The four Rebuild of Evangelion films in one folder were one series:
`- 1.11` read as episode 1, and every file took the folder's name, so
the whole folder matched a single film. A dash followed by a decimal is
now part of the film's name, and a folder whose files are all
unnumbered, outside any grouping folder and each named differently
gives each file its own title.
AniList needs every query word in a title, and a film's version number
is not in it, so a search that finds nothing is retried without it.
All four films now match their own entries.
A crawl now keeps .srt/.ass/.ssa/.vtt files and pairs each with a video by
name: the video's name followed by a dot or nothing, the same inside a
Subs folder, the Subs/<video>/ layout, and failing those the only video in
the folder. The words past the video's name give the language and the
picker's title. They go in a new sidecars table (schema v12), rebuilt by
every crawl.
The desktop opens them while mpv opens the video and adds them with
sub-add through the stream registry once the file has loaded. One in the
preferred language is turned on when subtitles are on and the file has
none of its own playing; the rest wait in the subtitle menu.
No client crawled a share unless someone pressed Refresh, so a new
episode stayed out of the library until then, and nothing said how stale
the library was.
Catalog schema v11 adds share_crawls: when each share's last crawl
finished and was attempted, how many files it stored and how many were
new, and why the last attempt failed. A failure keeps the last good
crawl's numbers.
Every client now crawls, quietly, any share not looked at for six hours:
the desktop at launch and every half hour while open, the phones at
launch and on their existing catch-up timer. A quiet crawl says only
news ("anime: 3 new files") and leaves failures on the share's row
rather than in a toast.
Share rows on all three show "120 files - crawled 2 h ago", or the last
failure until a crawl succeeds.
Metadata was only fetched after adding a share (mobile) or from the
Settings buttons (desktop), so a show or episode that turned up in an
existing share stayed bare. Every crawl is now followed by a match run
when enrichment is on; `plan` makes no requests for a library that is
already matched. A stored episode guide is fetched again when a file
newer than it has no named entry, and daily for two weeks while an
aired episode is still unnamed.
Removing a share now carries its watch state to any other share holding
the same file, `pstr remove` drops the share's catalog rows, and the app
forgets catalog rows of shares no longer configured — an older `pstr
remove` left them, and every episode showed twice.
iOS, for a slow iPad:
- swipe back on a settings page uses its own edge strip; on the whole
page it also pressed the row under the finger and opened the metadata
sheet
- "Change match" from the tile menu opens once the menu has dismissed,
not after a guessed 350 ms that a slow device missed
- the activity banner tracks each running line by itself (also Android),
so overlapping work no longer leaves it spinning forever
- buttons, tiles and rows draw their press; marking watched shows at
once; remove/repair share, sign out and deletions show the banner;
the player says "Opening…" while the stream opens
- posters are read into a reserved buffer instead of a byte at a time
- Desktop shows Proton's CAPTCHA instead of failing the sign-in: a wry webview
in a child process of the app (`pstr-app::verify`), confined to
https://*.proton.me, hands the token back and the sign-in is retried. Linux
packages and CI gain WebKitGTK 4.1; the Windows zip and MSI ship
WebView2Loader.dll, without which a -gnu build does not start.
- The desktop keyring splits a secret too long for one entry across several,
so a Proton session with many key salts survives Windows' 1280-unit cap.
Short secrets stay one entry, so every existing secret still reads.
- `WatchSync` runs against a small `HistoryDrive` trait and is tested end to
end against an in-memory Drive: two devices, newer wins, no re-upload of an
unchanged history, a missing file written again, an unreadable own file
never overwritten, a newer-format file left alone.
- An installation's file unwritten for 90 days is retired by the next device
to sync, after its entries are merged and carried in that device's file.
- Removing a share keeps its watch state: sync would bring it back from the
other devices anyway, and re-adding the share picks up where it left off.
- Watch history syncs once more on desktop exit (bounded to 3 s) and, on
Android, through a WorkManager job enqueued when the app leaves the screen.
- The second-factor prompts say security keys are not supported.
