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
Commit 90bf598 rewrote docs/DECISIONS.md and docs/CHANGELOG.md with only the
fourth pass's entries. Both files are back as they were at 8e8f802, with those
entries appended, and the proxy-cache row the new one replaces is marked
superseded.
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
The Windows README removed only the prompt's own account from a data folder
created by an unelevated console run, which missed a console run by a
standard user. It now removes the account the service's refusal names.
The DECISIONS rows on OneDrive sync roots and on the service's start-up
log say which later rows partly supersede them.
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
A desktop sharing the archive that imports into an existing sequence lists the
new frames in the sidecar itself and keeps its ThumbsBuilt and ProxiesBuilt.
The server scan cleared the flags only when SequenceAdopter.Reconcile found
unindexed photos, so it queued no image pass: the new frames, and every frame
they renumbered, answered 404 on /thumb and /proxy until a pass was forced.
A scan now also clears AnalyzedLight, ThumbsBuilt and ProxiesBuilt when the
sidecar's frames version differs from the index row's, decided under the
sequence lease, and its follow-ups queue the light analysis and both passes.
Writers that change only part of a sequence wrote the whole index row back
from a sidecar read under a lease they had already released (decode tiers) or
without any (verify stamp): an ingest or a PATCH committed in between was
undone, and a row written from a sidecar the index had not synced claimed its
frames version, hiding the change from the scan. FrameTiers.RecordAsync and
ArchiveVerifier now write the sidecar and only their own columns
(SequenceRepository.RecordTiersAsync, RecordVerifiedAsync) under one lease.
They and the image passes record the folder's scan stamps only
while the index holds the sidecar's frames version
(ArchiveIndexer.RecordStampsInStepAsync), so the next scan still sees frames
another process added.
An image pass decided ThumbsBuilt/ProxiesBuilt from the sidecar it read at
its start and set the flag again after a scan had cleared it for photos added
meanwhile, so no pass was queued for them. MarkAsync now holds the lease for
the sidecar and the index, and sets the flag only while the sidecar lists the
photos the pass read at the same indexes; otherwise it clears it.
A rebuild deleted every row whose folder was not among those listed when it
began, so a sequence that got its first frame or was created by an upload
session while it ran lost its row (404 for the Uploader, a second '(2)'
sequence for the session). A row is now checked again under its lease, as it
and its folder are then, before it goes.
Tests: a desktop's ArchiveIngest commit followed by a server scan (Data and
Server, /thumb and /proxy answer 200 for every frame); a thumbnail pass with
a scan's reconcile in between; tier records and a verify stamp interleaved
with a PATCH and with another writer's frames, on SQLite and PostgreSQL; a
rebuild during which uploads place and create sequences.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
With Keep proxies offline, the agent refreshed a proxy cache only when the
server's sequence had ProxiesBuilt. That flag waits for every frame, and a
missing or damaged frame never gets a proxy, so a HEIF sequence with one
such frame kept the 320-px thumbnails cached in place of proxies forever,
although the server offered the other frames' proxies sent by clients.
For such a cache the agent now reads the manifest once per change of the
server's update time and queues the background download when a frame is
offered as a proxy that the cache holds only a stand-in for (or nothing).
The test server's fake serves proxies sent to it, as the real one does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The desktop editor of a server sequence relied on the server's
sequenceChanged event to go read-only when frames changed. Scans publish
none, and an event missed while the socket reconnected was never noticed,
so a save made meanwhile passed the settings-version check and wrote
positions in the old order over the ones the server had moved.
The PATCH now carries expectedFramesVersion, the frames version of the
proxies the editor holds. On 409 the editor reads the manifest: other
frames make it outdated (read-only, Get proxies; the edit stays pending),
the same frames at a new version retry at that version. After a lost event
socket the frames are compared again. Servers that predate the field
ignore it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
An editor open on a sequence of this PC while earlier photos arrived (a
sync download, an import, a scan) kept its frames and positions in the old
order, while the index moved the stored positions with their frames. The
next edit of exclusions, the hero or keyframes was merged as the editor's
own value and stored in the old order, so other photos were excluded and
the Library's hero and the project's named different frames.
LocalSequenceSource records the frames version it read the frames at (row
and frames read as one state) and refuses a save under the sequence lease
when the row's frames version moved (FramesChangedException); the index
edit also names that version. EditorSession then reads the sequence again
and moves its project, its basis and every undo and redo step the way the
index moved the stored copy (ProjectSettings.FollowFrames: by frame id when
frames were only inserted, otherwise unchanged; waiting photos placed),
and saves in the new order. A read after a job follows the frames the same
way, and saves wait for a read under way.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A server folder adopted with the same photo twice ('IMG_0001.JPG' and
'IMG_0001 (1).JPG') holds two frames where the PC keeps one. Translation
by photo sent every position of that photo to its first copy, so the
server's exclusion of the duplicate excluded the PC's only copy, a push
gave the server back one exclusion instead of two, and keyframes of both
copies piled on one frame.
FramePhotos now lists every frame holding a photo and tells them apart by
their order (Occurrence). A position on the n-th copy goes to the n-th
copy of the receiving copy, or waits as a pending frame with its
occurrence when that copy holds fewer; occurrence 0 is not written, so
older projects read the same.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The Windows README told an administrator to run icacls /reset /T on a data
folder created by an unelevated console run, which let every local user read
the database, the TLS key and the temporary admin password until the next
successful start. It now takes ownership with takeown /A /R and removes only
the console account's entries with icacls /remove:g, so nobody else gains
access at any step.
The 2026-10-12 Start with Windows decision said the uninstaller leaves the
StartupApproved value; setup removes it with the entry, as the later Task
Manager state row says.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A library synced with its Sync button lies beside the account's UserFolder
(%USERPROFILE%\<tenant>\<site> - <library>), so a portable copy unpacked
there kept .\data inside a synchronised folder. OneDriveFolders.SyncRoots
also reads HKCU\Software\SyncEngines\Providers\OneDrive\*\MountPoint, where
OneDrive registers every folder it syncs (the location Office reads), and
takes the registry root as a parameter for tests on a scratch key.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A server import stages its copies in this PC's cache folder, yet a full disk
there read 'The destination is full.' beside every photo, and a refused
write 'Titano may not write to the destination.', which sent the user to
the server. IImportSink.StagesOnThisPc (true for ServerImportSink) selects
Import.Failed.CacheFull and Import.Failed.CacheAccess instead.
A failed archive import job showed the job engine's raw text (SQLite, file
system or framework English) as the Import screen's line. ImportFailures.Job
keeps the archive engine's own localized sentences and shows 'The import
stopped before the end.' for anything else; the raw text stays in the
tooltip and the logs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Settings › Archive reads archive.json without the settings lease each time it
is shown and after a waited edit. On a slow share such a read could return
the file as it was before a template, proxy or thumbnail edit that was being
written, then put the old value back on the row and in this PC's defaults
while the archive kept the edit. ArchiveReadGate now drops a read that
overlapped an edit being written; the rows are read again at once, or when
the last edit being written ends.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
PUT /api/v1/frames/{id}/proxy raised Kestrel's 1 MiB limit only inside the
handler, after its 404 and 422 answers and after authentication's 401 and
403. A proxy above 1 MiB left unread then closed the connection, and the
client retried a network error instead of reading the answer. The route now
carries IRequestSizeLimitMetadata (ArchiveImages.MaxProxyBytes) like the
chunk route, so routing applies it before anything answers.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Ten new strings (proxy pass log line, import failure sentences, two server
errors, the OneDrive data notices and their tooltips), translated and
independently reviewed per language against each glossary.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Positions crossing copies could still land on other photos in three
windows:
- A push translated a row snapshot's settings against local frames
listed later, after a manifest round trip: an import of earlier photos
meanwhile moved the row's positions, and the server stored shifted
exclusions, hero and keyframes. SyncPositions now reads this copy's
row with its frames as one state (frames listed between two reads of
the row at the same frames version) and translates that row's
settings, at the settings version being pushed. The conflict
comparison and ApplyLocallyAsync compare against that same row. When
the copy was edited since, or its frames keep changing, nothing is
sent or applied and the next comparison translates again (Keep mine
reports a changed conflict; PushAll sends an edit saved meanwhile).
