Open
Conversation
Add MultipartSyncParser (streaming MIME parser), MultipartSyncClient (batched API orchestration), SyncRequest model, and syncBookmarks() Retrofit endpoint. Migrate delta-detected metadata reloads to POST /bookmarks/sync with with_json=true, falling back to legacy paginated GET when updatedIds is unavailable. Add Room schema v11 with content_package and content_resource tables (destructive migration for article_content). Update v2 spec document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add download button to save log zip to Downloads via MediaStore API. Wrap log content in SelectionContainer for drag-to-select copy support. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add ContentPackageManager for per-bookmark file storage with atomic staging/commit. Add LoadContentPackageUseCase that fetches full offline packages (JSON+HTML+resources) via multipart sync, with fallback to legacy LoadArticleUseCase. Add ContentPackageDao for Room persistence. Add OfflineContentPathHandler serving local files through WebViewAssetLoader. Update BookmarkDetailWebViews to use per-bookmark offline base URLs and intercept resource requests for local serving. Generate picture wrapper HTML when server HTML is absent. Add androidx.webkit dependency. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Wrap MediaStore.Downloads usage in API 29+ check with pre-Q fallback using public Downloads directory. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Coil's AsyncImage cannot resolve URLs with the synthetic offline host (https://offline.mydeck.local/). Convert these URLs to file:// URIs pointing to the actual local files so images render correctly. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Post-branch item: investigate progressive display of article content instead of spinner-until-ready, especially for image-heavy articles. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Update DAO queries to include Picture bookmarks (type='photo') in content-eligible queries, not just hasArticle=1 - Migrate BatchArticleLoadWorker and DateRangeContentSyncWorker to use LoadContentPackageUseCase (multipart) with fallback to legacy LoadArticleUseCase on transient failure - Add sourceUpdated freshness check: when content is DIRTY but the existing package's sourceUpdated matches the bookmark's current updated timestamp, restore DOWNLOADED state without re-fetching - Add storage usage display and "Clear All Offline Content" button to Sync Settings with confirmation dialog - Add storage management string resources to all 10 language files Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When a picture bookmark has offline content from a content package, the wrapper HTML already contains the image. The toHtml() method was also prepending the metadata imgSrc, causing the image to appear twice. Skip the manual image prepend when articleContent comes from a content package (offlineBaseUrl is set). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The <figure> element used in picture wrapper HTML inherits default browser margins (~40px), making the image and caption narrower than the title/description above. Reset figure margin to 0 and add styling for figcaption and .picture-caption across all four theme templates. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The offline storage size display only refreshed on init and after clearing. Now it also refreshes when: - Date-range download work transitions from running to completed - Content download counts change in the DB (via detailedSyncStatus flow) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add CachedAnnotationEntity Room table and DAO (v12 migration) - Build AnnotationHtmlParser to extract annotation metadata from <rd-annotation> HTML - Populate annotation cache during ContentPackageManager.commitPackage() - Make fetchAnnotations offline-first in BookmarkDetailViewModel - Remove legacy annotation sync for content-package bookmarks (fix color flash) - Update annotation-driven refresh to use multipart re-download - Add comprehensive AnnotationHtmlParser unit tests - Update spec with Stage 3a architecture and test requirements
The Readeck multipart sync endpoint strips all attributes from <rd-annotation> elements, returning bare tags. This broke highlight colors, offline listing, tap-to-edit, scroll-to-annotation, and selection-based edit detection. - Add AnnotationHtmlEnricher to patch bare tags with attributes from the annotations REST API during content package commit - Add AnnotationDto color/note fields (API returns them despite spec) - Use API color as authoritative source in fetchAnnotations - Dismiss annotations sheet before scrolling to highlight - Show snackbar when attempting highlight CRUD while offline - Add offline-annotation-crud-spec for future queued mutation support Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…sist server-provided value; ui: offline placeholder for video embeds; debug: include omit_description in Details dialog