A Proton account is now optional alongside public links. Signing in (password,
then a second-factor code and a mailbox password where the account has them;
a CAPTCHA gate is solved in the Android app) stores one session blob in the OS
credential store under `proton-account`, writing back every refreshed token
since refresh tokens are single-use. With an account, folders of the viewer's
own Drive — My Files, other devices, shared with me — can be added as shares
through the same `ShareClient` surface as a public link.
Watch history syncs through a `proton-stream` device in that Drive: one
`watch-history/<installation>.json` per installation, so no two devices ever
write the same file. Each sync reads the others and merges per episode, newer
`updated_at` wins, with a deterministic tie-break. It runs at launch, after
sign-in, whenever the player closes and every five minutes. A session that can
no longer refresh signs the app out instead of failing on every launch.
The CLI gains `login`, `logout`, `account`, `browse` and `sync`.
Also in this change:
- Android offline downloads no longer look stuck (B64): two transfer at once,
through a copy stream with no read-ahead, with a smoothed transfer rate.
- Choosing a metadata match by hand now enriches the title too, instead of
keeping the old match's episodes and artwork.
The catalog schema is unchanged.
The AniList query now keeps the start date, season, banner, popularity,
status, next airing episode, tags, main studios, director and story
relations, as TitleDetails in one JSON column (schema v10). Matches
stored without them are fetched again by id, never searched, so
hand-picked matches stay; refreshing the same entry keeps its episodes
and fanart. Enrichment records the entries it walked. node_first_seen
remembers when each file appeared, across recrawls.
pstr_core::franchise groups titles that share a provider id. The grid
folds each franchise into its lead with a +N badge, and the title page
lists the whole franchise in release order. pstr_core::shelves picks
rows by airing status, director, studio and genre. A row needs three
franchises, so one franchise's titles never make a row on their own.
Desktop:
- Covers are the default grid (Settings > Appearance > Library tiles).
- The hero and title band use fanart or banner decoded at 1920 px.
- New sorts: recently added, newest, highest rated, most popular.
- Titles can show the share's, English or romaji name; search matches
all of them.
Android carries TitleDetails through a hand-picked match; its screens
are unchanged.
AniList's per-season lookup searched for "<title> 2nd Season" and the
scorer picked the base entry again, so Attack on Titan seasons 2-4 and
Jigokuraku season 2 were captioned with season one's names. Episode
names came from streamingEpisodes, which is partial and numbered however
a streaming site counted. Bleach S17 and Mushishi S02/S03 had none.
AniList now walks SEQUEL relations from the match and takes each
entry's episodes from ani.zip, which carries TVDB's season, number and
absolute number. An episode row keeps all three plus its place in the
AniList entry. EpisodeGuide::get matches whichever numbering a file
uses and never answers from another season.
Episodes and fanart are stored as one enrichment. The episode list is
replaced even when empty, and the enrichment is recorded as asked, so a
film is not re-asked every run. Storing a different match drops the old
entry's episode rows. AniList's 4.75:1 bannerImage is no longer stored
as the 16:9 backdrop; ani.zip's 1920x1080 fanart is used instead.
Schema v9 adds the columns and deletes the AniList episode rows and
banners. Matches are kept. Desktop, Android and the CLI now share
service::plan and MetadataService::work. The settings text names
ani.zip. Audit and roadmap in docs/METADATA.md; bug entry B61.
- ImageCache drops the picture asked for longest ago past 256 MiB and
uploads at most six textures a frame
- posters and thumbnails larger than a card are stored scaled down, so
a cached grid no longer decodes and resizes full backdrops every launch
- a saved watch state patches the library in place instead of rebuilding
it from the catalog; the offline index reuses the library just built
- download byte progress sends one row, not a sorted copy of every row
- the CJK fallback font is found and read off the UI thread
- the watch-state upsert uses a cached statement
A track picked on one episode now carries over to the next by its title as
well as its language, so e.g. "Signs & Songs" vs "Full" subtitles in the same
language stick. Catalog migration V8 adds the title columns; the Android
bridge keeps a stored title only while its language is unchanged.
nvrl