Repository navigation
Show chapter names, and say which chapter Continue opens - #84
Merged
Merged
Conversation
A downloaded chapter's file is named from its number alone, so the scanner's title for it was the number said twice. Sources do give real names, and the downloader already wrote them into each CBZ's ComicInfo -- nothing read them back. setBookMeta now stores the landed copy's name, filtered so a source that also just says "Chapter 12" does not replace one restatement with another, and rows show it beside the number. The series page also names the chapter the Continue button opens, with the page position when one is part-read. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A chapter downloaded before the downloader learned to record the source's name is stuck titled after its own file, which is named from the number alone. The listing written on each check is the same data for the same chapters and is already in hand, so replaceListing now fills in any book title that is still a bare restatement of its number. Only from a listing title that is not itself one: plenty of sources genuinely call every chapter "Chapter 12", and swapping one restatement for another is churn. A title set by hand never matches the pattern, so an override is safe. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The sweep is how most chapters arrive, and its Landed record carried the group and the source but not the chapter's name, so a chapter fetched overnight stayed "Chapter 12" until the next check's listing heal. It now passes the name to setBookMeta like the add and download paths do. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
persistScan's upsert re-derived every book title from its filename, and it runs on every add, every sweep and every manual scan -- so each time anything anywhere was fetched, every name in the library went back to "Chapter 12" until that series' next check healed it. The scan now keeps the row's title when the filename says nothing but the number and the row holds a real name; a file that carries a name of its own still wins. Tests: the name filter, the landed stamp, the listing heal and the rescan. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…ptions Library ratings could not express "keep ecchi off the shelf" without moving series between libraries to get a display outcome. Two admin settings say it directly -- genres to treat as adult, sources to treat as adult beyond what their extension declares -- and a per-series adult_exempt keeps one wrongly tagged classic visible without turning the whole filter off. Still purely a surfacing filter: nothing here changes who may open a series. browsable() cannot bind parameters, so the configured names are held to a strict character shape (letters, digits, space, ' - + / & ( ) . ! :) on the way in, again at the interpolation, and quoted with ' doubled on the way out. GET /api/series/:id returns overrides.adultExempt so an editor can seed it, and the settings PATCH and series meta PUT are described in openapi.yaml. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Admin -> Settings gets an "18+ filter" section after Notifications: one chip per genre the library holds and one per source, each saving on its own. Both lists are fetched with the reveal on, because with "Show 18+" off a ticked genre or source is exactly what the listings stop returning -- a picker built from them would lose the chip the moment it was lit. The series "Edit details" dialog gets the matching "Always show" checkbox. Strings in all eight locales, and a note in USAGE.md. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The keydown handler had no ArrowLeft/ArrowRight cases at all, so in paged mode arrows fell through to the browser's native scrolling of the snap-x track: the first press nudged it a few pixels and the snap pulled it back, so it took two presses to land on the next page. Bind them (plus PageUp/ PageDown/Space) to the same scrollBy the tap zones use, with preventDefault so the native scroll stops competing. A focused form control (the page slider, the settings sheet) keeps its own arrow keys and Space in both modes, so it can still be nudged by keyboard. USAGE.md lists the reader's keys. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
With a keyboard, pressing a letter or digit while no text field or dialog has focus opens the command palette with that character already typed, so searching the library is just "start typing". Ctrl/Cmd+K and "/" still open it empty. Modifier chords, punctuation, IME composition, claimed keys and open aria-modal dialogs are left alone; the reader is excluded as before. The palette input now focuses in a layout effect so the second keystroke isn't lost to <body>. The palette's action list, including its Desktop filtering (DESKTOP_HIDDEN.paletteKeys), is unchanged. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
AdultToggle renders on its own only when the account holds an 18+ library, so
an install whose only adult content was a genre or source on the 18+ filter
hid those series with no Show 18+ button on Library or Home. GET
/api/adult-filter answers { configured } -- only the flag, never the lists,
and false for an account capped below 18 -- and Library and Home pass it to