…add FreshnessMarkerUseCase and spec addendum; update workers, parser, content package handling, and detail screen
…, and progress bar - Option D: performFullSync() now persists bookmark metadata from the list endpoint paging response (zero additional network requests) - FullSyncWorker skips loadBookmarksUseCase when full sync already persisted metadata; LoadBookmarksWorker INITIAL trigger uses performFullSync directly - Video bookmarks now participate in automatic/manual/date-range content sync (removed type != 'video' exclusion from DAO queries and LoadContentPackageUseCase) - Preserve embed/embedHostname in insertBookmarks when incoming data has null (list endpoint omits embed; multipart sync includes it) - Add refreshBookmarkMetadata() as fallback for video bookmarks opened before any multipart sync has populated embed - Offline video embed placeholder now matches actual embed host instead of hardcoded YouTube/Vimeo allowlist - Reader loading overlay: indeterminate bar for already-downloaded content (WebView render), no duplicate bar during on-demand fetch (top-level determinate bar covers that) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ress The determinate progress bar was invisible because it was drawn before BookmarkDetailScreen in the Box (earlier children are painted behind later ones). Moved it inside the Scaffold content Box as the last child so it renders on top of the loading overlay, positioned just below the top app bar. Also fixed the bar stalling during the network fetch phase by adding a ticker coroutine that advances progress in small increments (5% every 300ms) while the multipart response is in flight, with a monotonic progress floor to prevent backwards jumps. Additionally, the ReaderLoadingOverlay now only shows the indeterminate bar when content was already cached at screen open — not after an on-demand fetch transitions to WebView rendering. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…and design rationale Add Implementation Status section documenting completion of all four stages plus Stage 3a. Add Retained Legacy Endpoints section explaining why three GET-based endpoints remain (article fallback, creation polling, error flag refresh) with rationale for each. Update Stage 4 rollout description to reflect actual scope. Expand field notes with Option D resolution details and video content package inclusion rationale. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
All content from the addendum (parser fixes, atomic commit, timestamp normalization, worker batching, freshness marking, annotation enrichment) is now reflected in the main spec's Implementation Status and field notes. The remaining follow-ups listed in the addendum are complete. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Move to docs/archive/: - filter-ui-filtered-list-spec.md (v0.11.1, fully shipped) - sync-error-media-follow-up-spec.md (superseded by sync-multipart-adoption-spec) - sync-content-api-omit-description-spec.md (superseded by sync-multipart-adoption-spec) - sync-architecture-optimization-spec.md (superseded by sync-multipart-adoption-spec) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…age 4
Fix BookmarkDto @SerialName mismatch: server returns "embed_domain" but
the DTO was mapped to "embed_hostname", causing embedHostname to always
deserialize as null. The field was undocumented in the OpenAPI spec but
confirmed present in both GET /bookmarks/{id} and POST /bookmarks/sync
responses via direct API testing.
New unit tests covering Stage 4 changes:
- performFullSync persists bookmark metadata from paging response
- performFullSync returns inserted count in SyncResult
- insertBookmarks preserves existing embed when incoming embed is null
- insertBookmarks overwrites embed when incoming embed is non-null
- LoadBookmarksUseCase with null/empty updatedIds returns success
without fetching (no-op guard)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
… with_json The root cause was Kotlinx serialization's encodeDefaults=false (the default) combined with SyncRequest.withJson defaulting to true. When callers set withJson=true it matched the default and was omitted from the serialized JSON, so the server never received the flag and returned no JSON parts in multipart sync responses. Content packages got HTML+resources but no metadata. - Change SyncRequest.withJson default from true to false so explicit true is always serialized - Add encodeDefaults=true to global Json config as belt-and-suspenders - Simplify resolvePersistedOmitDescription to always preserve existing value when incoming is null (list endpoint never sends omit_description) - Add BookmarkDao.updateOmitDescription for direct writes after content package commit, surviving concurrent sync races - Add omitDescription to BookmarkDebugExporter localState output - Add realistic CRLF+Content-Length multipart parser test matching server format - Update BookmarkRepositoryImplTest to match simplified resolve logic Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…d fix Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The legacy LoadArticleUseCase fallback in fetchContentOnDemand was running for all bookmark types. For Picture and Video, this called GET /article which marks hasArticle=false bookmarks as PERMANENT_NO_CONTENT — permanently blocking the content package path from retrying. Restrict the legacy fallback to Article type only. Picture and Video types now report the content package result directly without the destructive side-effect on contentState. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…clients, note visibility Three annotation/highlight issues fixed: 1. Annotation CRUD (create, recolor, delete) no longer triggers a full WebView reload. Color changes use pure JS DOM updates; create/delete use lightweight HTML-only multipart fetch with JS innerHTML replacement. 2. Annotations added via other clients now sync to the reader HTML when opening a bookmark. New syncAnnotationsForContentPackage() compares API annotation snapshot against cached snapshot and triggers HTML-only refresh if changed. 3. Annotations with notes were completely hidden by CSS display:none targeting the data-annotation-note attribute. Replaced with a ::after pencil icon indicator matching the Readeck web UI. New spec: docs/specs/highlight-compose-fix.md Updated: sync-multipart-adoption-spec.md with field notes for this fix, embed_domain SerialName mismatch, and picture/video PERMANENT_NO_CONTENT. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Videos with an embed but no article content (hasArticle=false) were incorrectly routed through fetchContentOnDemand, which marked them PERMANENT_NO_CONTENT. Combined with a race against metadata refresh (which provides the embed data), this caused the auto-switch to View Original on first open. Second open worked because embed was already in Room. Skip fetchContentOnDemand and PERMANENT_NO_CONTENT error state for videos that have an embed fallback. Videos with neither article nor embed still get the correct failure path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Await refreshBookmarkMetadata before evaluating content state for videos with missing embed. Previously the metadata refresh was fire-and-forget, creating a race where contentLoadState was set to Failed (and the UI switched to ORIGINAL mode) before the embed arrived from the API. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: add pre-merge fix specs and correct validation test attribute names - branch-pre-merge-fixes.md: documents two required/recommended fixes found in final code review (unconditional deleteResources() in ViewModel, wrong KDoc in rewriteToRelativeResourceUrls), plus supplementary findings on partial enrichment lock-in and test spec attribute name errors - branch-validation-tests.md: correct attribute names from data-id/data-color to data-annotation-id-value/data-annotation-color; add partial enrichment lock-in as a documented known limitation to test for https://claude.ai/code/session_01LcSEVkLCCLLGue3GCLpzZv * docs: add pre-merge guardrail + unit test implementation specs Detailed implementation instructions for: - Guardrail 1: fix partial enrichment lock-in in AnnotationHtmlEnricher (replace enriched-tag early-exit with bare-tag check) + add 50% mismatch threshold that aborts enrichment and returns original HTML when match quality is too poor - Guardrail 2: add relative-URL safety check to ContentPackageManager .deleteResources() — aborts with warning if HTML still references local resources via relative paths - Unit test specs: AnnotationHtmlEnricherTest (10 cases including lock-in regression), LoadContentPackageUseCaseTest (5 cases for refreshHtmlForAnnotations), ContentPackageManagerTest (5 cases for deleteResources guardrail) - Build verification checklist https://claude.ai/code/session_01LcSEVkLCCLLGue3GCLpzZv * docs: add Readeck API documentation gaps, improvement proposals, and dump script docs/readeck-api/api-documentation-gaps.md Six documented gaps in Readeck API docs discovered during v0.12.0 development: POST /bookmarks/sync request/response format undocumented, rd-annotation attribute stripping undocumented, embed_domain missing from OpenAPI spec, omit_description absent from list endpoint, field coverage differences not documented, GET vs POST /bookmarks/sync relationship not explained. docs/readeck-api/api-improvement-proposals.md Five proposals for improving the sync API for mobile offline reading clients: include annotation attributes in sync HTML, add since filter to POST sync, add omit_description and embed_domain to list endpoint, add error parts to multipart response, document max batch size. docs/readeck-api/readeck-api-dump.py CLI script for dumping and comparing Readeck API endpoint output. Supports legacy article HTML, annotations, single bookmark JSON, GET sync IDs, and POST multipart sync with full binary-to-readable multipart parsing. Includes --compare for diffing runs before/after server changes. https://claude.ai/code/session_01LcSEVkLCCLLGue3GCLpzZv --------- Co-authored-by: Claude <noreply@anthropic.com>
…ess; add TODO guidance Fix 1 proposed replacement now checks updateHtml() return value before calling deleteResources() — updateHtml() can fail on I/O error, leaving old relative-URL HTML on disk. Deleting resources after a failed HTML write reproduces the broken-images bug. Supplementary finding section updated with a recommended TODO comment to add inline in AnnotationHtmlEnricher.kt at the partial-enrichment lock-in site, so the limitation is visible to developers working in that file post-merge. https://claude.ai/code/session_01LcSEVkLCCLLGue3GCLpzZv
Guardrail 1 (AnnotationHtmlEnricher): replace the early-exit guard with a bare-tag-specific check to fix partial-enrichment lock-in, and abort enrichment when fewer than 50% of annotations are matched to avoid silently committing poorly-enriched HTML. Guardrail 2 (ContentPackageManager.deleteResources): abort deletion if the current HTML still contains relative resource URLs (./filename), preventing broken offline images when callers skip the HTML-refresh step. Also covers the BookmarkDetailViewModel Fix 1 caller path that gates deleteResources on a successful updateHtml(). Unit tests added: - AnnotationHtmlEnricherTest: threshold (above/below/boundary), lock-in regression (partially-enriched HTML re-enriched), entity decoding - LoadContentPackageUseCaseTest: legacy endpoint preferred, multipart fallback, URL rewrite applied/skipped based on hasResources, failure path does not overwrite HTML - ContentPackageManagerTest: relative URL aborts deletion, absolute URL allows deletion, no-package/empty-resources no-ops, no HTML file allows deletion Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…versal LoadBookmarksWorker: when delta sync fails or no cursor is available, deltaUpdatedIds is null and loadBookmarksUseCase.execute(null) was a silent no-op, leaving bookmark data stale until the next scheduled full sync. Now falls back to bookmarkRepository.performFullSync() in that case, matching the same error-handling shape as the initial load path. ContentPackageManager.commitPackage: resource files were written using the server-provided path value directly (File(stagingDir, resource.path)) with no bounds check. A malformed or malicious path such as ../../../... could escape the staging directory and write to arbitrary locations within app storage. Now validates canonicalPath starts within stagingDir and skips with a warning if not. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…ess) WIP snapshot. Test-only storage limits (5/10/20 MB) and some tooling remain pending removal after testing is complete. Changes in this commit: - Offline reading regression: guard enqueueArticleDownload in BookmarkRepositoryImpl with isOfflineReadingEnabled() check at both call sites (direct and via pollForBookmarkReady). Adding a bookmark while offline reading is disabled no longer triggers a content download. - Image-only storage measurement: add getTotalImageResourceBytes() DAO query summing content_resource.byteSize for image/* MIME types, and calculateImageSize() on ContentPackageManager. Storage cap enforcement now measures image resources only, not total offline directory size. - Concurrent worker fix: switch ExistingWorkPolicy from KEEP to REPLACE in LoadBookmarksUseCase so a new enqueue cancels any in-flight worker rather than spawning alongside it. Add AtomicBoolean in-flight guard in BatchArticleLoadWorker as a secondary safety net. - Two-watermark hysteresis: enforcement now prunes to 80% of the limit (low watermark) rather than exactly the limit, triggered at 95% (high watermark). Adaptive batch sizing drops from 10 to 2 when image storage