed from the frame as read without
the lease, possibly hours earlier. A preview-only or undecodable tier, a
duplicate flag or a repair recorded meanwhile was reverted, and a photo
an upload repaired during the run was flagged missing again, so the next
copy moved the good file aside.
Under the lease the flush now merges only Missing and Corrupt into the
frame's current flags, and the index sets and clears only those bits with
the hash, after locking the sequence row. A check whose sidecar record or
file (size and modification time taken before it was read) changed since
is left for the next verify.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
FrameTiers.RecordAsync updated the frames of a sequence before its row,
while a scan's sync locks the row (FOR UPDATE) before rewriting every
frame. On PostgreSQL the two deadlocked and one job was aborted with
40P01. The tier writer now records the row first; the verify's integrity
flags (next commit) and maintenance's duplicate flags take the row lock
in the same transaction as their frames, so every writer of a sequence's
frames follows one order. A PostgreSQL test that deadlocked with the old
order covers it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Writers that change part of a sequence (decode tiers, image-pass flags,
analyses, verify flushes and stamps, duplicate flags, title/tags/rating/
notes edits, mass rename) now ask ArchiveIndexer.InStepAsync before their
own sidecar write and record the folder's scan stamps afterwards only when
the index held the folder as it was: the stored stamps are still the
folder's and the row holds the sidecar's frames. What another process
changed in the sidecar alone (frames, a new title) stays visible to the
next incremental scan, which writes the whole row from the sidecar.
EditSequenceInfoAsync reads sidecar and row under the lease and writes the
row through SequenceRepository.EditAsync at the settings version it read,
applying the edit again on a conflict, then the sidecar from the stored
row; settings and frames committed meanwhile stay. MassRenamer writes only
folder and title (RelocateAsync, whose null title now keeps the current
one). Verify flushes and duplicate flags run under one lease.
A scan that finds frames another writer listed in the sidecar itself
clears AnalyzedLight, ThumbsBuilt and ProxiesBuilt only in its own row and
only on a server (ImagesAfterScan); the shared sidecar keeps the writer's
flags, so a desktop's poll no longer makes the server redo its passes.
An image pass overtaken by an upload keeps its flag while a pass of the
same kind is queued for the sequence. ThumbPack.IsFrom accepts a thumbnail
drawn before a verify hashed its photo, in the pass, in SameFrames and in
the server's thumbnail lookup, which answered 404 for every such frame.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Telling copies of a photo apart counted a frame not hashed yet with the
same size and capture time as a hashed photo's copy, although it may be
another photo (a RAW of constant size in a burst without sub-second
times). Positions then shifted between copies: an exclusion waited forever
as a pending frame, or landed on the other photo. A hashed photo is now the
frames holding its hash; size and capture time count only when no frame
holds it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The agent recorded the server's update time for a cache of stand-in
thumbnails before the proxy download it queued, so a download that failed
(server restart, network drop) was never queued again once its backoff
ended: the server's sequence may never change again. A comparison that
offers proxies is now recorded only once their download is queued, and
dropped when that download ends other than done.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A render queued before the editor had seen frames inserted under it (a sync
download, an import of earlier photos, a scan without an event) embedded the
project's positions in the old frame order. Render now checks first that the
frames read are still the sequence's (ISequenceSource.FramesCurrentAsync: the
row's frames version on this PC, the manifest on a server). On this PC the
sequence is read again and the positions follow the frames; a server
sequence whose frames changed turns outdated and the render is not queued.
Reads of the sequence no longer overlap: a read waits for the one under way,
saves that meet new frames read them through the same slot, and they save
the project as it is once its positions followed. The frame on screen stays
on its photo when frames are inserted or removed before it.
Tests: the event-loss test waits for the listener's subscription before
dropping the stream, and the conflict tests flush once more after the
choice when a retry started before it was still failing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A read of the open archive's settings that an Archive row edit overlapped
was dropped and left to the edit's end, so a folder change could complete
while the rows still showed the previous archive. The read now waits until
no edit is being written and reads again, so the retarget and the reload on
showing Settings are done only once the rows show what the archive holds.
A smoke test edits a row while Settings reads the archive, with the
settings lease held elsewhere.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A deep analysis kept the report it had just made, measured from the frames
loaded at its start. Proxies that arrived during a long run make its verdict
on those frames' originals stale (every later reader drops the report), yet
the run applied its advice and recorded AnalyzedDeep, HasWarnings and the
unreadable frames' warnings. The light analysis the proxies queued could have
finished on another worker before, and nothing ran it again.
The run now loads the frames again under the sequence lock and keeps its
report only while it describes them (SequenceAnalyser.ReadReport's rule); the
Auto values are applied under that same lock as the status
(ProjectStore.UpdateLockedAsync), so a light analysis either ran before and
is not undone, or runs after. Its scan stamps are recorded only while the
index holds the sidecar's frames.
Test: a deep analysis of HEIF files without an Exif thumbnail, during which
their proxies arrive, records no deep status and leaves the warnings and Auto
values of the light analysis that read the proxies.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC