pdfjs-dist 5.7.x (bumped 5.4.530 -> 5.7.284 in #4143) moved the JBIG2
image decoder -- the codec used by virtually every black-and-white
*scanned* PDF -- from pure JS into a WebAssembly module the worker
fetches at runtime from `wasmUrl` (/vendor/pdfjs/). The `copy-pdfjs-wasm`
script only copied an explicit allow-list ({openjpeg.wasm,qcms_bg.wasm})
and silently dropped jbig2.wasm. cpx does not error on an empty glob, so
the missing decoder went unnoticed: scanned PDFs rendered blank (pages
still turned) in CI builds, while local `tauri build` reused a stale
pre-bump public/vendor copy and worked. Regression between 0.11.1 and
0.11.2.
Copy the whole wasm/ dir (mirrors the {cmaps,standard_fonts}/* fonts
pattern) so future decoders moving to wasm can't be dropped again, and
ship the *_nowasm_fallback.js files for graceful degradation.
Adds a regression test asserting every .wasm the bundled pdf.js
references is covered by copy-pdfjs-wasm.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A cover with `data-duokan-page-fullscreen` on <html> renders in
paginated mode but is blank in scrolled mode; only the first cover is
affected, other images are fine in both modes.
The paginator's fullscreen branch pins such images with
position:absolute and height:100% and forces their ancestors to
height:100%. That fills the fixed-height page when columnized, but in
scrolled mode the container height is `auto`, so height:100% resolves
to 0 and the cover collapses out of view.
Bump foliate-js to gate the fullscreen treatment on column mode and
reset any stale absolute pinning when a fullscreen-cover doc is laid
out scrolled (so toggling paginated -> scrolled also recovers). Add a
browser regression test + repro-4379.epub fixture.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
sync: empty-start/end range CFIs left by the cfi-inert skip-link bug (e.g.
epubcfi(/6/24!/4,,/20/1:58)) resolve to a section-spanning range and navigate
to the wrong end of the section. Add isMalformedLocationCfi and discard such
locations on the cloud-sync receive path (useProgressSync) and the kosync push
path (useKOSync) so they can't move the reader or propagate to other devices.
foliate 569cc06 stops generating them but does not repair already-synced values.
reader: bump foliate-js to 167757a to fix the white<->black background flash
when swiping between differently-colored pages; add a regression test for the
sliding per-view background segments.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a "Highlight Current Sentence" keyboard action (default Shift+M, in the
Text to Speech shortcut section) that persists the sentence TTS is reading
aloud as a normal highlight using the user's default style/color — no text
selection, eyes-off, silent, and idempotent (a repeat press on the same
sentence is a no-op rather than a duplicate).
Flow: the shortcut handler in useBookShortcuts dispatches tts-highlight-sentence
→ useTTSControl (which owns the TTSController) resolves the current sentence via
the new TTSController.getSpokenSentence() and relays create-tts-highlight
→ Annotator builds the BookNote with the pure, unit-tested buildTTSSentenceHighlight
helper and persists/renders it like any other highlight.
Closes#4085
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(kosync): compare reflowable conflicts via locally-resolved CFI percentage
KOReader reports progress as a percentage from its own pagination, which isn't directly comparable to Readest's progress. For reflowable books, resolve the remote XPointer to a local CFI and compute the equivalent fraction (getRemoteLocalFraction), comparing that against the local percentage and falling back to the reported percentage only when it can't be resolved locally (non-XPointer progress or a missing section). The resolved fraction also drives the conflict-dialog remote preview so the shown value matches what was compared.
Loosen the conflict threshold to 0.01 when the remote progress was last pushed from this same device (remote.device_id === local deviceId), so sub-page drift between a push and the next pull doesn't prompt. Render sync percentages with 2 decimals via formatProgressPercentage.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(reader): correct scrolled-mode reopen drift over background-image sections
Bump the foliate-js submodule to include the scrolled-mode reopen drift fix for sections with background images, and add a browser regression test plus its EPUB fixture.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(library): redirect to login on pull-to-refresh when signed out
Guard the pull-to-refresh handlers so an unauthenticated user is sent to the login screen instead of attempting a library pull and OPDS subscription check.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(memory): add kosync conflict + toc/scrolled-restore notes
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(reader): prevent CFI crash on inert-only section bodies
Reopening/paginating across a background-image or otherwise content-less section could crash with "Cannot destructure property 'nodeType' of 'param' as it is undefined" in foliate's fromRange, aborting the relocate so the reading position was never saved. Bumps the foliate-js submodule to 569cc06 (visible-range walker skips cfi-inert skip-links; isTextNode/isElementNode are null-safe) and adds a regression test reproducing the exact crash.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(kosync): keep auto-push working when a pull finds no real conflict
In the 'prompt' strategy, pullProgress set syncState to 'conflict' unconditionally on every pull that returned remote progress, even when promptedSync found no actual difference. Since auto-push only runs while 'synced', and a pull fires on every book-open and window re-activation, progress stopped being pushed. promptedSync now returns whether a real conflict was surfaced, and pullProgress only stays in 'conflict' for genuine conflicts (otherwise 'synced').
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(settings): show most recent sync time and reorder settings tabs
Library settings menu now reports the latest of the book/config/note sync timestamps as "Synced …" instead of only the books timestamp. Reorder the settings tabs so Integrations precedes TTS.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
generateKOProgress built its XCFI converter from the paginator's
primaryIndex and the rendered primary document, then converted
progress.location. Because #primaryIndex can lag behind the viewport
during scrolling, the CFI's spine section could differ from the
converter's, tripping XCFI's guard ("CFI spine index N does not match
converter spine index M") and silently dropping the progress push.
Route through getXPointerFromCFI, which keys off the CFI's own spine
index and loads the correct section's document from the book when the
rendered index doesn't match. Fall back to the cached config.xpointer
on failure.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
In scrolled mode, SectionInfo paints a solid `bg-base-100` `notch-area`
mask over the top safe-area strip at z-10. The Ribbon was also z-10 but
rendered earlier in the DOM, so the equal-z mask painted over the
ribbon's upper (unsafe-area) half — only the lower 44px showed. In
paginated mode the mask has no background, so the ribbon showed fully.
Raise the ribbon to z-20 so the whole ribbon stays visible above the
mask, and mark it pointer-events-none so taps still fall through to the
notch mask's scroll-to-top handler.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A single MDX commonly ships its resources across several MDD files sharing the MDX's filename prefix (e.g. `Name.mdd` for images, `Name-02.mdd` for scripts, `Name-03.mdd` for audio). The importer only grouped the exact-stem `Name.mdd`, so the other MDDs were dropped as orphans and their resources (notably audio) could never be loaded.