- The manifest read the sequence row, then paged its frames: an upload
committed in between paired settings of one frames version with
frames of the next. The server now lists them again while the frames
version moved during the listing, then inside a transaction holding
the row (on SQLite, on that transaction's connection, so no second
pooled connection waits behind the write lock). The client also
refuses to translate against a manifest whose frames are not each
listed once at 0..n-1.
- Server capture times were fetched only when something was unhashed,
so the same parked photo was written with or without capturedAt. They
are now fetched whenever the two frame lists differ; an equal project
then compares equal, and an entry can be placed on a photo that
arrives unhashed.
CompareOriginals counted unhashed server frames one-to-one against this
PC's photos, so two copies of one photo in a folder the server adopted
left Download on offer forever while the download fetched nothing. A
photo on either side is now there when any photo of the other side
matches it, as downloads and uploads find them. Paged frame listings
keep each frame once (an insertion between pages lists one again).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
TranslateFramePositions read the stored pending frames, then parked the
project's own hero when the receiving copy lacked its photo (leaving no
hero chosen), and only then placed pending entries: a stale pending hero
whose photo the copy held became the hero again and the newer choice was
lost. When a hero is chosen, stored pending entries no longer carry the
hero flag (an entry left with nothing to place goes) before positions
are translated, so only the hero parked now may wait.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
ArchiveIndexer.SyncAsync and the ingest's in-step append (and the first
frame of an API sequence) read the sequence row and upsert the whole row
from that read in one transaction. On PostgreSQL (READ COMMITTED) a
PATCH accepted between the read and the upsert was written back over:
its settings were lost and settings_version went back, so the PC that
pushed them later pulled the old settings over its edit.
The row is now read with SequenceRepository.GetForUpdateAsync (FOR
UPDATE, as EditAsync takes it). A PATCH committed first is then moved by
the insertion like any position; one arriving later sees the new frames
version and gets 409 frames_conflict.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Uninstalling (or an upgrade dropping a program) deleted the client's or the
Uploader's Run value but left its StartupApproved\Run value behind. A moved
entry keeps that state, so an entry the user turned off stays off.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- ArchiveImages.IsProxyOf compares capture times to the microsecond the index
keeps (sidecars and Exif keep 100 ns ticks), and for frames read through
proxies also applies the policy's size rule, so the proxy pass, uploads,
thumbnails, renders, analyses, tiers, manifests and downloads agree.
- Proxy headers are cached per file while length and write time are unchanged
(and the file has settled), so manifests cost a stat per proxy.
- FrameTiers.RecordAsync checks proxy presence when it writes the sidecar and
returns the sequence's frames as they are, so a deep analysis or render that
ran while proxies arrived no longer flags the frames preview-only again, and
the analysis advises from the current flags.
- Server manifests ignore ProxiesBuilt for frames the server cannot decode, so
frames added since the last proxy pass are asked for their proxies.
- GET /sequences/{id}/analysis/report answers 404 for a report that found
frames unreadable before their proxies arrived (DeepReports.ReadCurrentAsync).
- The deep checkpoint key marks only frames read from proxies, so checkpoints
written by earlier versions still resume.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The Uploader runs for the whole sign-in session and read its sign-in
entry only at start, so an entry Task Manager turned off meanwhile still
showed as ticked. Opening the settings panel now reads it again.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
AppEnvironment now carries AppPaths.SyncedFolder: when the per-user data
folder itself lies in a OneDrive sync root, Titano still starts (nowhere
is safer), logs it and shows the error toast "Data inside OneDrive" with
Open, instead of the false "Data kept outside OneDrive". The latter now
needs PortableSyncRoot (renamed from the old SyncedFolder meaning).
A portable copy now in a OneDrive folder whose .\data still holds an
index or settings.json gets that folder named in the log and in a toast
("Earlier data unused", with Open). Nothing is moved or deleted.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
GET /api/v1/archive/settings refreshes the served archive from
archive.json first, so the admin form (which saves every field) starts
from a desktop's change made since the last poll instead of writing the
old templates back. ServerArchive.RefreshAsync replaces the served
archive only when the refresh registered a change, and only through a
compare-and-swap against the archive it read, so a poll racing an
admin's PUT no longer puts the pre-edit row back for ten minutes.
RefreshSettingsAsync now ignores an archive.json a newer Titano wrote,
as documented, instead of registering what this version reads of it.
AppServicesTests now points the folder at an archive the index does not
list first, and checks that the archive no longer chosen gets no scan.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A retarget now returns the archive registered at its own folder, not at
the folder current when it finishes, so an edit made while folder A opens
no longer lands in the archive of a folder B chosen right after. The rows
and _archive follow only the latest folder; Reset section pushes to the
archive it asked about and reloads the rows when the folder changed. A
newer sidecar met while opening no longer faults the retarget silently:
it is reported like any open failure. The smoke test now waits for the
empty-folder retarget before editing, so the archives[0] fallback would
fail it, and covers the two-folder race.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Authentication answers 401 and 403 before the endpoint runs, with the chunk
unread; over Kestrel's 1 MiB limit the connection was closed and the
Uploader saw a network error instead of being asked to sign in. The chunk
route now carries IRequestSizeLimitMetadata for the largest chunk size a
server allows, which routing applies when it selects the endpoint.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A batched Thumbs or Proxies pass stopped at the first sequence whose folder
was renamed, deleted or damaged before the pass ran, and the others kept no
images until an administrator retried it. Each indexed sequence now gets its
own background pass, labelled with its title and de-duplicated against a
queued pass of its own, as after an upload.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The index's own predicate for sequences created without photos reads the
archive root's sidecar once per call and treats a damaged or unreadable one
as naming nobody. A torn root sidecar is therefore rebuilt under the
root-adopted sequence instead of a later POST /sequences row, and a sidecar
the server may not read fails the root folder alone instead of every scan,
rebuild, orphan check and scan follow-up. Maintenance leaves these rows out
of temporary-file purging and pack compaction, so the root's thumbs.pack is
compacted under the leases of the root's own sequence.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A proxy download caches a frame's thumbnail under its proxy name while
the server has no proxy of it (HEIF until a client or render node sends
one). The cache did not record which it was, so the thumbnail stayed
there for good after the server's proxies arrived (V-29).
- remote.json records each cached file's kind (proxy or thumbnail); a
later proxy download replaces thumbnails, and files of caches written
before kinds were kept, once the server offers the frame's proxy.
- A cache filled with thumbnails while the server had no proxies counts
as stale, so offline caches are fetched again when the proxies arrive.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A folder the server adopted and never verified lists no SHA-256. A
download of such a sequence failed ("did not arrive intact") because
every unhashed frame collapsed into one wanted entry and the empty hash
was refused, and desktops never sent the proxies the server asked for
those frames (V-9).
- Each unhashed server original is fetched by frame id into staging,
accepted on its size (two attempts, resumed with Range) and hashed
here; the archive ingest files it by that hash like any other.
- Server frames without a hash are matched to this PC's originals by
size and capture time: they are not downloaded twice, not uploaded
again, counted right in the comparison (each frame on its own), and
their wanted proxies are counted and sent from the matching original.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The Jobs screen's flush could take a render's end before the editor that
followed it had seen it, and announce it without the output file. It now
yields while the open editor follows that render (FollowsRender); an editor
closed in between still says it once, without the output.
Open on a job over more sequences than a Library page (a maintenance run over
a whole archive) now opens the Library as it is instead of looking every
target up and paging through the list to select them all. Selecting many
cards (Open's selection, Clear selection) counts the selection once instead
of once per card over every card.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Picking archive A, then B, then A again before B's read ended left A's read
marker set while its tree and figures were already cleared: Rebuild, Scan,
Verify and Maintenance were enabled and the rebuild dialog counted 0
sequences. ForgetRead now clears the marker too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A reply that is not the server's JSON (a reverse proxy's 502 page while the
container restarts, its 413 for a large body, a bare 500) reached the Import
screen as the HTTP reason phrase, in English, beside every photo. ErrorText now
uses an ApiError's message only when the server wrote it (a code that is not
TitanoClient's http_<status> fallback) and otherwise says what the status means:
too large, rate limited, not answering (gateway and timeout statuses) or a
server error. A card file that cannot be read while planning is listed with the
executor's catalog sentence instead of the system's text and full path.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Settings synced between a PC and its server sequence were stored as they
came, in the other copy's frame order. When the inserted photos arrived
later (a download, an upload), the index moved those already-correct
positions a second time, so exclusions, the hero and camera keyframes
landed on other photos (V-21, CD-2).