is above 90% of the limit (gate threshold). - Limit-hit text-only mode: once the storage cap is hit and pruned during a worker run, subsequent batches call restoreDownloadedStateIfHasContent() instead of re-downloading. Bookmarks with existing text content are restored to DOWNLOADED state via a DB update only (no network). This eliminates the download→prune oscillation and the resulting scattered image pattern in the bookmark list. - Image backfill via worker: reconcileImagePreferenceForManagedContent() no longer calls executeBatch() directly when turning images on. Instead it marks text-only bookmarks as DIRTY and lets BatchArticleLoadWorker handle the backfill, ensuring storage cap enforcement applies. - Sync Status split: offlineStorageSize replaced with separate textStorageSize and imageStorageSize fields, displayed as "Text: X" and "Images: Y" lines. - Test storage limits: MB_5 / MB_10 / MB_20 added to OfflineImageStorageLimit enum and toLabelResource(). English strings only — to be removed before release. - Spec: docs/specs/image-storage-limit-refinements-spec.md added capturing all five issues, root causes, fixes, and a TODO for picture bookmark image storage counting against the cap when Download Images is off. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…tion plan (#133) * Add unified offline content architecture spec and implementation plan Proposes a simplified offline content model that replaces the two-tier (text-only / full) managed offline design with a single full-package-only tier. Covers content state semantics, three management policies (storage limit, newest-N, date-range rolling window), bookmark card icon behaviour, annotation implications, settings layout, migration strategy, and future considerations (collections-based scope, on-demand cache eviction). Implementation plan breaks the work into 13 independently mergeable slices with model recommendations (Haiku / Sonnet / Opus) for each. https://claude.ai/code/session_017MmFgCXCCArrdtLs5B9pcU * Add Material You compliance and user guide requirements to offline spec Notes that the sync settings screen must be brought to full Material You parity (Bookmark Sync section is the reference) and that settings.md and your-bookmarks.md user guides must be updated to cover the new offline reading controls and card icon semantics. https://claude.ai/code/session_017MmFgCXCCArrdtLs5B9pcU --------- Co-authored-by: Claude <noreply@anthropic.com>
…#134) * Add unified offline content architecture spec and implementation plan Proposes a simplified offline content model that replaces the two-tier (text-only / full) managed offline design with a single full-package-only tier. Covers content state semantics, three management policies (storage limit, newest-N, date-range rolling window), bookmark card icon behaviour, annotation implications, settings layout, migration strategy, and future considerations (collections-based scope, on-demand cache eviction). Implementation plan breaks the work into 13 independently mergeable slices with model recommendations (Haiku / Sonnet / Opus) for each. https://claude.ai/code/session_017MmFgCXCCArrdtLs5B9pcU * Add Material You compliance and user guide requirements to offline spec Notes that the sync settings screen must be brought to full Material You parity (Bookmark Sync section is the reference) and that settings.md and your-bookmarks.md user guides must be updated to cover the new offline reading controls and card icon semantics. https://claude.ai/code/session_017MmFgCXCCArrdtLs5B9pcU * Fix paradigm violations: remove per-bookmark content management references Spec: clarify that the Bookmark Detail dialog exposes no per-bookmark content removal action; add the incomplete onRemoveDownloadedContent dead-code cleanup (both ViewModels + full composable callback chain + string resources) to the explicit removal list. Plan: expand Slice 3 to cover the full dead-code cleanup of the per-bookmark removal action; rewrite Slice 11 to remove the incorrect instruction to retain the overflow menu item. https://claude.ai/code/session_017MmFgCXCCArrdtLs5B9pcU --------- Co-authored-by: Claude <noreply@anthropic.com>
Add comprehensive KDoc documentation to clarify DOWNLOADED + hasResources semantics: - BookmarkEntity.ContentState enum: Document two DOWNLOADED sub-states - ContentPackageEntity: Clarify hasResources field meaning - BookmarkListItemEntity: Explain offline availability indicators - ContentPackageManager: Add unified architecture overview Remove orphaned sync_remove_downloaded_content string resources from all language files. Audit identifies areas where contentState checks need hasResources consideration for later slices.