`groupBundlesByStem` now attaches every `.mdd` whose stem starts with an `.mdx` stem at a separator boundary to that MDX bundle (longest prefix wins on overlap); the boundary check prevents false merges like `dict.mdx` claiming `dictionary-words.mdd`. The runtime provider, contentId, and replica sync (binary upload + manifest apply) already treat `files.mdd` as a list, so multi-MDD bundles sync across devices with no further changes.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(dictionary): keep other dictionaries usable when System Dictionary syncs to an unsupported platform
`dictionarySettings.providerEnabled` is whole-field synced across devices, so enabling System Dictionary on macOS/iOS sets the flag on web/Linux/Windows too. There the row is hidden and the feature is a no-op, but the settings UI read the raw flag and locked every other dictionary's toggle read-only. Gate the lock on `isSystemDictionaryEnabled(settings)` — the same platform-aware check the annotator uses — so it matches real lookup behavior and never triggers where the system dictionary can't run.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(dictionary): dispatch system dictionary handoff by native OS (fixes iPad)
iPadOS sends a desktop "Macintosh" user agent, so the UA-based `getOSPlatform()` reported iPad as 'macos' and the handoff invoked the macOS-only `show_lookup_popover` Rust command that iOS never registers ("Command show_lookup_popover not found"). Derive the OS from the app service's `is*App` capability flags (sourced from the Tauri OS plugin, correct on iPad) via a new synchronous `getInitializedAppService()` accessor, so iPad routes to the iOS plugin command path.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(ui): constrain reader View Options dropdown to h-8
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(agent): note System Dictionary platform-detection and synced-flag patterns
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(agent): add agent notes for cache, reading-ruler, foliate touch
Add project-memory notes and index entries:
- manage-cache-ios-layout: iOS container layout and what Manage Cache clears
- reading-ruler-line-aware: line/column-aware reading ruler internals
- foliate-touch-listener-capture-phase: capture-phase gesture suppression
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(reader): pad sidebar and notebook for the device status bar (#4089)
Top-anchored slide-in panels (sidebar, notebook) only applied status-bar
top padding when isFullHeightInMobile was true. On a tablet/desktop
(isMobile === false) that gate collapsed the padding to 0, so a visible
system status bar overlapped the panel's top toolbar and made its icons
inaccessible.
Extract the inset math into getPanelTopInset() and gate it on
(!isMobile || isFullHeightInMobile) so non-mobile panels clear the status
bar like the reader header, while a partial-height mobile bottom sheet
(which doesn't reach the top of the screen) stays flush.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(reader): keep footer bar clear of the pinned sidebar
On a mobile tablet in portrait, forceMobileLayout renders the footer bar
with position: fixed, anchored to the viewport, so left-0 w-full spans
the whole window and slides under a pinned sidebar — the progress / font
/ TTS controls end up obscured.
Anchor the footer inside the book's grid cell (position: absolute) when
the sidebar is pinned, mirroring the header bar. The flex layout already
offsets the grid cell by the sidebar's real rendered width, which honors
the sidebar's min-w-60 floor and 45% cap that a stored-width offset would
miss. The slide-up panels are absolute within the footer container, so
they shift and narrow with it and their animation is unchanged. The
switch only happens when the sidebar is pinned, so phone (< 640px) and
unpinned tablet-portrait class names stay identical.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(ui): refine reader side panels and their empty states
Closes#4089
Add a shared EmptyState component (large muted icon, title, and an
optional hint or action) and use it for the empty annotations, bookmarks,
and notes panels in the sidebar and notebook, replacing the ad-hoc
"No … yet" placeholders.
Polish the surrounding chrome: switch the bookmark toggler to the Ri icon
set with responsive sizing, crop the HighlighterIcon viewBox to its
artwork to remove the asymmetric bottom padding, and tune mobile sizing
and spacing across the panel headers, tab navigation, and footer nav bar.
Translate the new empty-state strings (No Notes, No Annotations, No
Bookmarks, and their hints/action) across all 33 locales and drop the
obsolete "No … yet" keys.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a "Manage Cache" item to the library Advanced Settings menu (native
mobile apps only) that opens a modern dialog showing the combined size and
file count of the app's reclaimable storage, with a confirm-gated clear that
reports per-file progress.
- iOS clears Cache + Temp + Documents/Inbox; Android clears Cache + Temp.
- Multi-source helper (getCacheEntries/getCacheStats/clearCacheEntries) with
unit tests; per-file failures are counted, never abort the run.
- Dialog uses the centered-hero + btn-contrast design language, theme-neutral
progress, and is e-ink correct.
- i18n: new strings translated across all locales (+ en plural forms).
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Snap the reading ruler to real rendered text lines instead of stepping by a
fixed arithmetic height, so the band always frames whole lines.
- Snap to actual line geometry from the relocate range; the band is sized
dynamically to the text block plus symmetric padding (round(fontSize *
lineHeight * 0.3)), capped at (lines + 1) line heights so a tall image inside
a block can't expand it to cover the whole figure.
- Drop block/container rects (Range.getClientRects aggregates multi-line element
borders) so paragraphs aren't merged into one giant line and skipped.
- Column-aware in multi-column layouts: the band spans one column at a time and
advances column by column.
- Confine the band to lines at least half visible within the viewport.
- Scrolled mode: snap to lines, and at a view edge scroll the view and realign
the band to the start/end of the new view (works for vertical-rl too, which
scrolls horizontally); paging snaps the view edge between lines so text isn't
cut or repeated.
- Vertical writing mode: correct band centering and drag direction; Up/Down keys
move the ruler while Left/Right turn pages (taps always move the ruler).
- Page turns keep the first/last line: forward lands on the first line of the new
page, backward on the last line; the relayout re-snap anchors on the band's
leading edge so it never skips a line.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: design spec for gesture-based brightness control (#3021)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: revise brightness-gesture spec per /autoplan review (#3021)
CEO+Design+Eng dual-voice review. Key fixes: capture-phase listener
(bubble-phase could not suppress foliate paginator), opt-out toggle,
18px threshold, selection guard, brightness seed race, rAF teardown,
e-ink stepped overlay, contrast capsule, perceptual curve reuse,
listener-level test harness.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(reader): swipe-to-adjust brightness gesture on mobile (#3021)
Left-edge vertical swipe adjusts screen brightness on iOS/Android, with a
Sun-icon progress overlay. Capture-phase non-passive listener suppresses the
foliate paginator / page-flip / UI-toggle handlers; selection guard, strip
reservation in scrolled mode, eager brightness seed, rAF throttle + teardown.
Opt-out toggle in Settings > Behavior > Device (default on). Perceptual curve
shared with the menu slider. Pure-helper + listener-level tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(reader): detect brightness-swipe edge by screenX; i18n + shorter label (#3021)
On-device fix: paginated mode lays the iframe doc out as wide side-by-side
columns, so clientX/documentElement.clientWidth are document coordinates and a
left-edge touch on a later page never fell inside the strip (armed stayed false).
Detect with screenX against the parent window width, matching usePagination.
Also: translate the two new setting strings across all locales and shorten the
toggle description to 'Slide along the left edge'.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
rsvp-controller.test.ts calls controller.start() in ~20 tests but never stops
the controller. start() schedules a countdown (setInterval, 500ms x3) that then
schedules the recurring word-advance (setTimeout). On the real clock those fire
~1.5s later — after the test file's jsdom env is torn down — and
emitStateChange's dispatchEvent(new CustomEvent(...)) runs against a stale realm
and throws. Vitest reports it as an unhandled error and fails the whole run
intermittently on CI:
TypeError: Failed to execute 'dispatchEvent' on 'EventTarget': parameter 1 is
not of type 'Event'.
at RSVPController.advanceToNextWord -> emitStateChange (Timeout._onTimeout)
Fake only the timer functions (setTimeout/setInterval + their clears) for this
suite via vi.useFakeTimers({ toFake: [...] }); useRealTimers in afterEach
discards any still-pending fakes. The tests assert synchronously and never
advance playback, so faking the timers changes no assertion; Date/performance
stay real so CFI/position checks are unaffected. Test-only; production behaviour
is unchanged.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The two #4112 scrolled-mode preload browser tests asserted that the
previous-previous section (F-1) was not loaded after navigating to a section,
checking it after `waitForFillComplete()`.
The eager backward buffer (minPages) is suppressed while `#display` is
stabilizing, but pulls F-1 (and beyond) in on the first scroll events once the
fill settles. So the post-fill assertion raced the buffer: it passed locally but
failed intermittently on CI (loaded set [1,2,3,4,5]).
Assert the precondition the instant `#display` stabilizes — where the buffer is
provably suppressed and only the immediate-previous section is loaded — before
the fill settles. The eager buffer loading F-1 later is exactly what the second
test wants once it navigates back. Test-only; production behaviour is unchanged.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
On a hard refresh the TOC sidebar occasionally (~1 in 10) scrolled to and
highlighted the current chapter, then rewound to the very top of the list.
It is a scroll-position race in TOCView, not a progress/sectionHref reset (the
reading position stays correct throughout). OverlayScrollbars resets the wrapped
Virtuoso viewport's scrollTop to 0 when it initializes (deferred). Its
`initialized` callback re-scrolled only to `initialScrollTarget.index`, captured
at mount — and on a fresh refresh `progress` is not available yet, so that index
is 0 and the reset is never corrected. Whether OverlayScrollbars initializes
before or after the auto-scroll to the reading position is the timing race that
made it intermittent.
Re-apply the scroll to the current active item (via refs mirroring the live
flatItems/activeHref) in the `initialized` callback, falling back to the
mount-time index. Refs are used because OverlayScrollbars binds the callback at
mount and fires it later, so it must read the latest active item.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(export): make annotation export link type configurable
Add an Annotation Link selector (App / Web) to the Export Annotations
dialog. Defaults to the app deeplink in the native app and the universal
web link on the web, so web exports no longer emit readest:// links that
only the desktop/mobile app can open. The default markdown template now
uses the configurable annotation.link variable.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(annotations): move Moon+ Reader import into a dedicated Import Annotations modal
Replace the single 'Import from Moon+ Reader' menu item with an 'Import
Annotations' entry (below 'Export Annotations') that opens a dedicated
modal listing import sources. Currently lists Moon+ Reader; the boxed-list
layout makes adding future providers a one-row change.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Bump foliate-js: compensate the scroll position when a previous section is prepended above the viewport, so the browser's scroll-anchoring suppression at scrollTop 0 no longer drifts the view into chapter n-1; preload the previous section on backward navigation and eagerly while scrolling toward the top; and stop the blank-screen flash on adjacent navigation in continuous scrolled mode.
Split paginator-multiview.browser.test.ts into paginator-scrolled and paginator-paginated, adding regression tests for the drift, previous-section preloading, the eager backward buffer, and the flash-free transition.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds a primary/secondary sort pair so users can group by author and have
each author's books drilled-in list sort by series, without touching the
sort menu each time. Closes#4307.
- New "Then by..." picker in the library view menu (None + same keys as
primary). Secondary acts as tiebreaker for the global sort, and as the
in-group ordering when the user drills into a non-series group.
- Smart defaults derived from groupBy, surfaced as "(Auto)" in the menu
and resolved at sort time so user picks are never overwritten:
- groupBy=Author + secondary=none -> Series
- groupBy=Series + librarySortByAuto -> primary becomes Series
- librarySortByAuto flips off as soon as the user makes any explicit
primary pick; subsequent groupBy changes then respect that choice.
Settings: librarySortBy2, librarySortByAuto. URL: ?sort2 syncs the
secondary; auto is settings-only (no URL representation).
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Windows WebView2's native share UI, invoked via tauri-plugin-sharekit,
synchronously blocks the Tauri command thread waiting on
ShareCompleted/ShareCanceled callbacks. When the user dismisses the
picker without those callbacks firing, the app freezes and has to be
killed from Task Manager. Skip the native share path on Windows and
fall through to the save dialog, mirroring Linux.
* fix(library): suppress Android image callout on book covers
Long-pressing a cover on Android could trigger the WebView's native
image callout at the same time as the bookshelf's own 500ms long-press
handler for multi-select, causing apparent freezes. `-webkit-touch-
callout: none` doesn't inherit, so the existing `.no-context-menu`
rule on the item container never reached the cover `<img>`. Apply the
callout suppression to descendant images/anchors and disable native
drag on the cover.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(library): show modal for multi-file import failures
When a batch import yields more than one failure, the previous toast
crammed every filename onto a single line that often overflowed and
truncated. Add a dialog that lists each failed filename with its
error reason, dedupes the message into a header banner when every
file failed for the same reason, and falls back to the existing toast
for single-file failures.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(library): sort manage-group modal by most recent activity
The Group Books modal listed groups in store-insertion order, which made
recently-active groups hard to find in libraries with many groups. Sort
each level desc by the newest `updatedAt` across the group's books,
propagating up the path so a recently-touched book in
`Literature/Fiction` keeps `Literature` fresh too. Extract the index as
`buildGroupNameUpdatedAt` in libraryUtils for reuse and unit testing.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(library): tighten select-mode action bar and header polish
- SelectModeActions: switch the narrow-viewport grid from 3 columns to
4 (with the delete action explicitly placed in column 2) so the icon
set stops wrapping awkwardly on phones below ~500px.
- LibraryHeader: keep the "Select All" / "Deselect" label on a single
line so it doesn't wrap and shove the underlying button taller.
- SetStatusAlert: drop the hover bg on the small-screen cancel button
and rely on text-color contrast so it stops flashing a tinted disc
on mobile taps.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Closes#4222.
Three changes that together stabilize Readest sync across devices:
**Stop the artificial updatedAt bump in useProgressAutoSave**
saveConfig unconditionally bumps config.updatedAt = Date.now(), and
useProgressAutoSave used to fire saveConfig on the very first relocate
after book open — even though that relocate just reflects the position
loaded from disk, not user action. The stale local config then looked
"newer" than a fresher server-side push, so the next auto-push
overwrote the other device's real progress via last-writer-wins. The
hook now snapshots the loaded location and skips saveConfig when the
in-memory location still matches it.
**Retry the first config pull with backoff + release the gate**
useProgressSync gates pushes behind a successful pull (so a brand-new
import can't clobber the server's real progress). But handleAutoSync
only re-arms on progress.location changes, so a single failed pull
(Android cold-start contention, Wi-Fi/LTE handoff, captive portal)
used to block every push for the whole reader session. The new
pullWithRetry retries on backoff (1500/4000/10000 ms) and releases
the gate after exhaustion — server-side last-writer-wins still
protects the cross-device case (a stale local push with an older
updated_at loses to a fresher server record). sync-book-progress
events reset the chain so manual pull-to-refresh recovers cleanly.
Fetch timeout bumped from 8s to 15s to better tolerate slow networks
in that cold-start window.
**Server piggybacks books.progress off configs push**
/api/sync POST now updates books.progress + books.updated_at for each
upserted config, gated by .lt('updated_at') so a concurrent newer
books push is never downgraded and a missing row is a silent no-op
(useBooksSync still seeds new rows from the library page). The
in-reader syncBooks round-trip is dropped — the reader now sends one
POST per auto-save instead of two, and the books row stays consistent
with config pushes even while a reader stays open (#4198).
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Fixes#4339. PostHog telemetry was previously enabled by default for every
install, which surprised privacy-conscious users on self-hosted setups.
Behavior for new users only (existing users are migrated to a decision that
preserves their current `telemetryEnabled` setting):
- 90% are silently opted out on first launch.
- 10% see a one-time consent prompt; accepting opts in, declining opts out.
Decision is persisted via a new `readest-telemetry-decision` localStorage key
so subsequent boots don't re-roll. PostHog now inits with
`opt_out_capturing_by_default` so brand-new users never ping before the
decision is finalized.
New `TelemetryConsentDialog` uses the project's `btn-contrast` (theme-neutral)
CTA and `eink-bordered` chassis so it renders correctly under `[data-eink]`
without color-mode-only assumptions. Strings translated across all 33 locales.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(dict): restore MDD eager init; CSP-safe Vocabulary.com audio handler
Two MDict fixes:
1. **Regression**: #4334 turned on `MDD.create(..., { lazy: true })` so
the audio / resource side of every MDict would skip the upfront
key-block decode + sort. That's the right trade-off on the MDX side
(millions of headwords, ~80 s init saving), but it doesn't transfer
to MDDs: js-mdict's lazy lookup binary-searches `keyInfoList`
envelopes with `comp` (localeCompare), and MDDs store keys in
byte-sorted order with `\` prefixes and case folding. The two orders
disagree just often enough that `mdd.locateBytes('sound.png')`
misses on resource paths that do exist — the icon's CSS
`background-image: url(sound.png)` then stays unresolved and the
speaker glyph disappears. Bisected to 93abca89 (#4334). Drop the
flag for MDDs; MDXs keep the lazy win.
2. **New**: a CSP-safe replacement for the `onclick="v0r.v(this,'KEY')"`
audio handler used by Vocabulary.com-derived MDicts. The dict's
`j.js` is never loaded inside our shadow root (we don't execute
MDX-supplied JavaScript), so without intervention the inline
handler would throw `ReferenceError: v0r is not defined` on every
click. The new helper parses the audio key from any matching
onclick, strips that one attribute, and binds our own listener that
probes the companion MDD for `<KEY>.mp3` / `<KEY>.m4a` / etc. and
plays the bytes through an `<Audio>` element. Other unrelated
inline handlers are deliberately left alone — they were no-ops
already (their helper JS isn't loaded), and CSS rules sometimes
match on `[onclick]` attribute presence.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(sync): skip binary upload when sync category is disabled
`publishReplicaUpsert` already gates the metadata row on
`isSyncCategoryEnabled(kind)` so a user who turned dictionary / font /
texture sync off in Manage Sync never publishes a row for that kind.
`queueReplicaBinaryUpload` had no such gate, so the binary side still
fired: a 250 MB MDict bundle or a few hundred MB of font files would
still upload to cloud storage with no replica row to reference them.
Mirror the gate here. The `kind` parameter (`dictionary`, `font`,
`texture`) lines up with `SyncCategory` names verbatim, so a single
`isSyncCategoryEnabled(kind)` check at the top of the function is all
it takes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Make the dictionary import path usable on large bundles and bring the
multi-device flow up to par.
Import perf
- Skip `MDX.create()` at import time. The factory triggers full init —
decompresses every key block and sorts millions of keys with
localeCompare just to expose the header. Replace with a tiny
`readMdxHeader()` that only reads the small XML header for Title /
Encoding / Encrypted (saves ~17 s on a 250 MB MDX on web).
- `partialMD5`: read the 9 sample slices in parallel rather than
sequentially. Each freshly-picked-File slice round-trip on Chrome
costs ~100 ms cold; parallelisation collapses 9 of them into one.
- Native fast-path in `nativeAppService.writeFile`: when the source is
a `NativeFile`, delegate to Tauri's `copyFile` rather than streaming
the file through `NativeFile.stream()`. Streaming a 250 MB body
through 1 MB IPC chunks on Android took ~100 s; native copy is bound
by disk throughput instead. Exposes `NativeFile.getNativeLocation()`
for the FS layer to use the underlying path + baseDir directly.
Lazy lookup
- Pass `lazy: true` to `MDX.create` / `MDD.create`. The js-mdict change
in this PR skips the upfront decompress-every-block + sort during
init (~80 s on the same 250 MB bundle) and decodes only the relevant
key block on demand per lookup. First-lookup main-thread block
drops from ~81 s to ~230 ms. (Closes #4228.)
Raw .dict
- Drop the import-time gate that flagged non-gzip dict bodies as
`unsupported`. The runtime body loader (`loadDictBody`) already
probes the gzip header and falls through to a passthrough buffer
for raw files, so the gate was the only thing preventing raw
`.dict` bundles from importing on devices that received them via
cloud sync. (Closes#4179, partially addresses #4248.)
Import-flow UX
- `handleImport` now always surfaces a toast for every non-cancelled
attempt: picker errors, missing app service, no-op imports, and
unsupported-but-imported bundles each get their own message instead
of failing silently.
- Call `markAvailableByContentId(newDict.contentId)` after
add/replace so the "Bundle is missing on this device" warning
clears immediately — no need to close-and-reopen the panel.
System Dictionary
- Drop the cascading toggle behavior in `setEnabled`. Each provider's
enabled flag persists independently; exclusivity is enforced at
lookup time. Toggling System on/off no longer wipes the user's
preferred set of in-app providers.
- Render non-system rows as read-only when System is on (toggle still
shows what's queued to restore; tooltip explains the lock).
- `isSystemDictionaryEnabled` short-circuits to `false` on platforms
where the handoff isn't implemented. `providerEnabled` is whole-
field synced across devices, so a flag set on macOS would otherwise
leak to a Windows device with no way to look up a word.
js-mdict
- Submodule bump to e6dbc99 which adds the opt-in `lazy: true`
`MDictOptions` flag (skip `_readKeyBlocks` + post-init sort; new
`lookupKeyBlockByWordLazy` path on `MDX` and `MDD`). Eager mode is
unchanged and every existing js-mdict test still passes.
i18n
- 208 new translations across 33 locales for the new UX strings.
Closes#4228Closes#4248Closes#4179
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(library): make bookitem-main shrink to match cover in fit mode
Closes#4234. In fit mode the bookitem-main kept its 28/41 aspect
regardless of the cover image's natural aspect, leaving extra padding
beside (portrait covers) or above (landscape covers) the cover. The
select-mode overlay and icons drifted away from the cover edge.
BookCover now reports the loaded image's natural aspect ratio. BookItem
overrides the bookitem-main's aspect-ratio with the cover's aspect so
the box hugs the cover exactly, and proportionally shrinks book-item
width for portrait covers so the info row icons align with the cover's
right edge.
Also wrap the TTS "Back to TTS Location" pill with whitespace-nowrap so
long translations (e.g. German "Zurück zur TTS-Position") expand the
button width instead of overflowing the fixed height.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(library): scope cover shrink to bookitem-main, leave info row at cell width
Per review, the width shrink should only apply to the cover row so the
title and info icons keep their original cell-wide layout. Move the
width style from .book-item to .bookitem-main alongside its aspectRatio
override.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Closes#3914. Switch the page-fullscreen image to object-fit: contain
(and SVG preserveAspectRatio meet) so cover images fit the page and
center without cropping. Also adds a duokan-image-gallery-cell layout
helper and drops the mobile marginBottomPx override.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Prevents 'Please log in to continue' error when importing fonts or images
locally without logging in. The replica upload is now gated on
getAccessToken() so unauthenticated users are never enqueued.
Non-PDF fixed-layout EPUBs visually scale the iframe via CSS transform
while keeping native dimensions inside, so getClientRects() returns
unscaled positions. The SVG overlayer was sized in CSS pixels with no
viewBox, leaving annotations and TTS highlights drawn at scale-1
displacement from the actual text.
Bumps foliate-js to set a matching viewBox on the overlayer SVG, and
adds a regression test covering spread/scroll/PDF/empty frames.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(library): in-place import with cloud sync and symmetric local delete
Adds an `inPlace` option to importBook so a source file inside a
registered external library folder is referenced directly via
`book.filePath` instead of being copied into Books/<hash>/. Sidecars
(cover, config, nav) still live under Books/<hash>/.
ingestService routes through shouldImportInPlace, which marks an
import in-place when the absolute source path lives under any of
`settings.externalLibraryFolders` and is NOT inside a per-root
`Books/` subtree. The Readest data dir (`customRootDir`) is
intentionally excluded — that directory is Readest's home and
should freely hold hash copies; in-place is for user-registered
roots (Duokan, Calibre, Moon+ Reader, an iCloud mirror, …).
Cloud sync treats in-place books as first-class:
- uploadBook reads bytes from (book.filePath, 'None') when set.
The cloud key is unchanged, so a peer downloading the book
lands it under Books/<hash>/ as a normal hash copy.
- useBooksSync strips `book.filePath` before pushing — it is a
device-local path that is meaningless on any other device.
- ingestService no longer skips upload for in-place books;
autoUpload / forceUpload behave like any other book. Only
transient imports opt out.
- deleteBook 'local'/'both' now physically removes the source
file at book.filePath (base 'None'). Local-delete semantics
are symmetric with hash-copy books: the local copy is gone,
the cloud backup remains, a future pull restores under
Books/<hash>/. removeFile errors are swallowed.
New `SystemSettings.externalLibraryFolders?: string[]` (no UI yet;
registration entrypoint lands in a follow-up). Added to
BACKUP_SETTINGS_BLACKLIST alongside `localBooksDir` /
`customRootDir` so device-local paths don't ride cloud backups.
Tests: cloud-service, ingest-service, and backup-settings suites
cover in-place delete, multi-root matching, per-root `Books/`
guard, and the backup-strip.
* feat(library): one-tap "read in place" toggle in folder import
Surface the in-place / copy choice as a single "Read books in place (don't copy)" checkbox in the Import-from-Folder dialog. When the user opts in, the chosen directory is registered in `settings.externalLibraryFolders` and ingestService's `shouldImportInPlace` will route the books straight to importBook with `inPlace: true` — no copy into Books/<hash>/, sync still works, local delete still removes the source file (the symmetry was set up in the previous in-place commit).
User experience:
- First-time users hit the toggle once per library folder. The choice is also persisted to localStorage so subsequent dialog opens default to whatever they picked last.
- Repeat imports from a folder that's already registered as an external library folder force the toggle ON and disable it, with a help line explaining that imports from this folder are always in-place. The check is exact-string (after path normalization) so registering /Users/me/Duokan only locks the toggle for that exact path — picking /Users/me/Downloads after Duokan still shows the toggle in its normal state.
- URL-ingress / drag-drop replays go through `runFolderImport` without the dialog and default `readInPlace: false`. They still benefit from in-place automatically when the dropped path lives under an already-registered root, because that decision is made by `shouldImportInPlace` based on settings, not by the dialog flag.
Mechanics:
- ImportFromFolderResult gains `readInPlace: boolean`. ImportFromFolderDialog gains an `initialReadInPlace` prop (seeded from the new `readest:lastImportFolderReadInPlace` localStorage key) and an `isRegisteredExternalRoot` predicate it uses to render the locked / unlocked toggle.
- runFolderImport calls a new `registerExternalLibraryFolder` helper that appends the chosen directory to `settings.externalLibraryFolders` and persists settings, but only when `result.readInPlace` is true. `isRegisteredExternalRoot` does the inverse lookup the dialog needs. Both helpers normalize paths the same way `shouldImportInPlace` does so the predicate matches the ingest layer.
- The new feature has no effect for users who never flip the toggle: `externalLibraryFolders` stays empty, the path-prefix check in `shouldImportInPlace` returns false for every import, and books continue to be copied into Books/<hash>/ exactly as before.
Self-healing for externally-removed in-place books:
Once the dialog lets users opt their library into in-place mode, the source file becomes a piece of state Readest doesn't control — another app may rewrite it (e.g. Duokan persisting reading progress into the epub), the user may move it in Finder, or an external drive may unmount between sessions. Previously, clicking such a book would navigate into the reader, fail inside loadBookContent's `fs.openFile(book.filePath, 'None')` with a low-level IO error, flash an "Unable to open book" toast, and auto-bounce back to the library — leaving the stale library record in place so the next tap reproduces the same dance.
BookshelfItem.handleBookClick now probes availability before navigating, but only for purely-local in-place books (`book.filePath && !book.uploadedAt && !book.deletedAt`). If `appService.isBookAvailable` returns false — which for in-place books means the recorded `book.filePath` no longer exists at the OS level — we dispatch `delete-books` for that hash and show an info toast explaining the removal, instead of opening the reader.
Scope is intentionally narrow:
- Cloud-synced books still flow through `makeBookAvailable`'s on-demand download path; missing local copies trigger a re-download, not a deletion.
- Hash-copy books (no `filePath` set) are not probed: a missing Books/<hash>/ file under normal use signals a bug or filesystem corruption, not user intent, and silently dropping the record would hide the real problem.
- The dispatched delete-books event reuses the existing Bookshelf deletion path, so sidecar metadata and selection state are cleaned up the same way as a user-initiated delete. For in-place books that path doesn't touch any file outside Books/<hash>/, so the now-missing source location (or whatever the user did with it externally) is left alone — symmetric with 165f15a6.
* fix(library): centralize book content resolution
---------
Co-authored-by: Huang Xin <chrox.huang@gmail.com>
* fix(opds): show 'Open & Read' for publications already in the library
PublicationView only flipped the acquisition button to 'Open & Read'
when the user downloaded the book within the current component
lifecycle — the downloadedBook state started as null on every mount
and was never reconciled against the actual library. So reopening the
detail page (after navigating away in the OPDS browser, or coming back
from the reader) always showed 'Download' / 'Open Access' again and
asked the user to re-download a book they already had.
Two real-world feeds (m.gutenberg.org and ManyBooks) exposed two
distinct failure modes that made naive title/author equality unusable:
(1) Gutenberg's OPDS entries never carry <dc:identifier>. The only work
identifier lives in <atom:id>urn:gutenberg:1342:2</atom:id>, which
foliate-js's opds.js parses into metadata.id — a field
getMetadataHashInfo never reads. So the identifier candidate list
was empty and identifier-overlap matching silently skipped every
book.