- Settings are translated by photo (SHA-256, else size and capture time)
to the frames of the copy they go to: the server's before they are
compared or applied here, this PC's to the server's frames as its
manifest lists them before a push (SyncPositions).
- A position on a photo the receiving copy lacks waits in project.json as
a pending frame and takes its place when the photo arrives, on an
insertion or an append in step with the index; the Library's hero
frame follows. A hero chosen meanwhile wins, and advice does not fill
a hero while one waits. Old projects without the member are unchanged.
- Pushes send expectedFramesVersion; PATCH /api/v1/sequences/{id} answers
409 frames_conflict when the frames changed since, and the next
comparison translates again.
- A pulled payload equal to the local copy is settled without the
archive: an offline or read-only archive no longer fails the
comparison (V-10).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- Renders, analyses, the server's manifests and proxy downloads take a proxy only
when its Exif capture time is the frame's (ArchiveImages.IsProxyOf); one left at
the index by a frame renumbered since counts as missing and is asked for again
(V-24).
- A frame whose original the host reads through its proxy (HEIF on the server) is
preview-only only while it has none: the light analysis and the proxy pass keep
the flag, the preview-only count, the light finding and the advisor's output cap
in step with the proxies, and an arriving proxy queues a light analysis on the
server (V-1, V-25).
- The deep checkpoint's content key includes whether each frame is read from its
original or its proxy, and analysis.json lists the frames measured from proxies;
an unreadable verdict on a frame now read from its proxy makes the report stale
(V-26).
- Thumbnails drawn from proxies are compared with the archive's thumbnail size
instead of 320 px (V-27).
- A resumed thumbnail or proxy pass sets ThumbsBuilt/ProxiesBuilt only when the
frames an earlier run did still have theirs, and the proxy pass logs frames
waiting for a client's proxy apart from unreadable ones (V-28).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The releases README told the operator to make Administrators the owner of a
data folder created by an unelevated console run. The server checks the owner
of every entry inside, and the TLS key and password file keep a protected DACL
that grants the console user, so it now gives takeown /A /R and icacls /reset
/T, and says that deleting the folder also deletes the SQLite database.
DECISIONS: the trusted-proxies row no longer endorses a Docker subnet, and the
sign-in entry cleanup row says an upgrade that renames a program moves its
entry rather than removing it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- While the open archive's archive.json cannot be read, the rows show the index's templates and proxy size and the
thumbnail size stays this PC's: the stand-in 320 no longer replaces the default of new archives. Settings ›
Archive loads through RefreshSettingsAsync, so opening it also brings the index up to a change made elsewhere.
- With no archive folder the rows edit only this PC's defaults, never the first archive of the index.
- An Archive row edited (or Reset section) while the archive at a just-chosen folder is opening waits for it and
goes to that archive (or stays a default when the folder holds none), instead of being undone when it opens.
- A portable copy whose data had to leave its OneDrive folder says so at start, with Open on the data folder.
- Opening Settings re-reads Start with Windows, which Task Manager may have changed meanwhile.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
BRIEF §6.2 forbids a SQLite index in a synchronised folder. AppEnvironment.Detect now resolves the data folder
through Core.Windows' AppPaths and OneDriveFolders: a portable copy whose .\data would lie under a OneDrive sync
root (a OneDrive folder, or a Desktop or Documents folder backed up into one) keeps its data in
%LOCALAPPDATA%\Titano and stays portable otherwise; AppEnvironment.SyncedFolder names the OneDrive folder so the
user can be told. The pure choice lives in Core/IO/DataLocation (tested on Linux); reading the OneDrive accounts
key no longer fails start-up when it cannot be read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Task Manager's Startup tab keeps the Run value and writes a StartupApproved\Run value whose odd first byte means
off. The client and the Uploader now show such an entry as off; ticking the box removes the marker so the entry
runs again, unticking removes it with the Run value, and re-pointing the command after an update leaves it as it
is. Core/IO/StartupApproval holds the rule; the registry hive is injectable for tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- The desktop's rescans run for the whole session and each one targets the archive at Settings › Archive's folder
(as the Import and Archive screens do), so an archive created on the first run, chosen in Settings, relinked or
imported with settings is rescanned from the next tick on, and Rescan every always reaches a live timer. The
folder of the start is reopened (relink, interrupted renames) as soon as its archive.json is there, so an
unplugged drive or a folder without an archive no longer stops polling for the session.
- ArchiveService.RefreshSettingsAsync reads archive.json and registers the archive again when its name, templates
or proxy size changed elsewhere (a server's admin panel, another PC); null when the file cannot be read, is newer
or belongs to another archive. Desktop rescans and the server's polling call it first, so imports, uploads and
the admin panel follow the shared settings without a restart.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The chunk endpoint raised the request body limit only before reading a chunk. A
chunk sent again after a lost answer, or to an unknown or finished upload, was left
unread above Kestrel's 1 MiB default, so Kestrel closed the connection instead of
delivering the answer. The limit is now raised before any answer, to the larger of
the upload's and the server's chunk size, and never lowered.
PUT /api/v1/archive/settings answered 500 when archive.json held a thumbnail size or
proxy quality out of range, which the admin API cannot set. It now answers 409
archive_policy with the policy to correct in details (Server.Error.ArchivePolicy).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A scan can index a sequence inside a folder the plan moves after the dry run was
shown (photos copied into a subfolder). The folder would then move with it while
its index row stayed behind. Apply now lists the index again before anything moves
and under each moving step's leases just before the move, and throws
StaleRenamePlanException naming the newcomer. A resumed journal drops such a step
while its folder has not moved yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A sequence created through POST /api/v1/sequences keeps an empty folder until its
first frame. Scans no longer adopt loose photos at the archive root under its id,
and a rebuild, the orphans report and fix, verify and the scan follow-ups no longer
treat it as the root sequence, as missing or as a sequence to analyse. All of them
use ArchiveIngest.AwaitsFolder.
On the server (ArchiveServiceOptions.ImagesAfterScan), a scan that indexes sequences
without thumbnails or proxies queues background Thumbs and Proxies passes for them,
as an upload does. A scan that finds photos added or removed clears ThumbsBuilt and
ProxiesBuilt together with AnalyzedLight.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A finished render's toast is claimed once (JobsViewModel.ClaimRenderToast):
the editor that followed the render to its end says it with its output,
also when the sequence cannot be read again (then without Show in
folder) or the editor closed meanwhile; any other render is said by the
Jobs screen. Before, the Jobs screen stayed silent whenever an editor had
the sequence open, so a failed reload or a render the editor did not
follow ended without a completion toast.
Open on a job over several sequences (toast, Windows notification, Jobs
row) shows the Library with every one of them selected, clearing a
search or filter that hides one and reading pages until each is listed,
instead of opening the first in the editor.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Picking another archive, or the shown one leaving the index, now also
clears the previous archive's tiles, charts, last scan, recent runs and
rail badge until the new one is read, so a slow probe or a failed read
never shows them under the new archive's name.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A photo the executor could not copy now carries one localized sentence
(ImportExecutor.FailureText: its own checks, a full destination, access
denied, else "The photo could not be read or copied.") on the Import
screen and in the import job's log. The photos a stopped folder or
server import never reached show the stop's cause (ImportFailures.Stop:
the server's own sentence, not answering, untrusted certificate, card
gone, destination full or not writable), and a failed index check reads
"The index could not be read or written.". A refusal without words
reads as the catalog's sentence. The system's text stays in the log.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
STATUS no longer claims every --capture screen is pixel-diffed: with no
baseline committed the step compares nothing. Next steps now starts
with accepting the first baselines from ci-screenshots. DECISIONS marks
the folder-scoped Run cleanup row as superseded by the log-based rule
and drops the branch-merge notes from that rule's row.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Moving the test into ArchiveLanguageTests turned "" into a raw,
invisible U+200F, so the line read as Assert.Contains("") and GitHub
flagged the file for hidden bidirectional text.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
RemoveLeftovers deleted every Titano or Titano Uploader Run value that
started a file the upgrade removed, including a client or Uploader the
new version only renamed or moved. The new program then found no entry
to follow and Start with Windows was silently off. Setup now points such
an entry at the program of the same component in the new payload (its
shortcut target), keeping the options and the value name, and removes
it only when the component is no longer installed. Uninstall still
removes the entries of the programs its log lists.