Update content descriptions to reflect unified offline content architecture: - Outline icon: 'Text available' (on-demand text cache) - Filled icon: 'Available offline' (full offline package) Add new string resources to all language files as English placeholders. Icon logic already correctly implemented based on hasResources flag. Verification: - assembleDebugAll: PASSED - testDebugUnitTestAll: FAILED (pre-existing BookmarkRepositoryImplTest issue) - lintDebugAll: FAILED (pre-existing missing translation issues)
- Remove image toggle row from BookmarkDetailDialog and backing ViewModel action/state - Remove downloadImagesWithArticles preference key from SettingsDataStore/Impl - Remove BookmarkDetailViewModel.onRemoveDownloadedContent() and BookmarkListViewModel.onRemoveDownloadedContent() (dead code) - Remove onRemoveDownloadedContent callback parameter from full composable chain: BookmarkDetailScreen, BookmarkDetailTopBar, BookmarkDetailMenu, BookmarkCard, BookmarkCards, BookmarkListScreen, BookmarkDetailsDialog - Remove sync_download_images string resources from all language files - Update LoadContentPackageUseCase to treat images as always enabled (removing isDownloadImagesEnabled() calls; full cleanup in Slice 4) - Update ContentSyncPolicyEvaluator and LoadBookmarksUseCase to remove downloadImages constraint gating - Update SyncSettingsScreen/ViewModel to remove downloadImages state - Remove tests for removed functionality
- Remove ContentPackageManager.deleteResources() and its relative-URL safety check (containsRelativeResourceUrls, RELATIVE_RESOURCE_URL_PATTERN) - Remove LoadContentPackageUseCase.executeTextOnlyOverwrite() — the HTML-refresh + URL-rewriting downgrade path - Remove unused settingsDataStore injection from LoadContentPackageUseCase - Simplify fetchAndCommit() to always call fetchContentPackages - Clean up executeForceRefresh() signature (forceResources param removed) - Remove BatchArticleLoadWorker hysteresis pruning, adaptive batch sizing, limitHitDuringRun branch, restoreDownloadedStateIfHasContent call, contentPackageManager param, and BATCH_SIZE_NEAR_LIMIT/watermark constants - Remove BookmarkDao.restoreDownloadedStateIfHasContent() DB query - Remove ContentPackageManagerTest (all tests were for deleteResources) - Fix LoadContentPackageUseCaseTest constructor to match updated signature ContentSyncPolicyEvaluator unchanged — already clean after Slice 3.
…efresh
Remove the `hasResources: Boolean` parameter from
`LoadContentPackageUseCase.refreshHtmlForAnnotations` and
`fallbackMultipartRefresh`. URL rewriting to relative paths now always
occurs unconditionally, reflecting the unified model where the only
supported offline state is a full package (HTML + resources).
Guard the call site in `BookmarkDetailViewModel.refreshAnnotationHtml`:
only invoke `refreshHtmlForAnnotations` when `hasResources == true`.
Text-cached bookmarks (hasResources=false) fall through to the existing
`executeForceRefresh` fallback, avoiding broken relative image URLs for
bookmarks that have no local resource files.
Update `LoadContentPackageUseCaseTest`:
- Remove `hasResources` argument from all call sites and mock matchers.
- Remove the now-invalid `urlRewriteNotAppliedWhenNoResources` test
(behaviour no longer exists).
- Rename `urlRewriteAppliedWhenHasResources` → `urlRewriteApplied`.
- Update `fetchHtmlOnly` mock matchers to drop the second argument.
Update `BookmarkDetailViewModelTest`:
- Remove `hasResources` argument from `refreshHtmlForAnnotations` mocks.
- Add `coEvery { contentPackageManager.hasResources(any()) } returns true`
to `@Before` setup so `cachedHasResources` is populated correctly and
the hasResources guard passes in annotation-lifecycle tests.
- Add per-test `contentPackageManager.hasResources("123") = true` overrides
in the two annotation tests that verify `refreshHtmlForAnnotations` is
called.
