The flake no longer takes yt-dlp from nixpkgs; a dedicated `ytDlp`
derivation fetches the upstream release binary directly and pins both
`version` and an SRI `hash`. That makes the previous workflow inert: it
ran `nix flake update nixpkgs` and compared `nixpkgs#yt-dlp.version`
before and after, so it could churn the lockfile forever without ever
moving the version we actually ship.
The workflow now reads the pinned version straight out of the `ytDlp`
block in flake.nix, asks the GitHub API for yt-dlp's latest release tag,
short-circuits when they already match, downloads the new release to
recompute its SRI hash (required — the hash is part of the derivation's
identity, so the URL cannot be changed alone), and rewrites the three
pinned fields under a sed range address scoped to that block so sibling
pins like ublockLite and isdcac are untouched. It asserts only flake.nix
changed and that the new version appears exactly twice before opening
the PR.
The nix flake wrapper pins a yt-dlp via ARCHIVR_YT_DLP, but yt-dlp rots
fast — extractors break within weeks of a pin. Add a resolver that probes
`--version` on both the pinned binary and a user-installed copy under the
mutable state dir, and runs whichever is newer.
Version strings are YYYY.MM.DD, so plain string ordering is chronological.
Ties resolve toward the state dir: a user who installed it there did so
deliberately. ARCHIVR_YT_DLP_FORCE bypasses the comparison entirely, and
with no candidate at all we fall back to bare `yt-dlp` on PATH — exactly
the previous behaviour.
Resolution is cached in a OnceLock so `--version` costs one subprocess per
process, and all four inline env::var lookups now go through it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two independent problems, one commit:
1. Stale binary. nixpkgs-provided `pkgs.yt-dlp` on the pinned
nixos-unstable rev is 2026.03.17 (Mar 2026). yt-dlp itself
releases days-to-weeks, and YouTube frequently rotates the
player-signature / client surfaces the older builds request
(`android_vr` is the current casualty), which returns HTTP 403
mid-download for the format specs archivr passes (`-f
bestvideo+bestaudio/best`). Even bumping the nixpkgs input would
leave us dependent on that channel's yt-dlp cadence.
Fetch the upstream zipapp directly instead
(github.com/yt-dlp/yt-dlp/releases/download/<ver>/yt-dlp), wrap so
`python3` and `ffmpeg` are on PATH, and pin version+hash in one
place. Bumping is: change version, replace hash from
`nix hash file <url>`.
2. Missing pin in server wrapper. `archivr-cli` was already wrapped
with `--set ARCHIVR_YT_DLP` + a PATH prefix; `archivr-server`
was NOT — it only pinned single-file, chrome, and the tweet
scraper, silently falling back to whatever `yt-dlp` the user
happened to have on PATH. Server captures therefore inherited
the user's (often stale) system yt-dlp regardless of the flake
pin. Same wrapper flags now apply to both binaries.
devShell keeps `pkgs.yt-dlp` for now: the dev shell is a
convenience, not a release surface, and matching wouldn't fit in this
commit without duplicating the derivation across let-scopes.
Text captures land as `.md` (Markdown) or `.txt` (plain) blobs, but
PreviewPanel only dispatched on video/audio/image/pdf/html extensions,
so opening a text entry hit the 'No preview available' fallback with
the raw artifact path exposed.
- New `TextPreview` component fetches the primary artifact as text,
renders it in a monospace `<pre>` with word-wrap, and shows the
entry title on top and the MIME as a small trailing tag. Handles
loading/error states.
- `PreviewPanel` gains a `TEXT_EXTS` set + a branch that dispatches
to `TextPreview` for `md` / `markdown` / `txt`.
- CSS is padded and centered to ~780px so a text note reads like a
document rather than an edge-to-edge terminal dump.
v1 intentionally does NOT parse Markdown: keeping frontend deps at
react+react-dom only. Bump to a real Markdown renderer if we start
capturing Markdown-authored notes.
Tweet and tweet_thread entries store their payload under artifact_role
`raw_tweet_json`, not `primary_media`. `build_summary_input` filtered
strictly for `primary_media LIMIT 1`, so both cases silently failed
with 'entry X has no primary_media artifact to summarize'.
Threads compound the problem: the tweet scraper writes ONE json file
per status, so even a fixed lookup that took the first row would
summarize only the initial tweet and lose the rest of the conversation.
Fixes:
- New `load_summary_artifacts` helper returns every artifact for a
role in insertion order.
- For entity_kind `tweet` / `tweet_thread`, load all
`raw_tweet_json` artifacts (falling back to `primary_media` for
archives predating that role convention).
- Iterate artifacts, extract text per file with the existing
markdown/html/json branches, then join thread pieces with a
`---` separator so the model sees a real paragraph break between
statuses instead of one flowing document.
Single-tweet entries produce one piece and the separator never
renders. Non-tweet entries behave exactly as before.
The text row shipped with semantic classnames (`capture-text-inputs`,
`capture-text-title`, `capture-text-body`, `capture-text-mime`,
`capture-text-icon`) but no CSS rules. Falling through to the parent
`.capture-row-main` flex-row (`display: flex; align-items: center`)
meant the title, textarea, and mime-select stacked as intrinsic-width
boxes centered on the tall body, producing a layout where the body
floated to the top-right, the title box appeared BELOW it, and the mime
selector rendered as an unstyled OS dropdown.
Fix:
- `.capture-text-row .capture-row-main` uses `align-items: flex-start`
so the leading icon and trailing × pin to the top of the block.
- `.capture-text-inputs` is now a full-width column-flex container with
proper gaps.
- `.capture-text-title` reuses the 44px input height and typography of
`.capture-input`; `.capture-text-body` gets a 140px min-height,
vertical resize, and matching border/focus treatment.
- `.capture-text-mime` is styled as a small chip with a custom caret
so it matches `.capture-quality` and stops looking like a raw
`<select>`. Sits in a right-aligned footer under the body.
- `.capture-text-icon` gets a 44px column so it aligns with the title
input; remove button gets a small top-margin for the same reason.
Rebuilt static bundle bumped as well (`index-BLxoi9rt.css`,
`index-CQcpPA_I.js`).
Two related fixes for the codex_cli summary provider:
1. Executable discovery. `ARCHIVR_CODEX_CLI` was already respected, but
without it the code resolved to bare `codex` and relied on PATH.
The ChatGPT desktop app installs codex at
`/Applications/ChatGPT.app/Contents/Resources/codex` and does not
put it on PATH, so users who only have the desktop app saw
'No such file or directory' with no hint. `resolve_cli` now walks
env override → a small set of well-known absolute paths → HOME
/.local/bin/<bare> → bare fallback. Same treatment applied to
claude_cli for symmetry (/opt/homebrew/bin/claude, /usr/local/bin/
claude, HOME/.local/bin/claude).
2. Clean output. `codex exec -` writes a runtime header ("OpenAI
Codex vX", session id, sandbox, model), the assistant reply, and a
footer ("tokens used", replay of the reply) to stdout. The JSON
extractor took the first '{' from the *user prompt echo* and the
last '}' from the trailing replay, producing invalid text that
fell through to the "raw text under summary" fallback path. Now
uses `--output-last-message <tempfile>` and reads only the final
assistant message. Fallback (positional prompt) uses the same
flag. Tempfile is cleaned up on all paths, incl. spawn failure.
New "Summary" rail section between the URL/Preview controls and .meta-list.
A completed summary renders as bold tl;dr, body paragraph, tag chips, and a
provider · model footer; missing or failed shows Generate; pending/running
shows an inline spinner and polls GET every 1500 ms until terminal.
- api.js: fetchEntrySummary + requestEntrySummary. The POST helper unwraps
ApiError's { "error": ... } body so the missing-env-var message reaches
the user verbatim rather than as a bare status code.
- ContextRail.jsx: state seeds from detail.latest_summary so the section
renders immediately on selection. Polling is anchored on the summary
status rather than started inside the click handler, so a job still
running when the user navigates away and back is picked up again. A
transient poll failure is swallowed — the next tick retries, and a real
failure arrives as status === 'failed'.
- Regenerate passes force:true only when a completed summary is already
shown; otherwise the request can take the server's 200 cache-hit path.
- Provider choice persists in sessionStorage under archivr:summary:provider,
with try/catch around both accessors for private-mode browsers.
- Public sessions never see the selector or the Generate button, and the
section renders at all only when a completed summary made it through the
server's visibility gate.
- styles.css: .rail-summary-* only; spacing and the action button reuse
.rail-section and .rail-rearchive-btn. The spinner honours
prefers-reduced-motion — the text alone conveys the state.
- AGENTS.md: document the summary env vars alongside the existing
external-tool convention.
Smoke-tested end to end against a scratch archive with a seeded markdown
entry: claude_cli produced a real summary (pending → running → completed in
~11s); a local mock server exercised the openai_compatible transport and
confirmed the Bearer header, model, and system/user role split on the wire;
unconfigured providers return 400 naming the exact variable; a video entry
returns 400 "v1 unsupported"; a repeat POST returns 200 from cache without
adding a row.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
GET is read-only and gated exactly like entry detail, so a guest can read a
summary only for an entry whose content they could already read. POST
requires ROLE_USER, matching capture / tags / patch / rearchive; no auth
roles change.
Both the provider config and the content extraction resolve on the request
thread, before spawn_blocking. That is what lets a missing env var come back
as a synchronous 400 naming the exact variable, and an unsummarizable
artifact (video, audio) as a 400 saying so, rather than becoming a
background job the caller must poll only to learn about a config typo.
The pending row is claimed before spawning so the 202 can name a summary_uid
the client can poll immediately. summarize_entry owns the
pending → running → completed/failed transitions for that same row — the
cache key is identical, so both upserts resolve to one row — leaving the
handler to catch only the case where it fails before recording anything.
When !force and an identical cache key already completed, the existing row
comes back as a 200 with no new work.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Per-entry LLM summaries as a regenerable child record, not a column on
archived_entries and not an on-disk artifact: an entry may carry several
summaries (one per provider/model/prompt version), any of which can be
discarded and recomputed. Generation is manual-only — nothing in capture.rs
calls into this module.
- database.rs: entry_summaries table + index, EntrySummaryRecord, and
upsert/update/find/latest helpers mirroring the capture_jobs style.
provider_model is stored as '' rather than NULL because SQLite treats
NULLs as distinct inside a UNIQUE index, which would stop the CLI
providers (no model) from ever deduping on the cache key.
- summarizer.rs: SummaryProvider trait with four implementations —
Anthropic Messages API, OpenAI-compatible chat completions, `claude -p`
and `codex exec -`. Configuration comes from env vars only (never TOML),
matching how yt-dlp / single-file / tweet-scraper are resolved, which
also keeps API keys out of anything the archive persists.
- archive.rs: EntryDetail gains latest_summary, populated by one extra
LIMIT 1 query in get_entry_detail. EntrySummaryView aliases the DB row
rather than duplicating it.
Implementation notes:
- No tokio in core. CLI timeouts are enforced structurally: stdout is
drained on its own thread and handed back over a channel so the calling
thread can recv_timeout and kill an overrunning child; stdin is written
on a third thread so a 48 KB prompt cannot deadlock against a child
waiting for us to read.
- HTML is reduced with regex rather than a parser: html5ever is not in the
tree, and a model tolerates imperfect whitespace. Paired tags are spelled
out per tag because Rust's regex engine has no backreferences by design.
- reqwest is declared with only the `blocking` feature here, so bodies are
serialized via .body(value.to_string()) instead of widening the
workspace dependency for .json().
- input_sha256 holds a SHA3-256 digest via hash::hash_bytes, the tree's one
hashing primitive; the content is truncated to 48 KB *before* hashing so
the cache key describes exactly the bytes the model saw.
Tests: no mockito/wiremock in dev-deps, and adding a mock HTTP server for
one JSON shape is a poor trade, so the two halves that can actually break
are tested directly — request-body builders and response parsers — leaving
only reqwest's own transport uncovered. Plus schema idempotency, cache-key
dedupe, cascade-on-delete, provider_from_env happy/missing-var paths, HTML
and tweet extraction, output normalization, and the CLI runner's stdin
round-trip, timeout kill, and nonzero-exit paths.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- Add submitTextCapture API client function with same error handling as submitCapture
- Create makeTextItem() factory for text capture state
- Implement CaptureTextRow component with title, body textarea, and MIME selector
- Add 'Add text' button in capture dialog toolbar
- Update handleArchive to filter and route text submissions
- Modify submitBgJob to detect and submit text items via submitTextCapture
- Skip probe and conflict checks for text items
- Reuse job tracking and batch settlement for text captures
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
- Add CaptureTextBody struct for title, body, and optional MIME type
- Implement capture_text_handler with validation for empty fields and MIME type
- Route text submissions to perform_text_capture() in background
- Reuse existing capture job tracking and polling infrastructure
- Default MIME type to text/markdown when not specified
- Include route tests covering happy path, validation, auth, and error cases
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
- Add downloader/text.rs module with save() function that stages and hashes text content
- Support text/markdown and text/plain MIME types with .md and .txt extensions
- Add perform_text_capture() function for capturing user-supplied text
- Validates title (non-empty, max 500 chars) and body (non-empty, max 2 MiB)
- Creates blob records and entries with source_kind='text', entity_kind='document'
- Includes comprehensive unit tests for markdown, plain text, and validation
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
list_entries_for_collection was missing AND e.parent_entry_id IS NULL,
causing child entries (playlist/channel videos) to appear alongside
their parent in the flat archive view. list_root_entries and
search_entries already had this guard; the collection path did not.