(2) Author strings disagree across the boundary: the feed emits
'<author><name>Austen, Jane</name>' (Lastname-first) while the EPUB
inside the same feed ships <dc:creator>Jane Austen</dc:creator>.
Plain string equality (even after normalization) never matched.
Add findExistingBookForPublication in app/opds/utils/findExistingBook.ts
with a layered matcher:
- Pass 1: full metaHash equality (the strongest signal — same fingerprint
the bookService uses internally).
- Pass 2: identifier overlap. collectPublicationIdentifiers() splices
metadata.id into the candidate list alongside metadata.identifier, so
Gutenberg's 'urn:gutenberg:1342:2' feeds into identifierKeys(), which
extracts the '1342' digit-tail (>=3 digits, so the trailing ':2'
version suffix is ignored) and matches the EPUB's
'http://www.gutenberg.org/1342' → '1342'.
- Pass 3: tolerant title + author match. hasAuthorOverlap() does a
token-set fallback: each name is split on whitespace/comma/semicolon,
single-letter tokens (initials) and year-range tokens ('1775-1817')
are dropped, and we require >=2 shared tokens by default. So
'Austen, Jane' ↔ 'Jane Austen' match via {austen, jane} but
'Author A' ↔ 'Author B' don't (both collapse to {author} after
dropping single-letter tokens — we track raw token count to refuse
the single-token shortcut when discriminative parts were filtered).
Genuine mononyms still match on a single shared token because raw
count is 1 on both sides.
OPDSPerson[] is fed through the library's own formatAuthors so the
produced author string matches Book.author byte-for-byte (Intl.ListFormat
output, locale-aware). Soft-deleted books are skipped so removing a copy
locally reverts the button.
Wire it in opds/page.tsx by subscribing to useLibraryStore.library
(rather than the snapshot reads inside handleDownload) and memoizing
existingBookForPublication. Pass it to PublicationView, which now
seeds downloadedBook from the prop and resyncs when the prop changes
(switching publications inside the browser) — but does not clobber a
download success it observed locally before the parent recomputed.
Covered by unit tests for the helper, including the real Gutenberg
URN-in-atom-id case, ':version' suffix on the URN, 'Lastname, Firstname'
vs 'Firstname Lastname', year-range stripping, and the 'Author A' vs
'Author B' non-match.
* fix(opds): dedupe downloads by source url
---------
Co-authored-by: Huang Xin <chrox.huang@gmail.com>
* feat(reedy): Appendix A · Phase 2.1 — ChatModel + model registry
First piece of the deferred agent runtime, building alongside the shipped
Phase 1 MVP per the locked decision (no rip-and-replace; legacy + Reedy
MVP + agent paths coexist).
ChatModel is a thin interface over Vercel SDK's LanguageModel that
surfaces contextWindow / reservedOutput / supportsTools — the metadata
the M2.5 PromptContextBuilder and M2.6 AgentRuntime need for prompt
budgeting and tool-calling decisions.
createReedyModels(settings) returns the active (chat, embedding) pair from
the user's currently configured AIProvider, rewrapped in the Reedy
interfaces without touching the legacy AIProvider. CONTEXT_WINDOW_TABLE
hardcodes per-model metadata (Gemini 2.5 → 2M, GPT-5/Llama-4 → 128K,
local Ollama llama → 4K with tools off, etc.) with a conservative 8K
fallback for unknown ids.
No runtime wired yet — Phase 2.6 consumes this.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(reedy): Appendix A · Phases 2.2–2.3 — runtime event/error taxonomy + ToolRegistry
Phase 2.2 — pure types and factories for everything AgentRuntime.runTurn()
will yield:
turn_start | text_delta | tool_call | tool_result(ok/err)
| citation | memory_write | step_finish | usage | error | abort | done
ReedyError covers turn-level failures (context_overflow, turn_timeout,
model_error, provider_unavailable, stream_parse, abort, ...) with default
retryability per kind so the M2.6 streamLoop's retry-with-backoff can
decide without string-matching. ReedyToolError covers tool-execution
failures with a narrower kind union so the streamLoop can re-emit them as
{ tool_result, ok: false, error } events without losing structure.
Phase 2.3 — ToolRegistry that holds the runtime's tool catalog and adapts
each tool to the Vercel ai-SDK ToolSet streamText consumes. Every
registered ReedyTool gets wrapped with:
- Zod input validation → tool_invalid_args on mismatch
- Permission gate → tool_permission_denied (read auto-approves)
- Per-call wall-clock → tool_timeout (default 10s)
- AbortSignal propagation → tool_aborted on turn cancel
- Per-tool serialization → parallelSafe=false tools queue
The local AbortController per call composes the turn-level signal with
the per-tool timeout so we classify timeout-vs-abort correctly based on
which controller fired first, and listener cleanup avoids leaks.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(reedy): wire RetrievalBackend interface + metrics into the chat adapter
Phase 1B backend integration. Adds a RetrievalBackend interface so
TauriChatAdapter holds a uniform reference instead of branching across the
file, with two impls — LegacyIdbBackend (wraps the existing IDB ragService
unchanged) and ReedyBackend (lazy-opens reedy.db, adapts the active
provider's embedding model to Reedy's narrower shape, exposes a Vercel
`lookupPassage` tool). selectBackend() gates Reedy behind both
aiSettings.reedy.enabled AND isTauriAppPlatform() per plan D15 so the MVP
cohort is desktop-only.
The Reedy path streams via streamText({ tools: { lookupPassage }, stopWhen:
stepCountIs(3) }) with a status-aware system prompt that tells the model
how to phrase responses for each RetrieverStatus value.
Replaces the module-global `lastSources` + 500ms poll with a per-instance
ReedySourceStore keyed by a synthetic per-turn id the adapter generates,
so the Sources dropdown stops racing on global state. Both legacy and
Reedy backends now feed citations through the same store; the UI is
backend-agnostic via a shared SourceItem shape both ScoredChunk and
RetrievedChunk satisfy.
Adds the reedy_metrics table to the reedy migration (versioned with
app_version + session_id + turn_id per row) plus a ReedyMetrics writer
that ReedyBackend uses to record indexing-lifecycle and tool-use events.
Always-on local; no network egress. NoopReedyMetrics keeps construction
cheap before the DB opens.
Tests: retrievalBackend selectBackend gates, ReedySourceStore semantics
(append/replace/subscribe/clear), ReedyMetrics debounced batching +
exportBundle, and a TauriChatAdapter contract test that asserts the
Reedy/legacy code-paths pass the right args to streamText.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(reedy): UI wiring — settings toggle, clickable sources, feedback bundle
Phase 1B UI integration. AIAssistant now constructs the active backend
(legacy or Reedy) via selectBackend with the platform gate from M1.7,
wires a ReedySourceStore for the chat adapter, and routes Sources-dropdown
clicks to `getView(bookKey)?.goTo(source.cfi)` when the source has a CFI.
Legacy-path sources still render as static rows because they have no CFI.
AIPanel grows a 'Reedy Retrieval (Beta)' BoxedList with the toggle and a
'Send Reedy feedback' button that calls exportReedyMetricsBundle and
triggers a JSON download of the last 90 days of events. The toggle is
disabled on web with an explanatory description per plan D15.
