Conversation
Add design doc for a File System Access API backed sync provider. Feature-gated on Chromium browsers; handle persisted in a dedicated IndexedDB database owned by the provider module.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add filesystemCore stub to cloud-provider-core-registry (never called at runtime since FilesystemProvider.supportsWorkerDownload = false) - Widen VolumeMetadata.cloudProvider from narrow literal union to ProviderType, fixing cascading errors in cloud-fields.ts, library-placeholders.ts, and download-queue.ts Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
processDownload previously only handled supportsWorkerDownload=true. For the filesystem provider (which sets that to false), downloads were silently skipped — the queue item stayed in 'downloading' state forever and blocked all subsequent items. Add a main-thread branch that downloads via the provider's downloadFile method, extracts the CBZ with @zip.js/zip.js, and reuses the existing downloadSidecarEntries + processVolumeData pipeline.
The regex on error.message did not match Chrome's DOMException.message (which does not contain the string 'NotFoundError'), causing spurious DELETE_FAILED errors when a series folder was already gone during sync.
…ders The filesystem provider cannot upload from a worker because a FileSystemDirectoryHandle is bound to the window that received it via showDirectoryPicker — structured-cloning it to a worker does not carry the user-granted permission. Introduce a supportsWorkerUpload flag on SyncProvider (parallel to the existing supportsWorkerDownload) and, when a provider opts out, compress + upload the archive and any sidecars on the main thread via provider.uploadFile. Worker-capable providers (Google Drive, MEGA, WebDAV) are unaffected.
Previous fix bypassed the pool entirely and ran compression + upload on the main thread for filesystem. With a parallel queue that meant N volumes could load, compress, and buffer simultaneously with no shared-memory accounting and no provider concurrency cap. Instead, submit the same pool task used for real providers but in the worker's 'null provider' compress-only mode. The worker still does the CPU-heavy compression (throttled by WorkerPool + sharedMemoryManager); the main-thread onComplete then performs the single filesystem write via provider.uploadFile while the task's memory reservation is still held. Sidecars follow the same path. Worker-upload providers (Drive, MEGA, WebDAV) are unchanged.
Previous fix bypassed the pool and did blob read + decompression + processing on the main thread for filesystem. Even with the queue being serial that blocks the UI during decompression and leaves memory untracked by the shared memory manager. Instead, use the worker's existing decompress-only mode: prepareData reads the blob on the main thread (required — FileSystemDirectoryHandle is bound to the window) and passes it to the worker, which decompresses and returns entries. Memory reservation, provider concurrency gating, and shared memory tracking all apply uniformly with the existing worker-download path.
Adds design for a fourth real sync provider using Microsoft Graph API via MSAL.js. Multi-tenant (common) auth covers personal + work/school accounts. Worker-capable core follows the Drive/WebDAV pattern so WorkerPool throttling applies uniformly.
The Svelte 5 rewrite of QuickActions (replacing Flowbite's SpeedDial) dropped the background/shadow on the toggle button, leaving only a bare plus icon that disappears against light backgrounds. Fixes #249
feat: OneDrive and Local Folder sync providers + Google Drive re-auth overhaul
Rendered text must never overlap — whatever the quads claim, two visible lines painting the same pixels is a wrong layout. OPM 28 exposed two ways sub-threshold overlaps slipped through the re-capture gate (0.7): - p136: the detector re-captured one caption column as two half-width- offset quads with diverged hallucinated texts (bbox overlap 0.41), so both rendered stacked on the same ink. - p176: a wide slanted quad — detector noise, the print is upright — has an axis-aligned bbox that swallows the neighboring columns, and the merged-column wrap then painted across them. New final pass in layoutLines resolves every residual rendered-rect overlap. Clean single-column lines hold their ground; suspect/wrapped/ banded lines are clipped around every clean rect they touch and re-wrap in the widest surviving gap — which recovers the true column between the neighbors a slant-inflated quad claimed. Equal-trust collisions first try nudging apart within their quads' own slack (uniform size preserved; first round only, since nudging claims new ground), then split the contested span at its midpoint. Every clip yields a subset of the previous extent, so resolution never creates a new overlap and converges. Corpus sweep over 2,193 local .mokuro files (3.17M blocks): the old layout rendered overlapping text in 106,988 blocks (3.4%); all are resolved with zero residual overlaps and zero new violations. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
feat(reader): page gap for dual-page spreads + ctrl+shift+scroll (#234)
…n creation A dropped/timed-out response for the final chunk (or a retry exhausted against a transient status) was thrown as a hard failure even when Graph had already finalized the file server-side. Check by path+size before giving up. Also honor the Retry-After header on 429s instead of a fixed backoff, and retry transient failures when creating the upload session (previously only chunk PUTs were retried).
renameFile() overwrote an existing destination with zero conflict check, unlike WebDAV/OneDrive — now throws unless the occupant matches the source's size (idempotent-retry convergence, same rule as WebDAV). renameFolderInner() only renames the syncable files under a folder but then recursive-deleted the entire old folder, destroying any other file or subfolder a user had stored there. Now prunes only if verifiably empty, via the existing removeDirectoryIfEmpty helper.
Merging the export-to-file path with the new main-thread-upload (filesystem provider) path forced embedMokuroInArchive: false for both, regressing plain "export to file" — which previously always embedded OCR data — to the cloud-provider sidecar convention. Only the upload path should use a sidecar; export-to-file keeps the self-contained archive.
A manual "Reconnect" click is simultaneously the pointerdown that satisfies an armed gesture-retry AND the button's own click handler, so both could call requestAccessToken() on the same tick. Wire up the already-present (but previously unused) isRefreshing flag as a guard.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
adrian-tompkins
pushed a commit
to adrian-tompkins/mokuro-reader
that referenced
this pull request
Aug 2, 2026
…detection feat: Add import mismatch detection with detailed error modal
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
1 participant
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
See CHANGELOG.md [1.8.0] for the full user-facing list.