Files changed:
LoadContentPackageUseCase.kt
BookmarkDetailViewModel.kt
LoadContentPackageUseCaseTest.kt
BookmarkDetailViewModelTest.kt
When offline reading is enabled and a bookmark opens via the legacy text-only fallback (hasResources=false after fetchContentOnDemand), automatically enqueue a high-priority full-package download so the bookmark transitions from text-cached to fully offline. BatchArticleLoadWorker: - Add KEY_PRIORITY_BOOKMARK_ID input data key. - doWork() checks for the priority key first; if present, routes to processPriorityBookmark() which skips the DAO batch query and the isRunning guard, allowing priority jobs to coexist with the batch. - processPriorityBookmark() checks policyEvaluator constraints then calls loadContentPackageUseCase.executeBatch(listOf(bookmarkId)). - enqueuePriorityDownload() companion function: enqueues under unique name "content_priority_<bookmarkId>" with REPLACE policy to deduplicate repeated opens of the same bookmark. BookmarkDetailViewModel: - Inject WorkManager. - After fetchContentOnDemand resolves to Loaded, check cachedHasResources != true && isOfflineReadingEnabled(). If both true, call enqueuePriorityPackageDownload() which reads WiFi/battery constraints from settingsDataStore and delegates to BatchArticleLoadWorker.enqueuePriorityDownload(). BookmarkDetailViewModelTest: - Add WorkManager mock (relaxed) to field declarations, @before setup, and createViewModel() constructor call. Files changed: BatchArticleLoadWorker.kt BookmarkDetailViewModel.kt BookmarkDetailViewModelTest.kt
- Dissolve Sync Status section; redistribute last/next sync timestamps into Bookmark Sync section as a compact two-line block below the frequency ListItem - Add Bookmarks stats (My List / Archived / Favorites) using CompactStatRow in Bookmark Sync section for library context - Add ContentSyncStatusIndicator: surfaceVariant card with leading icon and bodySmall status text, absorbing the previous floating status text and purging indicator; error states use error color, syncing uses primary - Add 'What to keep offline' / 'Whether to download' / 'Storage' section subheadings using labelLarge for correct MD3 hierarchy - Replace StatusRow (ListItem-based) with CompactStatRow (flush label + right-aligned onSurfaceVariant value, no minimum height) in storage stats - Move Clear All Offline Content button inside AnimatedVisibility block - Remove trailing HorizontalDivider after OfflineReadingSection - Fix timestamp format: DateFormat.SHORT for time omits seconds - Fix all hardcoded strings: 6 new resource keys added to all 10 language files - Update BookmarkRepositoryImplTest for offline reading settings compatibility - Update GitHub workflow configs - Mark Slice 8 complete in implementation plan; add spec amendment Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
UI tweaks: - subheading labels (Bookmarks, What to keep offline, Whether to download, Storage) now use labelLarge + primary color per MD3 list-subheader pattern - group Sync frequency ListItem and Last/Next sync timestamps in a tight inner Column to remove the spurious gap caused by ListItem's internal padding - restructure Bookmarks heading+stats to match Storage exactly: Text(padding vertical 8dp) + Column(32dp indent, spacedBy 4dp) - remove trailing HorizontalDivider after Offline Reading section - move Last/Next sync out of ListItem supportingContent to avoid text wrapping next to the trailing frequency control Content changes: - remove test storage limits (5 MB, 10 MB, 20 MB) from enum, ViewModel, dialog, string resources, and all language files - add 500 as a Most recent bookmark count option - rename "Newest N bookmarks" to "Most recent" across all files - rename "Include archived bookmarks" to "Include Archive" - remove "policy" from user-visible strings; update max storage cap description to reflect renamed policy labels - update section and toggle descriptions to be non-redundant and avoid technical jargon User guide (settings.md): - restructure Offline Reading section to match new subheading layout - fix Bookmarks counts position (above button, not below) - replace "offline packages" and "on-demand text caches" with plain language throughout Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace broken !github.event.pull_request guard (always true on push events) with an explicit condition: run on workflow_dispatch, any pull_request, or pushes to main only. Feature branch pushes without an open PR no longer trigger checks or snapshot builds. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