StartupCommand.StartsFrom, left over from the superseded folder-scoped
cleanup, is gone; StartupCommand.Retarget replaces it and its misleading
test now covers how setup moves an entry.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The compose comments, the releases README and the Unraid template told
users to list the Docker network's subnet (172.17.0.0/16 on Unraid) in
TITANO_TRUSTED_PROXIES. That range holds the gateway, from which Docker
relays IPv6 clients and the host's own connections to published ports,
and on the default bridge every other container: each could again pick
its X-Forwarded-For and X-Forwarded-Proto. The guidance now names the
proxy container's fixed address on a shared user-defined network, warns
against subnets, the gateway and the default bridge, and sends a proxy
on the Docker host to network_mode: host (loopback). The server's
parse error no longer suggests a /12 network.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A service has no console, and a refused data folder or server.json throws
before the server's log file exists, so the SCM recorded only error 1064
and the entry to check was lost. ServiceMain and the stop path now also
write the reason to the Application event log (source TitanoServer,
advapi32 ReportEventW, no message file registered). The Windows README
says where to look and that a data folder created by an unelevated
console run must be deleted or given to Administrators first.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- The server's diagnostics bundle lists the SQLite compile options next to the database line when the index is
SQLite (BRIEF §6.2), as the desktop bundle does; before, they reached only one early log line.
- A deep analysis resumed after an interrupted run decodes the frames that run found unreadable again to confirm
them; a frame read from its proxy is now skipped, so the proxy's tier is never recorded as the original's.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Machine translations with each language's glossary, every one reviewed independently; the plural forms each
language's rule selects are included. Every culture now carries every key (i18n check: 0 errors, 0 warnings).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The single-instance test ran the second launch with Task.Run and waited on it. A task waited on before it starts
may run inline on the waiting thread, which owns the instance mutex, and a mutex lets its owner in again at once,
so the "nobody exits in time" check could see the launch get its turn. The launches now run on dedicated threads.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Every workstream's decisions (dated 2026-10-12) and changelog entries, the merge choices (no decode tier from a
proxy, Reset all and the sign-in entry, card-only post-actions, restore summaries as labels), the new playback and
installer checks in the phase rows, and how to try the new Settings pages, project files and notifications.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The service steps were only checked against in-memory records and registry copies. On an elevated runner the test
registers two throwaway demand-start services through the Service Control Manager, one starting a program of the
install folder and one of another folder, and checks that setup finds only the first, prunes it, and leaves the
second registered.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The running Titano listens for later launches as soon as it holds its data folder's mutex and keeps what arrives
until the window can take it, so a double-clicked .titano during a slow start is opened instead of dropped after
3 s. When the main window closes it stops listening (the mutex stays until the services have stopped), so a later
launch gets no answer and, instead of exiting, waits up to 30 s for the closing instance to exit and starts in its
place with the file. A mutex created by a Titano running with administrator rights no longer crashes a normal
launch: it counts as another instance and the launch exits quietly.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Services, sign-in entries and shortcuts are matched against the full paths of the files the install log or the
payload lists, never against everything under the install folder: an install into a shared folder such as D:\Apps
or /DIR="C:\Program Files" no longer stops, starts or deletes other vendors' services or removes their Run values.
The elevated step now takes the names the unelevated setup found (/SERVICES=... /NAMES=...) and reads each service
again, acting only when it still starts a program of the folder's log or of its own payload (or, for Prune, a
program of the folder that is gone).
Programs are asked to close before any service is touched. On uninstall the services are deleted only once the
programs have closed, just before the files go; a failed install or uninstall starts again the services it
stopped, and after an upgrade setup starts exactly the services it stopped, whatever their start type.
The Server component installs only for all users in a folder inside Program Files. In the window a custom folder
outside it is invalid while the Server is selected, with a tooltip saying why; silent mode refuses it. An upgrade
of an installation elsewhere cannot add the Server, and one it carried is dropped with its service.
Sign-in entries use StartupCommand (the same file as on fix/shell-notify) and are removed by
SetupRegistry.DeleteStartupEntries only when the client's or the Uploader's value starts a program the install log
lists; an upgrade that drops a program removes its entry too. StartupEntries is gone.
A shortcut that starts another product's existing copy is still left alone but now logged; uninstalling the
product it starts hands it to another product of the same scope that lists it and has the same program, so the
Uploader that stays keeps its shortcuts. Another registered Titano installation whose folder or log cannot be read
now counts as using every program's data, so "Remove my data" never offers it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The sample app shows a display path as its archive folder, and Settings › Archive edits only the archive at the
archive folder (never another one of the index), so the test points the folder at the archive it changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Any failure of the informational PRAGMA compile_options probe is logged
and gives an empty list; before, anything but a SqliteException was cached
by the lazy value and rethrown by every later open.
Tests now cover the decode-tier fields end to end: the Library's
preview-only chip and its plural tooltip, server cards and frames carrying
previewOnlyCount and the preview size over the API.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- Preview sizes come from probes and unscaled decodes: a size-limited decode
picks the smallest preview that reaches its edge (CR3 PRVW 1620x1080), so
renders and deep passes record its tier without a size and the output cap
keeps the largest preview's.
- A read error is not an undecodable file: ImageDecoder.Decode throws the
IOException when a provider could not read the file and none produced
pixels. Deep passes never flag frames they could not read or that
vanished, and frames only an interrupted earlier run found undecodable
are decoded again before they are flagged.
- Renders resolve stand-ins only for frames the pipeline asks for; frames
the read-ahead meets on the way (skipped by speed-ups) are left alone.
- A GoPro GPR finds its JPEG in the index when an import or upload kept it
in a sequence of its own (same original stem and capture time), for
probes, deep analysis, renders, thumbnails, proxies and the editor;
GPRs whose JPEG is indexed later are probed again.
- Thumbnails judge an embedded preview by its own JPEG header.
- PostgreSQL writes tier fields 1000 frames per statement.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The MakerNote interval-shooting hint (Panasonic TimerRecording) is kept as
FrameFlags.Intervalometer from indexing on, so it reaches the sidecar, the
index, sync and the protocol. When most frames carry it, gaps in the
cadence are the intervalometer's deliberate pauses and the advisor picks
adaptive timing (Advisor.Reason.Timing.Intervalometer, BRIEF §10.3).
Frames both analyses find undecodable are now one finding over the union
of both lists at the higher severity: the deep pass also lists missing,
corrupt and beyond-this-host frames, which the light analysis never flags.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The restore dialog's two summary lines count sequences and photos in one sentence, which a plural key cannot
express: they read "Backup: sequences 12, photos 3,400". Reset all returns the sign-in entry, kept in the
registry rather than in the settings file, to its default (off) too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- The archive's index row is written under the same lease as archive.json, so two edits cannot leave them in
opposite orders.
- A scanner holding the restore check's scratch copy no longer turns a passed check into a failure.
- Night frames and Star fields say "Enter a number" instead of "Enter a whole number".
- Folder and server sources start with Keep and no eject whatever Settings › Import says for cards.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Texts that count with participles or adjectives still put a bare number in
front of a word that agrees with it in Italian, so one failure read "1 non
riuscite": the import result and its Done toast, the archive health line and
the thumbnail and import job log lines. They now read as labels in English and
Italian ("Copied: 40 · failed: 1"), like the scan and verify log lines; their
old translations leave the other 36 catalogs until the next translation pass.
The uploader's progress keeps the mockup's English and uses labels in Italian,
and the admin dashboard's upload total no longer agrees with the number.
`i18n check` now also fails on an English text that tallies two or more counts
with participles or adjectives. A single "{0} failed" is left alone, since the
placeholder there is as often a name or a size as a count.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- Editor: a failed save showed what the server or the archive layer would say
for the exception's type, so the editor's own sentences were lost: frames
changed on the server or settings changed elsewhere read "The server does not
answer", a read-only account or a gone sequence "Something went wrong". These
failures are now EditorException (an IOException; ProxiesOutdatedException
derives from it) and the toast shows their sentence as it is. A project
written by a newer Titano is reported with the same sentence as a corrupt one
instead of the parser's English text.
- Library: a job posted to a server whose certificate changed since sign-in
threw out of the action unreported; the pin's refusal now reaches the error
toast ("The certificate was not trusted").