Thread accepts an onSourceClick callback and renders each source as a
button when its source carries a CFI, otherwise as a static div — so the
Sources dropdown is backend-agnostic via the shared SourceItem shape.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(db): add reedy schema migration with Tantivy FTS + lazy embeddings
Registers a new `reedy` migration set bound to reedy.db. Creates
reedy_book_meta + reedy_book_chunks with a Tantivy FTS index on
chunks.text (ngram tokenizer) and a per-book position index used by
BookRetriever. The vector embeddings table is intentionally NOT created
here — the indexer creates it lazily on first index so the vector32(<dim>)
column matches the active embedding model. Tests cover the migration
applies cleanly, is idempotent, and that the FTS index is queryable.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(reedy): retrieval primitives — DB, chunker, indexer, retriever, lookupPassage tool
Wires the MVP retrieval pipeline behind reedy.db, all under src/services/reedy/
and with no integration into the existing AI module yet (Phase 1B will do that).
- ReedyDb wrapper over DatabaseService: book-meta CRUD, lazy embeddings
table at the active model's dim, bulk chunk + embedding writes via batch(),
hybridSearch (brute-force cosine + Tantivy FTS + reciprocal-rank fusion
with 3× per-path over-fetch), per-book and global wipe. Internal write
queue serializes batch() calls so Turso's single-writer transaction guard
doesn't trip when BookIndexer runs across books in parallel.
- CfiChunker: TreeWalker over the section's DOM, ~maxChunkSize windows with
paragraph > sentence > word break-points, full epubcfi(/6/N!/…) anchors,
round-trip verified via CFI.toRange before each chunk lands.
- BookIndexer: per-book mutex, lazy embeddings-table creation, model.batchSize
embedding batches with dim assertion, terminal status transitions
(indexed | empty_index | failed). Re-indexing clears prior chunks via a new
ReedyDb.clearBookChunks helper.
- BookRetriever: status-typed results (ok | not_indexed | empty_index |
stale_index | degraded). Embedding has a 5s wall-clock budget; on timeout
it falls through to FTS-only with status=degraded.
- lookupPassage Vercel ai-SDK tool: Zod-validated query/topK, per-turn
composite-key dedupe, parallel-call serialization, 10s per-turn budget,
6000-char result clamp, status-with-hint passthrough, and a separate
serializeForModel that wraps each passage in <retrieved trust="untrusted">
with XML-escaped content so book text cannot escape the envelope.
59 unit tests; pnpm test + pnpm lint clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(db): forward DatabaseOpts to tauri-plugin-turso
NativeDatabaseService.open ignored its opts parameter, dropping any
experimental feature flags (e.g. 'index_method' needed for FTS / vector
indexes) and any encryption config before they could reach the Tauri
plugin. Translate DatabaseOpts to the plugin's LoadOptions shape and
forward as the single argument Database.load accepts. Skip translation
when no relevant opts are set so the existing path-string call shape is
preserved for callers without experimental/encryption needs.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(db): cover Turso vector primitives + add benchmark harness
Verifies the Turso functions Reedy retrieval depends on, with a
brute-force per-book kNN test that runs against every DatabaseService
backend (node, native, WASM):
SELECT vector_distance_cos(embedding, vector32(?)) AS d
FROM book_chunks
WHERE book_hash = ?
ORDER BY d ASC LIMIT k
This is the path Turso's own founder recommended in
tursodatabase/turso#3778 ("First, focus on efficient SIMD-accelerated
brute-force search") and what shipped at commit 1aba105df4f. Native
vector index modules don't exist in this engine: `libsql_vector_idx`,
`vector_top_k`, and `USING vector/hnsw/diskann/ivfflat` all parse-error
against @tursodatabase/database@0.6.0-pre.28 (libsql_vector_idx is a
libSQL/sqld fork feature; DiskANN was closed not-planned upstream in
#832). The test asserts cross-book isolation and nearest-first ordering
using only `WHERE book_hash = ?` and `ORDER BY` — no DDL, no identifier
interpolation, no index plumbing.
Also adds bench/ harness for manual perf checks:
pnpm bench [name] run benchmarks (refuses in CI)
pnpm bench --list list available benchmarks
pnpm bench --no-record skip results.jsonl append
pnpm bench --force override the CI guard
Uses Node 24's --experimental-strip-types so no tsx devDep is needed.
Appends one JSON line per run to bench/results.jsonl (gitignored, local
history; share by pasting tabular stdout into PRs/issues). Explicitly
NOT in CI — shared-tenant variance makes synthetic-benchmark regression
detection unreliable; production telemetry (reedy_metrics, plan §M1.9)
is the right tool for that.
First benchmark: vector-retrieval. Measured on M1 Pro:
400 chunks × 384 dim → 0.35 ms / query
400 chunks × 768 dim → 0.45 ms / query
2000 chunks × 768 dim → 2.23 ms / query
10000 chunks × 768 dim → 14.00 ms / query
400 chunks × 1536 dim → 0.70 ms / query
Per-chunk cost ~1.1 µs at 768 dim = ~1.4 ns/dim. NEON-class on
Apple Silicon, ~50× faster than scalar — confirms SIMD acceleration is
active in 0.6.0-pre.28. Per-query latency stays sub-ms at Reedy MVP
corpus sizes; the ceiling is ~10K chunks per book before phone-class
hardware notices.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* OpenRouter: new OpenAI-compatible provider with chat + embedding model
routing, health check via /models, and a fetchOpenRouterModels() helper
for the settings UI. API key, base URL and model fields are persisted in
AISettings, surfaced in AIPanel, indexed by commandRegistry, and added
to backupService's credential allow-list so the key round-trips through
encrypted backups.
* utils/httpFetch: introduce getAIFetch() as the single decision point for
outbound AI traffic. In Tauri it returns @tauri-apps/plugin-http's fetch
(Rust/reqwest transport, no renderer CORS preflight, no Android cleartext
block); on the web build it falls back to window.fetch. OllamaProvider
is migrated end-to-end — both ai-sdk-ollama streaming and the /api/tags
health probe — and the new OpenRouterProvider uses the same path, so any
future provider only has to call getAIFetch().
* Tests: unit tests for OpenRouter provider behavior (model selection,
availability, health check) and a backup-settings round-trip test
ensuring openrouterApiKey is treated as a credential field.
Fixes#4254. On app boot, Providers called applyBackgroundTexture
immediately after loadSettings() resolved, but the customTextureStore
was still empty — loadCustomTextures only ran later when ColorPanel
mounted or useReplicaPull seeded it during library render. The hook's
addTexture fallback re-derives the texture id from name and creates a
new entry with a different id whenever the saved id wasn't computed
from the current name (legacy imports, cross-device sync), so
applyTexture silently bailed out and no texture was mounted.
Seed customTextureStore.setTextures(settings.customTextures) in
Providers right after loadSettings() resolves — preserving the saved
ids — so applyTexture can resolve them at boot. Only custom textures
were affected; predefined textures (concrete, paper, etc.) worked
already because the lookup falls back to PREDEFINED_TEXTURES.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(reader): restore right-column clicks and selection in dual-page mode
Bumps foliate-js to readest/foliate-js@1ea2996, which stops marking
visible non-primary views as `inert` / `aria-hidden`. Adds a regression
test for the underlying visibility helper.