the existing toggle as alsoWhen, the way Discover passes its hidden-source
count. A settings save invalidates it.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A term submitted with a source chosen now asks only that source (&source= on search-all) instead of fanning out to every one and answering with everything, and the picker stays on screen so the filter is visible and clearable rather than silently narrowing the results. `source` only narrows the route's existing surfaceable() set, so an adult source without adult=1, one above the account's age limit, or an unknown id asks nobody and answers the empty shape. The search query's key now includes the chosen source; discoverSearch.test.ts and discoverAdult.test.ts pin the new key and the picker being mounted in both modes, and searchAll.int.test.ts covers the narrowed fan-out. Documented in openapi.yaml, docs/api.md and USAGE.md. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Upstream's reader already knows a series' readingDirection (the `rtl` state, from the series or the downloaded record) but used it only to flip the order of a double spread; the paged track always ran left to right. New reader pref `pagedDirection` (series | ltr | rtl, default series). With `series`, a RIGHT_TO_LEFT series lays paged mode's track out right-to-left: the next page is to the left (swipe right, tap the left edge, ←), and spreads put their first page on the right. Every other series reads exactly as before -- including every series on the built-in library, which reports WEBTOON. `ltr`/`rtl` override it for all series. Toggle in the reader sheet and Profile → Reading. The track now always states its `dir`. Under the Arabic UI it used to inherit dir="rtl" from <html>, and the positive-scrollLeft maths left the page counter on 1 and clamped a jump or resume to the first page. On a right-to-left track the footer row takes dir="rtl": the page slider fills and drags from the right, matching the page track, and previous/next chapter swap sides with their chevrons swapped to keep pointing outwards. The page counter stays LTR. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A source is the closest thing the app has to a format: a webtoon source wants continuous vertical scroll, a manga source wants paged right-to-left. One global default was wrong for half the library and the alternative was fixing every series by hand. Reader prefs now resolve global < source < series, and the settings sheet can pin the current look to the chapter's source (or forget it again). Stored and synced beside the per-series memory, under a key prefix that cannot be picked up by the per-series collector. Tests pin the prefix, the forget path and the pull; the three new strings are translated in all eight locales, and USAGE.md describes the button. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…t Continue Chapter list - The series page and the reader asked the books route for one page of 1000, so One Piece's chapters 1001+ never appeared in the list or in the reader's prev/next/jump list. lib/seriesBooks.ts fetchAllBooks() walks every page (tested). The smart-offline planner had the same cap server-side. - The list now renders 100 rows a page (lib/chapterPages.ts, tested) with a pager above and below (first/prev/range picker/next/last). It opens on the page holding the "Continue" chapter; a tap on the pager pins a page, and a new series, sort, or filter follows Continue again. A chapter chip in Sources & translations turns to the page holding that row before it scrolls to it. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
5 of 8 tasks
…per book On top of @Squeaks72's paging, which fixes a real bug on main -- Continue opened chapter 1 for anyone past chapter 1000: - The picker named pages by row position. Ghost, run and "show all" rows are rows but not chapters, and newest first labelled the page holding 1193-1094 "1-100". Pages are now named by the first and last chapter numbers they show, in the list's own order, and the count beside it is chapters. - The page followed Continue live, so expanding an older-chapters run, "Show all" or "Mark all read" flipped the list away from what the reader was doing. It is decided once per series, sort and filter, once the chapters and the sources' listing (whose ghost rows would shift it) have both arrived. - A scan between the two requests could hand the same book back at the 1000 boundary: two rows, one React key. Kept once. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ns keep their direction On top of @Squeaks72's right-to-left paging, which also fixes a bug on main (under the Arabic interface the paged track inherited RTL and the page counter stuck on page 1): - Resume and `?page=` links opened a right-to-left series on page 1. A streamed series learns its direction in the same tick it becomes ready, so the new flip effect ran right after the resume scroll with `current` still 0 in its closure. It now acts only on a real flip, tracked by the direction the track was last laid out for. - In paged mode the bottom bar follows the track both ways. On a left-to-right read under the Arabic interface it inherited RTL and ran opposite to its own pages. - Text inside the track (the missing-page caption, the Up Next and failure cards) took the track's direction: Arabic in an LTR paragraph, or English in RTL on a right-to-left read. It keeps the interface's, via useRtl(). - A double-tap zoom in paged mode centred scrollLeft, which there IS the page: it jumped to mid-chapter (page 1 on a right-to-left track). Centring is for the webtoon column only. - A right-to-left series forced to read left to right still swaps its spreads, on purpose: a double-page spread is one drawing and only reassembles with its first page on the right. Said so in the code. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…keys leave sheets and controls alone On top of @Squeaks72's paged-mode keys: - Steps are computed from the slide on screen, not "one width from wherever the scroll is": scrollBy stacked on a smooth scroll still in flight, so two quick presses or a held key landed one or two pages on. - A key with Alt, Ctrl or Meta is never the reader's. Alt+Left (browser Back) turned a page, and Ctrl+F opened the browser's Find AND toggled fullscreen -- the second one was already on main. - With a sheet open (settings, page grid, chapter list) keys no longer act underneath it, and Escape closes the sheet instead of leaving the reader; none of the three listens for Escape itself. - A focused button or link keeps Space and Enter: Space on a chapter row turned the page and swallowed the click. - Shift+Space goes back a page, in both modes. - On a right-to-left read (with the paged-direction change beneath this) Left and Right stay physical, like the tap zones, while Space, PageDown and Down mean "next" whichever way the pages run. - The page slider lets go of focus after a drag, so in the webtoon column Space scrolls again straight away. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # web/public/locales/ar.json # web/public/locales/de.json # web/public/locales/es.json # web/public/locales/fr.json # web/public/locales/ja.json # web/public/locales/pt-BR.json # web/public/locales/ru.json # web/public/locales/zh.json
…aks into the global one On top of @Squeaks72's per-source reader settings: - Choosing a look in the reader also saved it as the GLOBAL default: setPref saved the reader's live prefs wholesale, and those are the default with the source's and the series' settings laid over it. So "switch to paged, pin this source" left every other source opening paged too, and even a brightness change saved the source's look as everyone's. A change made with a title open now writes its look to that title only, and only what it touched to the global default, which is set under Profile -> Settings. globalPrefsChange() is the rule, and it has a test. - The default never applied to a downloaded chapter, online or off: the reader loads the offline record first and it names no source, so `chapters[0].sourceId` was empty. A series mixing copies from several sources flipped its look mid-read. It is keyed by the series' primary source now, learned when the series loads and remembered on the device for opening it offline. - The row could show a nineteen-digit extension id for a name. With no name there is no row. - The reading direction from the paged-direction change beneath this is part of the remembered look, per series and per source ("this manga site: paged, right to left" can now be pinned). - The settings sheet is capped at 85dvh and scrolls: with the two new rows it pushed its top off a short phone. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…itched off, and loses no keys On top of @Squeaks72's type-to-search: - Under a Japanese or Chinese interface the palette opened holding the raw Latin letter and the IME composed after it ("w" + "あんぴーす"): with nothing focused the browser does not route keys through an input method. There it now opens empty and focused, and the IME composes the title from the next keystroke. Safari's spelling of an IME keystroke (keyCode 229, key "Process") also counts as composing. - Single-key shortcuts must be possible to switch off (WCAG 2.1.4). Profile -> Settings -> "Type anywhere to search" does, on this device, and covers "/" as well. Ctrl/Cmd+K stays. - A letter typed while the palette was open but its box had lost focus was swallowed. The palette is now a dialog (role, aria-modal), so the global handler stands down for it, and the open palette puts stray letters into its box. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…aches every surface it promises On top of @Squeaks72's configurable 18+ filter, whose permission model holds (max_age_rating stays the only ceiling; hidden sources are still never asked; admin-only): - The genre list was pasted into SQL as literals behind a character whitelist -- the one place in the codebase admin-entered data met a query string. browsable() now names the column and reads the list with jsonb_array_elements_text, so no value can escape anything. That also fixed three things at once: - notification digests ignored the genres (notify builds its context without loading the lists); - case was folded in JavaScript on the way in and by Postgres on the way out, so a genre the two lowercase differently never matched while its chip showed ticked -- both sides fold in SQL now; - the whitelist was dropping