- Archive: picking another archive forced it offline until it was read, so the
rebuild dialog called it offline with the previous archive's count, and
disabled commands said the archive must be online. What was read is now kept
apart from availability: until the picked archive is read, Rebuild and the
online commands wait and say so, health fixes and search results of the
previous archive are cleared, and a running search is cancelled.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- Entrypoint: an existing .titano folder or subfolder that PUID:PGID cannot
write stops the container with a one-line error instead of a restart loop;
it is never chowned.
- Compose and README: with network_mode host, publish PostgreSQL on loopback
and set TITANO_DB_HOST to 127.0.0.1; behind a proxy, list it in
TITANO_TRUSTED_PROXIES (loopback only by default). Unraid template gains a
Trusted proxies variable.
- README: what the Windows server requires of its data folder and archive.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- Windows data folder: the folder and everything in it must belong to SYSTEM,
Administrators or the server's account and hold no link, checked before the
ACL is rewritten; otherwise the server stops and names the entry. server.json
is read only after its owner is checked on the open handle. Secrets are
deleted and created anew with the private ACL, so a planted file's owner
loses its rights.
- Windows archive: a folder the server creates is changeable only by those
accounts (users may read it) and a .titano folder it creates is private; an
existing archive others can change is logged as a warning at start.
- Proxy mode: only loopback is trusted by default, since LAN clients reach the
HTTP port from private addresses; a cookie sign-in that did not come through
a trusted proxy over HTTPS answers proxy_untrusted, naming
TITANO_TRUSTED_PROXIES.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A mass rename moved the folders of nested sequences while holding only the
lease of the sequence whose folder moved, so an ingest, an editor save or a
thumbnail pass of a nested sequence kept writing in the folder as it moved: on
Windows the move failed and left the rename pending; elsewhere the writer
recreated the old folder and the nested step then failed. A moving step now
takes the sequence lease, then the thumbnail lease, of itself and of every
sequence it carries, by id, checks them all before anything moves, and
records the carried sequences' sidecars and index rows before letting go, so
they are journaled together.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Wanted proxies were matched to this PC's originals by SHA-256 only, and
frames adopted from a folder stay unhashed until verified. A desktop that
uploaded an adopted HEIC folder counted none of the proxies the server asked
for, showed the link in sync and sent none of them. The upload now keeps the
hash it computes for each original the index has none for and makes the
proxies from those too; the comparison counts a wanted frame no hash matched
against an unhashed original this PC reads by size (the server names its copy
by its own template, so not by name).
The wanted count was recomputed only when frame or render counters moved, so
proxies a render node sent left the row "Local ahead" with an upload that
sent nothing, and proxies the server started asking for later were never
counted. A link whose copy may make proxies for the server now keeps the
server's update time of the last comparison (sync_links.diff_remote_updated_at,
migration 0006) and is compared again when it moved.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Earlier versions made the server's proxies of HEIF frames from their embedded
previews. Those carry the frame's capture time, so the proxy pass kept them
and the sequence stayed marked as having all its proxies; renders and
analysis then read thumbnail-sized files and no client or render node was
asked for a real one. The proxy pass now treats a proxy whose long edge falls
short of the policy's for the frame (its own size when smaller, a tenth of
slack, frames of unknown size not judged) as stale, and the server refuses
such a proxy when one is sent (422 proxy_too_small).
PUT /api/v1/frames/{id}/proxy also took a proxy of a frame the server
decodes itself, replacing its own proxy with any JPEG of the right capture
time; it now answers 422 proxy_not_wanted for those and for frames missing or
damaged. Its follow-up jobs are queued even when the sender disconnects once
the proxy is filed.
A render node cleared the server's Undecodable flag on every frame, so a
damaged frame the server's own render skips was downloaded and failed the
node's job. Only a format the server cannot decode and the node can loses the
flag now; other undecodable originals are neither fetched nor rendered unless
read from their proxy.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Index restore (ENG-06, UI-09):
- A staged restore now runs on the files before the index is opened, so it also
recovers an index that no longer opens; Settings > Advanced > Restore works
without an open index (the current one is shown as unreadable).
- The staged copy passes PRAGMA quick_check (before migration, in a scratch copy)
ahead of any backup: a damaged copy is set aside as .failed and never renews the
daily copies. The index it replaces is kept as a VACUUM INTO copy, or, when it
cannot be read, copied whole with its WAL as titano-replaced-<stamp>.db outside the
rotation. The staged file is renamed over the index, so it is never applied twice.
- Archive > Maintenance gains Restore database..., which opens Settings > Advanced.
Archive settings (UI-03, ENG-03):
- An archive.json with an empty or overlong name (drive roots) can be edited again:
the name read is made usable as at creation.
- Proxy and thumbnail sizes outside the presets (an admin's 3840) are listed and
shown with their number.
- Changing the archive folder retargets the rows: an archive there is registered
and rescanned; with none, the rows edit only the defaults of new archives. The
rows never fall back to another archive of the index.
- Reset all and Import settings no longer rewrite the shared archive.json; Archive >
Reset section asks first, naming the archive.
- PUT /api/v1/archive/settings: 409 archive_busy on a lease conflict, 409
archive_manifest for a damaged, newer or foreign archive.json (was 500).
Parallelism (UI-02):
- Auto is the job engine's default, half the logical cores (BRIEF 14.6), with each
job's own decode and analysis threads; a value the user sets is the number of
workers and caps those threads. The memory budget is shared: each job's frame
window gets budget / workers.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The server publishes one account's card progress at most once a second
(sooner reports are taken and dropped) and at most 120 card events of an
account a minute (429 rate_limited beyond), so an upload-only account
can no longer push job and sequence events out of every client's queue.
On the client, the end of a card import waits at most two seconds for
its reports to reach the server, then cancels whatever is still on its
way, so a cancelled import or one that lost its server ends at once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Reconcile and a scan that finds a sidecar with other frames than the
index now read the sidecar and index it while holding the sequence lease,
the one an editor holds while it saves its project. The save then either
lands before the move, which moves its positions, or reads the moved
settings and merges onto them; a save made in between can no longer
write positions of the old frame order back without a version conflict.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A re-upload that repairs a missing or damaged frame is now reported as
IngestResult.Repaired and adds nothing to the uploader's used bytes; the
quota checks at start and completion skip uploads whose target sequence
already holds the content. The start shortcut and crash recovery share
one rule (UploadRules.StandIn): an intact copy stands for the upload, the
target sequence's own copy first, except when the target holds the
content only in damaged copies, which the upload's bytes then repair.
A sequence adopted at the archive root that holds frames is never given
a new folder when its root sidecar is missing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- Toast.RunsFromNotification: a click on a Windows notification restores the window and runs the toast's
action only when the toast opts in (card detected: Import; interrupted import: Resume; job finished: Open;
navigation to Import or Jobs) and it is the only notification shown since the icon appeared. Cancel, Undo
and Retry stay in the window, and the notification text no longer offers them.
- Job endings are coalesced: finished jobs of one kind wait until no job of that kind the user waits for is
queued or running, then one toast says how many finished ("12 jobs finished", action Jobs); failures of one
update share one toast. Server > Upload all no longer raises one toast and one notification per sequence.
- Every failure that is not upkeep (scan, thumbnails, proxies, light analysis) is announced, including move,
adopt and user downloads; finished moves and adoptions are announced too.
- Rail badges: Library counts the sequences with warnings (the Warnings filter), refreshed after analyses;
Archive counts the index health lines that need attention.
- Tests: the balloon click path, the --minimized restore to a saved maximised window, batch and failure
coalescing, and the new badges.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The client's HKCU Run value now counts as on only when it starts this executable, so a portable copy, a
second install or a development build no longer shows the box ticked, and starting one no longer rewrites the
entry to itself. Refresh only follows an entry whose program an update renamed in this copy's folder, and
unticking removes only this copy's entry. Portable copies keep the choice in memory and hide the Startup row,
since they leave nothing in the registry.
The uninstaller removes the client's and the Uploader's sign-in entries when they start a program of the folder
it removes. StartupCommand in Titano.Core holds the Run key, the value names and the command parsing shared by
both.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Loc.SetCulture stores the new culture before it raises CultureChanged, so the
guard's exception left the process in the other language until the offending
test's own cleanup ran, or for good, and the parallel classes asserting English
failed beside it. Both test programs now restore English first (the switch back
passes the guard) and then fail the test, naming it and the language it chose.