… (Slice 13) Expand OfflinePolicyEvaluatorTest with boundary condition coverage for all three policies (STORAGE_LIMIT, NEWEST_N, DATE_RANGE) including secondary cap interactions, and add BatchArticleLoadWorkerTest covering the download-prune cycle, priority path, constraint gating, and error handling. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Without weight(1f), a long site name consumed all available row width, leaving 0dp for the reading time Text. Since that Text had no maxLines limit, Compose wrapped it one character per line, inflating the site row height ~144dp and pushing labels and action buttons off the bottom of the fixed-height card. Restore Modifier.weight(1f) on siteName in all five affected card variants (Grid: MobilePortrait, Narrow, Wide; Compact: Narrow, Wide) so reading time and download status icon are measured at their natural widths first. Also add maxLines=1 to reading time Text as a defensive guard. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…ghts list - toc-fragment-link-navigation-spec.md: production-ready spec with full WebViewTocBridge source, exact injection point, and step-by-step plan - highlights-nav-drawer-list-spec.md: supersedes _notes draft; addresses typed navigation routes, ARTICLE-gate bug fix, video annotation refresh fix, scroll-to-annotation wiring, and rail/drawer integration detail Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…otes Move implemented branch-specific specs to docs/archive/: - sync-multipart-adoption-spec.md (all stages complete) - granular-offline-content-management-spec.md (implemented 2026-03-26) - highlight-compose-fix.md (implemented and verified) - branch-guardrails.md, branch-pre-merge-fixes.md, branch-pre-merge-guardrails-impl.md, branch-validation-tests.md (guardrails implemented; tests exist) - branch-summary-sync-multipart-v012.md (historical record) Remove intermediate/superseded notes from git (preserved locally in _notes/): - CHAT_PROMPT_ALT_SYNC.md, CHAT_branch-architecture-analysis.md, CHAT_main-branch-architecture-analysis.md (transient analysis artifacts) - managed-offline-reading-spec.md (superseded by unified-offline-content-spec.md) - granular-offline-content-followup-spec.md (superseded by unified-offline pair) - managed-offline-reading-progress.md (progress tracker for superseded spec) Remaining in docs/specs/: 11 active specs (unimplemented or in-progress). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- workflow_overhaul_summary.md → docs/archive/ (WORKFLOW.md already reflects all changes described; summary is a historical record) - TODO-post-sync-branch.md → docs/specs/incremental-content-rendering-spec.md (rename to describe the feature: progressive article content loading) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
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
2 participants
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
Adopts multipart sync end‑to‑end and replaces legacy article caching with a unified offline content architecture. Bookmarks either have a full offline package (HTML + images) or they don't — managed automatically by a configurable policy engine. Also includes annotation caching, multiple reader/UX fixes, CI workflow improvements, and database migrations to support the new content model.
Major Changes
Multipart Sync + Metadata Migration
POST /bookmarks/sync.On‑Demand Content Packages
WebViewAssetLoader+ per‑bookmark base URLs.Offline Annotation Support
CachedAnnotationEntity+ DAO + schema migration.Unified Offline Content Architecture
Replaced an intermediate two‑tier model (text‑only cache vs. full package with per‑bookmark toggle) with a single‑tier policy‑driven approach:
OfflinePolicyEvaluator) with three modes:BatchArticleLoadWorker) runs the policy check → download → prune cycle, with constraint gating (Wi‑Fi only, battery saver) checked between batches.deleteResources,downloadImagesWithArticlesconditional).Reader & UX Fixes
Other Fixes & Improvements
BookmarkDtoembed_hostname→embed_domainserialization; added unit tests.omitDescriptionnot persisting (SyncRequestwas silently omittingwith_json).embed/embed_domainacross sync paths.Persistence + Migration
article_contenttable.content_package,content_resource, and cached annotation tables.CI & Build
ifconditions to prevent double‑runs on push+PR.Spec & Docs
Testing
./gradlew :app:assembleDebugAll./gradlew :app:testDebugUnitTestAll./gradlew :app:lintDebugAllNotes
has_article = true(offline transcript only; playback remains online).