real genres ("Boys’ Love", Thai, Devanagari). Genres are stored as typed. With no genre configured an uncorrelated check short-circuits the clause, so listings pay nothing. - "Find missing chapters" and Fetch answered 404 for a series the switch hid: POST /api/sources/fill/scan and /fetch looked the series up through browsable(). They act on a series someone opened, so they answer to visible() -- the permission. This was already true of 18+ libraries on main. - Source ids went through the genre whitelist, which dropped `_` and anything past 60 characters: the chip lit, then went dark. They get their own shape (sanitiseSourceIds). - The source hunt read only an extension's own flag, so a source the admin named adult could be followed onto a clean series by the failure hunt or the nightly repair. sweepAllowedFor now honours the list. - docs/api.md describes the named genres and sources, and the two by-id routes; the token checkboxes say "Include 18+ content", since they now cover more than libraries. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…never shown as one @Squeaks72's change showed chapter names and which chapter Continue opens -- both good -- by writing names into lib_books.title. But title is the FILENAME's: bookDto fills `name` and `metadata.title` from it, and on a library built by hand that is `One Piece v02 c012 [Digital]`, which became the name on most rows. And every screen that prints the title alone (reader header, History, offline Downloads, OPDS) would have lost the chapter number once a name filled it. - The source's name goes to its own nullable column, lib_books.chapter_name (a v0.46.0 block at the end of the DDL), and to `chapterName` on the book DTO. `title` is never touched again, so everything printing it reads exactly as before, and "Vol." is still decided from the filename. - The scan's upsert is back to main's `title=EXCLUDED.title`: nothing to protect any more, and no per-row regex over a 40k-chapter library. - chapterName() takes the volume, the chapter word and the number off the FRONT, in the ways sources write them ("Vol.3 Chapter 12", "Capítulo 12", "第12話", "12화"), and keeps what follows: "Vol.3 Chapter 12: The Return" is named "The Return". 12 does not match the front of 120 or 12.5. The listing heal uses the same rule in JavaScript, with an equality join, and fills only chapters that have no name yet. - The web shows `chapterName` and nothing else -- no fallback to the filename. - Continue no longer reads "page 0 of 23" after Mark unread. Each fix was proved by undoing it alone: writing the name into `title` again, the whole-string rule, the column dropped from booksSrc, and the web falling back to the filename each fail their own test. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It was inserted into the middle of the v0.40.0 block (after auto_follow_on_failure, and beside the age-rating columns), the same spot two other open pull requests had also chosen -- three-way conflicts on a line nobody meant to share. Each release's DDL is appended at the end under its own version comment, as v0.43.0 did. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…tings sheet speaks every language Two bugs found while reviewing the reader pull requests, both already on main: - In paged mode with the default "Collapse", a repeated page got a slide of its own that never rendered: the render window skips collapsed pages, and a paged track has no thin band to draw one as. The reader saw a blank screen with a faint page number. There, Collapse now shows the page like any other -- one swipe. Not removed, although the sheet's comment said so: removing would give the two modes different page orders, and nothing re-anchors the current page when the flow is rebuilt, so switching mode mid-chapter would land on another page. The comment now says what happens. - The reader settings sheet hard-coded English: Mode, Theme, Fit, Pages per view, Brightness, Page gap, Auto-scroll, the mode buttons and the theme names -- every one of them already translated in all eight locale files, just never asked for. The CSS `capitalize` that dressed up the raw theme ids is gone with them. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
_meta.strings is the number of keys, and the Continue line's key went in mid-file without it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
"Include 18+ libraries" became "Include 18+ content" when the admin's 18+ filter started hiding more than libraries. USAGE.md still sent readers to Profile -> Connections -> External readers -> Include 18+ libraries in this reader, a place the app no longer has, and consoleNamesInDocs caught it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ranch The PRs merge in order, so each branch carries the ones before it and merges without conflicts.
The PRs merge in order, so each branch carries the ones before it and merges without conflicts. # Conflicts: # web/public/locales/ar.json # web/public/locales/de.json # web/public/locales/es.json # web/public/locales/fr.json # web/public/locales/ja.json # web/public/locales/pt-BR.json # web/public/locales/ru.json # web/public/locales/zh.json
The PRs merge in order, so each branch carries the ones before it and merges without conflicts.