LanguageGuardTests runs a non-serial fixture through the runner and checks the
failure message, that English is back and that later listeners only hear of
English.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The MediaPlayer check ran on an STA thread with no exception handling: a WPF
NotSupportedException from new MediaPlayer(), an exception from Open or from an
event handler inside Dispatcher.Run ended the whole Titano.Core.Windows.Tests
process. Every outcome now goes through one Finish that closes the player and
stops the dispatcher once, and the exception is rethrown on the test thread with
its own type and stack trace.
Where Windows Media Player is missing the playback test passed with only a log
line. It is split into MediaPlayerOpensTheSoftwareH264Output, skipped with the
reason where Windows Media Player is missing or unusable, and
SourceReaderDecodesTheSoftwareH264Output, which runs on every Windows with Media
Foundation. MediaPlayerOpensTheHardwareHevcMain10Output also opens the hardware
HEVC Main10 output where a hardware encoder and an HEVC decoder exist.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A screenshot without a baseline only warned, so with no baseline committed every
screen was "new" and the CI step could not fail, and a screen added after the
first baselines could stay ungated for good. Now the baseline folder is armed as
soon as it holds a PNG: from then on a screen without its own baseline fails
(Verdict.Unbased, an error annotation). Before the first baselines the run says
plainly, in its warning and in bold in the job summary, that it compared nothing
and that no visual change can fail it.
The baselines README explains why the existing ci-screenshots pictures cannot
serve (captured before the clock was fixed: times and dates follow the capture
moment) and how the first set is accepted from the first green windows-latest
run of the reproducible capture.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Setup registered .titano with Titano.exe "%1", but nothing wrote such a file
and Titano ignored the argument (BRIEF §18).
- Core ProjectBundle: versioned UTF-8 JSON ("titano-project", format 1) with
the archive (id, name, path), the sequence (id, folder, title, frame
count) and the project settings as project.json holds them; newer or
foreign files are refused, frame positions apply only to the frames they
were saved for, saving never replaces an earlier file.
- Data ProjectBundles: the file for a sequence of the index, and the sequence
a file names (by id, else the same folder of the same archive).
- Editor: "Save project file" on the Export tab (sequences of this PC's
archives); opening a file shows its sequence and applies the settings as
one undo step with an Undo toast, or says in one sentence why not (not in
the library, unreadable, a job runs).
- Titano.exe opens its file argument once the index is open and runs once
per data folder: a later launch hands its file over a current-user named
pipe, lets the running window come to the front and exits.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- Windows services whose program lies in the install folder (Titano.Server
--install-service) are found in the registry and, through the Service
Control Manager, stopped before files are replaced (they join the
"programs running" prompt), started again after an upgrade that stopped
them, deleted when an upgrade drops their program and on uninstall. An
unelevated setup runs that step in an elevated copy (/SERVICES=...).
- The Server component installs for all users only (manifest allUsersOnly):
choosing it moves a new install to Program Files and disables "Just me"
with the reason; a per-user install cannot take it.
- Uninstall removes every HKCU Run value that starts a program in the folder
(the Uploader's "Start with Windows").
- A shortcut that already starts an existing program elsewhere is neither
retargeted nor logged, and a logged shortcut is deleted only while it
still starts a program of this folder, so the two products no longer
remove each other's "Titano Uploader" shortcuts.
- "Remove my data" lists and deletes only the data of this product's
programs that no other installed Titano product uses (the desktop app and
the Uploader share %LOCALAPPDATA%\Titano and its logs folder).
- The .titano file type is named in setup's language (Setup.FileType.*).
Setup tests run in Titano.App.Tests on Windows CI.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The fallback tested the builder's length, which the title line already made
non-zero, so release.yml (always --title) published notes that were only a
heading when a release held chore, ci, test or refactor commits alone. The
fallback now depends on whether a section or breaking entry was written.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- Archive: picking another archive while the screen was reading kept the old
archive's tree and figures while Scan, Verify and the set actions targeted the
new one, and reloads asked for during a read were dropped. A load requested
during a read now runs once more after it, a read whose archive is no longer
picked shows nothing, and a new pick clears the tree and the online state
until the picked archive is read.
- Import: a job that ended before the screen recorded its id (a destination
gone offline) left the screen copying forever; its state is read back once.
A replaced import result is disposed, so it no longer stays alive through the
language event with its whole plan.
- Library: the timeline's month readouts are made again on a language switch.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Findings, advisor reasons and a few UI texts put a count before a plural noun
without plural forms, so English and Italian read "1 frames are black". Each
such key is now a plural stem (.one/.other) whose count is its first argument;
findings and reasons, which travel as data, resolve through the new Loc.Text,
which picks the form from that argument (a fraction takes "other"), so stored
deep reports and project reasons keep their keys and arguments. Texts with two
or more counts (job log lines, "frames with a capture time: 12 of 300") read as
labels, which agree with no number. A single exposure change gets its own
finding instead of "1 exposure changes, typically 0 frames apart".
The converted and reworded keys leave the 36 other catalogs until they are
translated again; the dead plain Shell.JobBar.Ready goes too. `i18n check` now
fails on an English text that counts a noun without plural forms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Error toasts showed the exception's own message, often technical English from
.NET or the OS, under a translated title. The editor's failed save, a job posted
to the server from the Library, and the card check or delete after an import now
say why in one catalog sentence: what the server answered or that it does not
answer, that the card left, is locked or a file could not be read, and the
archive causes of ArchiveFailures for a local save. The details stay in the log.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Settings (UI-01, UI-02, UI-03, UI-09, RF-4):
- New Import, Analysis and Formats pages, in the order of Settings.html. Import holds
the defaults of new cards (session rules, subfolders, verification, duplicates,
after-import actions); Analysis the deep-analysis width cap, night and star-field
thresholds and deep analysis after import or scan; Formats the decode tier of every
format on this PC with a notice and Store install link per missing codec extension.
- Archive gains thumbnail size and rescan interval; its templates, proxy size and
thumbnail size are the open archive's own (archive.json) and new archives' defaults.
- Rendering quality, parallelism and background priority now drive the jobs: quality
maps to preparation size, read-ahead and default flow quality; parallelism sets the
engine's workers live and caps per-job threads; background priority raises the UI
thread while Titano is in front.
- Advanced gains Restore (daily backups, dry-run summary, destructive confirmation,
staged for the next start) and Reset all.
- The naming preview's sample title is a catalog string.
Engine (ENG-03, ENG-04, ENG-06):
- ArchiveService.GetSettingsAsync/UpdateSettingsAsync rewrite archive.json atomically
under a lease, then the index row; the server endpoint and the desktop use them, and
the desktop creates archives with Settings › Archive. ThumbnailPolicy in archive.json
sizes the thumbnails.
- DatabaseBackups: daily VACUUM INTO copies kept x3 on the desktop and on a server on
SQLite (DatabaseBackupService), staged restore checked and preceded by a fresh copy.
- Maintenance vacuums the SQLite index when 10 % or more is free (dry run reports it).
- JobEngine.SetWorkers; ProcessingOptions.Tune applies host tuning per job.
Archive setup toasts say why in a localized sentence (RF-1).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
- Screenshots: a capture that exits with an error or writes nothing fails the job
(the pictures it wrote are still published); the new "Screenshot diff" step
compares them with tests/visual/baselines, fails beyond the tolerance, warns
once for screens without a baseline and uploads the diff images.
- installer (windows-latest): publishes App, Uploader and Setup (self-contained
single file, no ReadyToRun), packs Titano 1.0.0 (with an extra legacy.txt),
Titano 1.0.1 and the Uploader installer, and runs
.github/scripts/installer-test.ps1: argument and not-installed exit codes,
silent install into a folder with a space, Add/Remove Programs, shortcuts and
the .titano file type, the installed programs drawing their screens, upgrade in
place removing legacy.txt, uninstall removing itself, user data kept, the
Uploader installer; the setup logs are printed whatever happens. The script
now resolves its paths from the PowerShell location.