When a dual-page spread crosses a section boundary, the right column
lives in a non-primary view. The previous a11y sync set both `inert`
and `aria-hidden` on that view — `inert` blocked link clicks and text
selection in the visible column, and `aria-hidden` hid it from
assistive tech while sighted users could still read it. The fix drops
`inert` entirely (screen-reader swipe-next is already handled by
`aria-hidden`) and applies `aria-hidden` only to views whose bounding
rect lies outside the visible container.
Fixes#4243 (right-column links unclickable in dual-page mode).
Fixes#4259 (text selection fails when an image section is on the left).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(reader): update paginator-multiview a11y assertions for new visibility-based aria-hidden
Browser tests previously assumed every non-primary view carried both
`inert` and `aria-hidden`. The fix in the preceding commit drops `inert`
entirely and only `aria-hidden`s views that are off-screen. Update the
two assertions to:
- check that no wrapper ever has `inert`;
- require `aria-hidden="true"` only when the wrapper's bounding rect is
fully outside the visible container, and require its absence when the
wrapper overlaps the viewport (the regression case from #4243 / #4259).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(cbz,i18n): ComicInfo metadata + CBZ page count + WebDAV i18n
Closes#4253 (ComicInfo.xml not read) and #4255 (CBZ shows "1 page left").
CBZ / ComicInfo (foliate-js submodule + Readest derivation):
- comic-book.js: find ComicInfo.xml in subdirectories too, parse
description / subject / identifier / published / series fields
beyond the prior name+position pair. Series Count populates the
canonical `belongsTo.series.total`; no top-level duplication.
- bookService.ts / readerStore.ts: derive `metadata.seriesTotal`
from `belongsTo.series.total` in parallel to the existing
series / seriesIndex derivation.
- ProgressBar / FooterBar / DesktopFooterBar: drop the hard-coded
`pagesLeft = 1` for fixed-layout books and compute it from
`section.total - section.current`. FooterBar uses
`FIXED_LAYOUT_FORMATS.has(bookFormat)` so CBZ picks `section`
(correct image count) instead of `pageinfo` (locations).
- ProgressBar: switch the remaining-pages text to "in book" for
fixed-layout titles (no chapter structure) and keep
"in chapter" for reflowable books.
WebDAV refactor for translation coverage:
- WebDAVBrowsePane / SyncHistoryPanel called `t(...)` (passed as a
prop) instead of `_(...)`. The i18next-scanner only looks for
`_`, so ~53 strings were unreachable and shipped in English to
every locale. Switched both components to call
`useTranslation()` themselves; helpers that aren't React FCs
take `_: TranslationFunc` so the scanner sees the literal calls.
- WebDAVClient.checkConnection now returns a `code` discriminator
(`SERVER_URL_REQUIRED` / `AUTH_FAILED` / `ROOT_NOT_FOUND` /
`UNEXPECTED_STATUS` / `NETWORK`); raw English `message` is
reserved for the dev console. New `formatConnectError` and
`formatSyncError` helpers in WebDAVForm translate via a switch
where each branch is a literal `_('...')`. Same treatment for
the sync-failure path that previously surfaced raw e.message.
- "Syncing 0 / {{total}}" is now parameterized as
"Syncing {{n}} / {{total}}" with n=0 at startup so the digit
formats naturally and the template can be reused mid-sync.
- "Cleanup · {{count}} book(s)" hard-coded options used unsupported
ternary; rewrote as plural-aware key.
i18n scanner fix (i18next-scanner.config.cjs):
- vinyl-fs walked into directories whose names end in source-file
extensions (Next.js route folder `runtime-config.js/`, Playwright
screenshot folder `*.test.tsx/`) and crashed with EISDIR.
Resolved by expanding globs via `fs.globSync` and filtering to
files only before handing to the scanner.
TypeScript-syntax sites that broke esprima during extraction:
- WebDAVBrowsePane / WebDAVForm: `(e as Error).message` and
`failed[0]!.title` inside `_(..., options)` arguments. Replaced
with `e instanceof Error ? e.message : String(e)` and
`failed[0]?.title ?? ''` — also runtime-safer.
User-facing em-dash cleanup:
- Removed em-dashes from translation keys across SyncHistoryPanel /
WebDAVForm / WebDAVBrowsePane / SyncPassphraseSection / send/page /
replicaCryptoMiddleware / AIPanel. Tagline in `layout.tsx` kept.
Locale translations:
- ~2400 translations applied across all 33 locales for the keys
that were either newly extractable, freshly worded, or
pre-existing but untranslated. Zero `__STRING_NOT_TRANSLATED__`
remain after the run.
Misc:
- next.config.mjs: drop `eslint.ignoreDuringBuilds: true` so build
runs the same lint as CI.
- Collection type: add `total?: string` for ComicInfo series count.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* ci(test): fix vitest invocation, run with 4 workers
`pnpm test:pr:web` was chaining `pnpm test -- --watch=false`, which
pnpm expanded into:
dotenv -e .env -e .env.test.local -- vitest -- --watch=false
The second `--` made vitest treat `--watch=false` as a positional
file pattern, not a flag. Vitest then fell back to defaults (in CI's
non-TTY env that still meant a one-shot run, so the suite passed),
but the worker pool was effectively serialized for big chunks of the
243-file run — wall ~90 s on a 4-vCPU runner where the parallel-sum
of phases was ~236 s (≈2.6× effective parallelism).
Replace the chained pnpm invocation with a direct call to
`vitest run --maxWorkers=4`, matching the 4 vCPUs the GH Actions
ubuntu-latest runner provides.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Email-in (`<user>@readest.com`) is now a paid feature. The other
Send channels — in-app /send page, mobile share-sheet, browser
extension — stay open to free users.
Three enforcement layers:
- `pages/api/send/address.ts` and `pages/api/send/senders.ts` return
403 with `{ code: 'plan_required', plan, requiredPlans }` for free
users. No `send_addresses` row is allocated on the blocked path.
`pages/api/send/inbox.ts` and `pages/api/send/inbox/file.ts` are
deliberately left open — they're shared with the file-upload and
extension channels.
- `workers/send-email` looks up `plans.plan` after resolving the
recipient and bounces (not silently drops) inbound mail for free
users with a one-sentence message pointing to upgrade plus the free
clip channels. Bounce rather than drop so a downgraded user
understands why their mail stops landing.
- `components/settings/integrations/SendToReadestForm.tsx` reads the
user's plan from the JWT before any API call. Free users see one
friendly card — headline, value prop, "View plans" CTA → /user, and
a softer line about the free alternatives — instead of address /
senders / activity sections of disabled controls. The
IntegrationsPanel NavigationRow stays visible so users can discover
the feature.
Single source of truth for the entitled tier set: `EMAIL_IN_PLANS` +
`isEmailInPlan(plan)` in `src/utils/access.ts`. Mirror copies live in
the Worker (no shared import surface) — keep them in sync.
Edge cases:
- Downgraded user: existing `send_addresses` row stays. All three
layers block; re-upgrading silently restores the same address.
- Loading flicker: `userPlan` starts as `null` so the loading skeleton
stays up rather than briefly flashing the upgrade card for a paid
user on a slow client.
12 new unit tests cover the gate on `/api/send/address` and
`/api/send/senders` (GET + POST blocked for free users, no Supabase
access on the blocked path, allowed for plus / pro / purchase).
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Huang Xin