The PRs merge in order, so each branch carries the ones before it and merges without conflicts. # Conflicts: # web/public/locales/ar.json # web/public/locales/de.json # web/public/locales/es.json # web/public/locales/fr.json # web/public/locales/ja.json # web/public/locales/pt-BR.json # web/public/locales/ru.json # web/public/locales/zh.json
The PRs merge in order, so each branch carries the ones before it and merges without conflicts. # Conflicts: # web/public/locales/ar.json # web/public/locales/de.json # web/public/locales/es.json # web/public/locales/fr.json # web/public/locales/ja.json # web/public/locales/pt-BR.json # web/public/locales/ru.json # web/public/locales/zh.json
The PRs merge in order, so each branch carries the ones before it and merges without conflicts. # Conflicts: # web/public/locales/ar.json # web/public/locales/de.json # web/public/locales/es.json # web/public/locales/fr.json # web/public/locales/ja.json # web/public/locales/pt-BR.json # web/public/locales/ru.json # web/public/locales/zh.json
The PRs merge in order, so each branch carries the ones before it and merges without conflicts. # Conflicts: # bff/src/lib/migrate.ts # web/public/locales/ar.json # web/public/locales/de.json # web/public/locales/es.json # web/public/locales/fr.json # web/public/locales/ja.json # web/public/locales/pt-BR.json # web/public/locales/ru.json # web/public/locales/zh.json
AngeloSha
added a commit
that referenced
this pull request
Sep 25, 2026
… extension engine Right-to-left paging (#90), one key per page (#92), reader settings pinned to a source (#95), type-to-search (#94), chapter names (#84), a chapter list that pages past chapter 1000 (#89), Discover keeping its source through a search (#87) and an 18+ filter an admin can widen (#86) -- seven merged with fixes on top, one as it was. The extension engine (Suwayomi-Server, headless) gets the desktop app's memory cap in every compose file (#72). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Owner
|
Thanks for this. Chapter names are something I've wanted for a while, and saying which chapter Continue opens is a really nice touch. I had to change how the name is stored before merging. So the name now has its own column, Each change is its own commit on your branch with the reasoning. Merged, and it shipped in v0.46.0. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Chapters never showed a name, only their number. The cause is in how they are stored. The downloader names every file from its number alone (
Chapter 12.cbz), and the scanner takeslib_books.titlefrom the filename, so the title ends up as the number again. The source's own name (out.chapterUsed.title) was available when each chapter landed, but nothing saved it.This PR fixes that in four places and shows the result:
setBookMetanow takes the landed copy'stitlefrom the add, download and nightly-sweep paths. A newchapterName()filter (bff/src/lib/library.ts) drops titles that only repeat the number (Chapter 12,Ch. 12,Episode 012,12). Without it, a copy from a source that says "Chapter 12" would replace a real name that an earlier copy supplied.replaceListing()already writes the source's listing for the same chapters, so it now fills anylib_books.titlethat is still just its number, using the listing title when that title is a real name. This fixes existing libraries without a migration, and a hand-set title never matches the pattern. On one real library this recovered names for about 866 chapter rows. Many sources only ever publish numbers, so plenty of rows stay number-only, and that is expected.persistScan's upsert rantitle=EXCLUDED.titleon every add, sweep and manual scan, which reset every stored name back to "Chapter N" whenever anything anywhere was fetched. Now the scan keeps the stored title when the filename says nothing but the number and the row holds a real name. A file that carries its own name (Chapter 4 - Title.cbz) is still titled from the file.Ch. 12 · The Sound of Thunder(web/lib/format.tschapterName(), the same filter). The series page names the chapter the Continue button opens, withpage 7 of 23when it is part-read.Type
Testing
Re-run after rebasing onto v0.45.0 (91b41aa): bff 1861 tests, 1848 pass, 9 fail, 4 skipped — the same 9 embedded-Postgres entrypoint failures
mainhas in this environment (main: 1857 / 1844 / 9 / 4), nothing new; web 432/432; both typechecks OK. The detail below is from the first run, on v0.44.0.I ran the upstream CI setup in
node:24-alpineagainst a realpostgres:16-alpine(TEST_DATABASE_URLset, so the SQL tests run instead of skipping), first on plainupstream/main(a8bc502) to get a baseline and then on this branch.maintsc --noEmittsc --noEmitThe same 9 failures appear on both sides. They are the
entrypointEmbeddedtests, which need an embedded Postgres binary that this sandbox lacks. There are no new failures.One upstream assertion changed.
updater.int.test.ts› "only the chapters that landed are stamped…" pins the exact shape ofr.landed. The sweep now also records the chapter'stitlethere, so the expected object gainstitle: 'Chapter 6'. The test's intent is unchanged: a book that did not land keeps its stamp. The full run had this as its only extra failure. After the fix I re-ran the affected DB files (updater,chapterNames,scan,seriesListing,scanlatorStamp) and all 77 tests passed, which gives the totals in the table.New tests:
bff/test/chapterNames.int.test.ts:setBookMetastores a real name, and a later bare title does not replace itpersistScanover CBZs on disk keeps a stored name, and a file with its own name still wins. This one fails against the oldtitle=EXCLUDED.titleupsert.web/test/format.test.ts:chapterNamefor the web.The new UI string (
{chapter} · page {page} of {pages}) is translated in all eight locale files.npm run test:i18nis a Puppeteer walk against a running server, so it could not run in this sandbox. The locale-parity unit test passes.Checklist
docs/USAGE.md)🤖 Generated with Claude Code