- PostgreSQL 16/17: postgres-ci-setup.sh prepares the service container and the
job exports the MD5, cleartext, trust, TLS and root-certificate variables.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
BRIEF §6.3 requires every authentication method and TLS mode against the
PostgreSQL service containers, but the MD5, cleartext and trust tests always
skipped in CI (no roles) and the TLS tests skipped at run time because the
official images start with TLS off. .github/scripts/postgres-ci-setup.sh prepares
a server container: a role per method with its pg_hba line ahead of the image's
SCRAM line (the MD5 password stored as MD5), and TLS with a self-signed
certificate for localhost and 127.0.0.1, enabled with ALTER SYSTEM and a reload
whose rejected lines fail the script. With TITANO_TEST_PG_TLS=1 a server that
refuses TLS fails the TLS tests instead of skipping them, and a new test checks
that verify-full accepts the server whose certificate chains to
TITANO_TEST_PG_ROOT_CERT.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
BRIEF §20.2 lists a MediaPlayer playback check after the Media Foundation encode
smoke test; the encoder tests stopped at the MP4 inspector. The new test encodes
two seconds of software H.264, opens the file with WPF's MediaPlayer and checks
its natural size and duration (failing on MediaFailed or a timeout), then
decodes the first frame with a Media Foundation source reader and checks its
size, the file's duration and the picture that was encoded. Without Windows Media
Player (Windows Server, "N" editions) the source reader decode is the check.
Core.Windows.Tests references WPF through its framework reference only, so its
implicit usings stay as they are.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
BRIEF §20.2 asks for visual regression of the --capture screenshots with a
pixel-diff tolerance; CI only published them. `Titano.Tools screenshots diff
<dir>` compares every PNG with tests/visual/baselines: a pixel differs when a
channel moves by more than --tolerance (8 of 255), and a screen fails when more
than --max-diff of its pixels differ (0.05 %), when its size changed or when a
baseline lost its screen; a screen without a baseline only warns. It writes a
diff image per screen with differing pixels (the baseline faded, the differences
magenta) and a Markdown summary, also to the GitHub job summary.
`screenshots accept <dir>` copies screenshots over the baselines. The baselines
folder starts with its README; the pictures come later from ci-screenshots.
The captures were not reproducible: the sample's jobs, renders, transfers,
cards, trash and remote sequences were dated from the clock, the Library
timeline and the Archive heatmap laid out around today, the Uploader captured
in the computer's language and an indeterminate bar animated. The sample now
dates everything from SampleData.Now, --capture freezes the new AppClock there,
formats are English, and the new Motion attached property (off in the captured
windows) keeps the indeterminate sweep still in the middle.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Four App tests switched the process-wide UI language inside parallel classes, so
another class's English assertion, or its constructor's reset to English, could
land in the middle of them; they now live in the serial ArchiveLanguageTests and
ImportLanguageTests. The App and Core test programs fail a test that leaves
English outside a serial class at the switch, through the new
TestContext.Current and TestContext.IsSerial.
The FrameWindow cancellation test counted unobserved task exceptions while other
classes ran beside it; it moves to the serial FrameWindowIsolationTests, and both
unobserved-exception tests drain the finalizer queue before they start counting.
The runner unwrapped TargetInvocationException only around the test method: a
skip or an assertion thrown by a test class constructor or field initializer was
reported as a failure reading "Exception has been thrown by the target of an
invocation". Constructors are unwrapped the same way now. TestRunner.Discover(Type)
and ExecuteAsync are public, so Tools.Tests checks the runner with nested fixtures.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The server decodes with managed decoders only, so HEIF originals had no
proxies, thumbnails from their small previews and no pixels for analysis or
managed renders, and nothing could send it proxies (ENG-05). Manifests now
carry the archive's proxy spec and, per frame, its flags and whether the
server wants its proxy. PUT /api/v1/frames/{id}/proxy (Edit or RenderNode)
stores a JPEG of at most 64 MiB that decodes and whose Exif capture time is
the frame's, then queues thumbnails and the proxy pass. An upload from the
desktop makes and sends the wanted proxies after the originals and counts
them as local changes; a render node sends those it can make from the
originals it fetched once the job is done. The server never makes such a
proxy from a preview, drops one left by a renumbered frame, takes thumbnails
from a proxy, and analyses and renders those frames from their proxies.
Render nodes fetched every frame of the manifest: frames missing, corrupt or
excluded on the server failed the download, and originals not hashed yet
failed the SHA-256 check (ENG-07). Nodes now skip those frames, fetch unhashed
originals into the job folder with a size check, judge decodability with their
own decoders, and use the proxy of a frame they cannot decode.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The server indexed only what arrived through its own uploads, so photos
already in an adopted folder and sequences another writer added stayed
unknown until an administrator asked for a rescan. ArchivePolling, registered
by the archive module, queues an incremental scan of the online archive at
start and every ten minutes at background priority, unless a scan or rebuild
of it is waiting, running or paused (ENG-13).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
BRIEF §8.2 tiers 2 and 3 had no writer: PreviewOnly and Undecodable were never set,
so the advisor could not cap the output at the preview size and one unreadable frame
failed a whole render.
- ImageDecoder.Probe names a file's tier without decoding it (JPEG, PNG and TIFF
without opening the file); the light analysis probes frames whose tier is unknown,
and the deep analysis, renders from originals and thumbnail/proxy passes record
what decoding actually found (FrameTiers). Sidecar frames gain "tier", "previewW"
and "previewH"; the index gains frames.preview_width/height and
sequences.preview_only_count (migration 0005, SQLite and PostgreSQL).
- SequenceContext takes the preview size from preview-only frames, so the advisor's
output cap applies; the light analysis reports format tiers (preview-only frames as
information, undecodable ones as a warning under the deep analysis' key, listed
once); a sequence that gains undecodable frames warns at once.
- Only a file in a format the host decodes natively is flagged undecodable; files
beyond the host's decoders keep their tier without the flag, so render nodes and
proxies can still supply them.
- A frame that stops decoding during a render is replaced by its nearest planned
neighbour and logged instead of failing the job; the render fails only when no
planned frame decodes.
- The Library card shows a preview-only chip with the count (local and server
sequences: SequenceDto.PreviewOnlyCount, FrameDto preview size).
- GoPro GPR decodes through the JPEG the camera writes beside it when that is larger
than the embedded preview (BRIEF §8.1).
- Thumbnails use a stored embedded preview only when it reaches 320 px; a smaller one
sends the frame through the host's decoders instead (BRIEF §8.5).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A download whose .part already held the whole original (the job stopped while
hashing it) or more asked the server to resume past the end, got 416 and
failed on every retry; such a part is now hashed again, and one the server
cannot resume starts over (CD-1).
Auto-sync pulled the server's settings with the copy of the sequence read when
the comparison began, so an edit saved on this PC meanwhile was overwritten
unseen. The payload is now applied in one step under the sequence lease and
only while the local settings version is the one read; otherwise nothing is
written and the change is settled as a conflict (CD-3).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
BRIEF §6.2 asks for the probe at startup. The options are read once per process
from an in-memory connection, logged with the library version and added to the
desktop diagnostics bundle; nothing depends on them (no FTS or JSON use).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
BRIEF §9 lists exposure mode, intervalometer hints and sRaw flags among the MakerNote
fields; no vendor parser read them. The vendor exposure mode (Canon, Olympus,
Fujifilm, Panasonic, Sony) now fills ExposureProgram when Exif has none, Panasonic's
TimerRecording becomes FrameMetadata.IntervalShooting, and Canon's SRAWQuality and
Nikon's ImageSizeRAW / small NEF compression become FrameMetadata.RawSize. Nikon keeps
its interval-shooting state in the encrypted ShotInfo, which stays undecoded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
Both 0003_server_uploads.sql files still pointed at the 0005 names they had before
the migrations were renumbered without gaps. Comment-only: migrations are keyed by
version, not by content.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A mass rename or move of a sequence whose folder holds other sequences moved
their files without updating their index rows, and a target inside another
sequence's folder could nest one sequence in another. The plan now carries the
nested sequences (listed in the dry run, moved with the folder that holds them,
inner folders leaving first) and refuses a target inside a sequence folder with
a localized NestedRenameException (CD-4). The Archive screen counts only the
selected sequences in its summary.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
CompactPacksAsync rewrote thumbs.pack under the sequence lease only, while
thumbnail jobs append to the pack under the thumbs-<sequence id> lease, so a
compaction racing a thumbnail pass could drop the entries just written.
Compaction now takes the thumbnail lease too (CD-5).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
BRIEF §14.3 lists cardEvent among the WebSocket events, but the server had
none. A client importing a card into the server now reports its start,
progress (at most once a second, a report still waiting replaced by the
next) and end to the new POST /api/v1/uploads/cards (Upload permission;
import id, card name, state, done, total, failed), and the server publishes
a CardEvent naming the account and client to every connected client; the
protocol client parses it and offers ReportCardAsync. A report that fails is
logged and never stops the import; a server without the endpoint gets no
more reports (SRS-10).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The project names frames by position (exclusions, hero, virtual-camera
keyframes), and an upload, import, scan or sync download of earlier photos
merges them by capture time and renumbers the frames after them; nothing
moved the project, so a render excluded the wrong photo and the project's
hero no longer matched the Library's.
project.json keeps its format: when the index takes a sidecar in which frames
were inserted before indexed ones, the index's copy of the project has its
positions moved from the indexed order to the new one in the same transaction
as the frame rows, and project.json mirrors it unless another writer changed
it meanwhile. The JSON is edited in place, so members a newer Titano wrote
survive; only an insertion is followed (if an indexed frame is gone, the
positions stay). The settings version stays, so a synced copy receiving the
same frames makes the same move (CD-2).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A sequence created with POST /api/v1/sequences had no folder, and the ingest
looked for its sidecar at the archive root: every upload into it failed with
503 commit_failed, so the Uploader's "New sequence" target never worked. Its
first frame now claims a folder by the rules of a sequence created by an
upload (folder template with its title, unique variant, sidecar with the
row's id, title, tags, rating, notes and project), and the row takes the
folder in one transaction (DOCKER-1).
A frame flagged missing or corrupt, or whose file is gone or has another
size, no longer stands for an upload: the start short-circuit and crash
recovery match only intact frames still on disk, and the ingest repairs such
a frame with the verified copy (same id and position, flags cleared, a damaged
file kept beside it as name.damaged-<time>) instead of calling it a
duplicate. The Uploader deletes a source only after the server received and
verified its bytes in that upload, never on "already on the server" (SRV-1).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
POST /sequences/{id}/renders and its complete call accepted the Upload
permission, so an upload-only account could file render outputs, which count
against no quota, into any sequence. Client renders are sent by the sync of
an editor's copy: both calls now require Edit (SRV-4).
The chunk endpoint capped the body at the current chunk size, while an
upload keeps the size it started with; after an administrator lowered the
setting, every chunk of an upload in flight got 413 until the upload expired.
The service now raises the limit to that upload's own chunk size before it
reads the body (SRV-5).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
The render upload refusal render_exists had no Server.Error.RenderExists
resource, so clients showed the raw code. ApiErrors builds that key at run
time, which the scan of used keys cannot see: i18n check now maps every
snake_case code literal the server passes to ApiErrors, Outcome.Fail and its
Fail and TooMany helpers to its Server.Error.* key and fails when one is
missing (SRV-6).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
UI-05: the main window's TaskbarItemInfo mirrors the job bar (progress, sliding while
the total is unknown) and turns red while a failure waits on the Jobs screen.
UI-06: the Uploader's tray icon moves to Titano.Ui (Interop/TrayIcon, with a balloon
click callback and error balloons) and the Uploader keeps using it. While the main
window is minimised, hidden or behind another program, every toast (card inserted,
auto-import, job ended, server messages) also becomes a Windows notification; clicking
it restores the window and runs the toast's action. The icon exists only while such a
notification can be clicked. Settings › General gains Notifications (on by default)
and Startup: Start with Windows writes the HKCU Run value "Titano" with --minimized,
and the window then waits minimised on the taskbar.
UI-07: Ctrl+1…6 and Ctrl+, navigate from anywhere in the window, as the rail tooltips
say (not behind the first-run dialog). The editor's 1:1 view moves to Ctrl+Shift+1;
its tooltip key is renamed so no translation keeps the old shortcut.
UI-08: the Jobs screen raises one toast when a long operation ends (render, deep
analysis, verify, rebuild, maintenance, upload: Open; failures: Retry, or Import for a
failed import), whichever screen started it. Background work and scans end silently,
and a finished render of the sequence open in the editor is left to the editor's own
toast with its output file.
UI-10: the Jobs badge adds failures not seen on the Jobs screen yet; the Server badge
counts settings conflicts waiting for a choice.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
RF-3: the anonymous usage counts checkbox (Welcome and Settings › General) promised a
daily ping that does not exist; BRIEF §21 puts that consent inside monetization, which
ships disabled until the human flips TitanoAdsEnabled=true. Directory.Build.props now
defines the property (default false) and stamps it into every assembly as metadata;
ProductFeatures.AdsEnabled reads it, and both checkboxes and the Settings search
keyword follow it. The unused Shell.Welcome.Consent sentence is removed from every
culture.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
DESK-1: SyncHost's dispose, session switch and stop now continue on the thread pool
(ConfigureAwait(false)) and keep the running agent in one reference. A signed-in user
whose archive is not polled no longer waits 10 s at every exit, after which the job
engine and the index were never closed.
DESK-7: ShellViewModel.Dispose disposes the Sequence section first, so the open
editor stops its runners and closes its source before the engine and the index go.
DESK-8: Cancel stays available while the server answers a sign-in. It now cancels
the request; an answer that still arrives changes nothing (the connection flow
ignores answers for a client it dropped), and a sign-in that completed just before
the cancel was handled is signed out again. The form no longer sets a result after
it closed, and an unexpected failure shows inline instead of escaping async void.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
A bind source Docker creates on demand (the compose default, a new
TITANO_ARCHIVE, the image's anonymous volume) is root-owned, so the server,
running as PUID:PGID, crash-looped with a stack trace (DOCKER-2). The
entrypoint now gives an archive mount root that user cannot write to
PUID:PGID when it is empty or owned by root (the folder only, never its
contents), and otherwise exits with a one-line error; it creates and owns
TITANO_DATA_PATH instead of a hard-coded /data (DEPLOY-1), and --help and
--healthcheck touch neither folder. The runtime image empties
ASPNETCORE_HTTP_PORTS, so Kestrel no longer warns "Overriding address(es)"
at every start (DOCKER-6).
Unraid template: the database host defaults to empty and required, since
container names do not resolve on the default bridge (SRS-08, DOCKER-3); a
Public HTTPS port variable goes with a remapped HTTPS port (DOCKER-4); the
time zone description no longer claims logs use it (DOCKER-8). Compose,
.env.example and the public README document the archive folder's owner,
public ports, trusted proxies, that LAN discovery needs host networking
(DOCKER-5), the Unraid database host, and that the Windows service
registers only from Program Files (SRS-06).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
In TLS mode proxy the server accepted X-Forwarded-For/Proto/Host from any
peer, so a client reaching the HTTP port directly could pick a new address
per sign-in (defeating the 5/min/IP limit), forge audit addresses and claim
HTTPS; the admin never warned (SRS-03, SRV-3). Forwarded headers now count
only from TITANO_TRUSTED_PROXIES (addresses or CIDR networks; default
loopback, 10/8, 172.16/12, 192.168/16, fc00::/7), and the Dashboard and
Settings show a "Not via proxy" chip with a tooltip when a page did not
come through a trusted proxy over HTTPS (ProxyWarning in DashboardDto and
ServerSettingsDto).
TITANO_PUBLIC_HTTP_PORT and TITANO_PUBLIC_HTTPS_PORT name the ports clients
reach when a container's ports are published as other host ports: the
admin's HTTP-to-HTTPS redirect and the LAN beacon use them instead of the
bound container ports (DOCKER-4, DOCKER-5). --help lists TITANO_LOG_LEVEL,
the new variables and the container's PUID, PGID and TZ (DOCKER-7); a test
keeps it in step with the variables the server reads.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC
On Windows the data folder sits in %ProgramData%, whose ACL let every local
user read the database, the password-less TLS key and the temporary admin
password, and create a server.json the LocalSystem service would load
(SRS-02, SRV-2). The folder now gets a protected DACL (SYSTEM,
Administrators, the server's account) at every start when it names anyone
else, which also replaces what the files inside inherited; server.pfx and
initial-admin-password.txt are written with their own protected DACL, and
with mode 0600 elsewhere. A folder whose ACL cannot be changed stops the
start with a clear message.
--install-service registered a LocalSystem service from any folder, so a
per-user install or an unpacked zip handed SYSTEM to that user (SRS-06). It
now refuses, with a localized one-line reason, unless the program folder is
inside Program Files and no account besides SYSTEM, Administrators,
TrustedInstaller or the registering administrator can change it or any file
below it (under dotnet, the host's folder too).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iLMBNwBE2GnphAsu4cjaC