Also removes garnix from flake.nix nixConfig.
Previously, any WebPage capture with via_freedium=true was routed
through the Freedium mirror regardless of the URL's host. This caused
non-paywall sites like borretti.me to be fetched through Freedium,
which is wrong — Freedium only knows how to handle a specific set of
publications.
Add is_freedium_supported_url() with a static allowlist of the 7 hosts
Freedium explicitly supports per its homepage announcement:
Medium, NYT, WaPo, Bloomberg, Reuters, Economist, Financial Times
The gate matches on the bare domain or any subdomain (e.g.
towardsdatascience.medium.com). Unparseable URLs fall through to a
direct fetch. The existing freedium-mirror.cfd re-wrap guard is kept
as a belt-and-suspenders check after the new allowlist test.
Fixes: https://borretti.me/article/notes-on-managing-adhd archived via
Freedium despite not being a Medium article.
Measured actual GitHub display: 838×279px at 1365px viewport.
Previous content (mark=216px, wordmark=132px) rendered at 83px/51px —
too small for the available space.
New target sizes calibrated to display dimensions:
mark 216 → 285px target → 110px at 838px display (+32%)
wordmark 132 → 170px target → 55px height at display (+35%)
tagline 62 → 78px target → 30px at display (+26%)
A in mark 160 → 210px target
Reduced inter-element gaps to keep content within 724px height;
fills ~84% of banner, leaving ~23px display margin each side.
At GitHub's ~840px README width, 2172×724 renders at 840×280px.
Previous element sizes (mark 130px, wordmark 80px) became 50px and 31px
at display — too small to read clearly.
Scale everything ~1.65× to fill ~80% of banner height:
mark 130 → 216px at target (≈ 83px at 840px display)
wordmark 80 → 132px at target (≈ 51px at display)
tagline 38 → 62px at target (≈ 24px at display)
A in mark 96 → 160px at target
Proportional gap increases maintain the same visual rhythm.
Previous version had three inelegancies visible on GitHub:
- Full-width ghost rule read as an artifact dividing the banner in half
- Top terracotta bar looked like site header chrome
- Radial gradient created compression banding at CDN quality
v5 design:
- Terracotta rounded-rect logo mark (130px) with Paper 'A' inside —
the actual brand element per the style guide, not a floating letter
- Flat Shell / Paper background — compresses losslessly, zero banding
- No top bar, no ghost rule, no gradient gimmicks
- Same crisp Cormorant Garamond SemiBold + 2x supersample + Lanczos
- Both dark and light variants updated with identical layout
Replace app-icon-style horizontal lockup with centered stacked
composition: large standalone terracotta 'A' mark → thin parchment
rule → wide-tracked cream wordmark → amber italic tagline.
Edge-darkening vignette clears the center and frames the bokeh
atmosphere without muddying the text zone.
Render generated-image.png (blurred library bokeh, prompt 3) as the
dark-mode README banner with logo mark, wordmark, tagline, and
terracotta top rule composited on top via Pillow.
Switch README dark-mode <source> from banner-dark.svg to banner-dark.png.
Light-mode banner stays as the clean Paper SVG.
- Add logo (SVG lockup), tagline, and brand-colored badges at top
- Replace dev checklist with prose Features section
- Add Architecture section with binary overview and on-disk layout
- Consolidate Supported Inputs into a table with inline shorthand examples
- Rewrite env vars as a scannable table
- Restructure into clear Quick Start → Architecture → Inputs → Config → Deployment → Development flow
- Remove personal motivation text and developer-facing in-progress notes
* Add Archivr brand identity and favicon
- Brand style guide (docs/branding/style-guide.html): logo, clear
space, don'ts, color palette, typography, in-context mockup
- Logo assets (docs/branding/assets/): lockup SVGs for dark/light
backgrounds, standalone mark, favicon.svg, multi-size favicon.ico
- Wire favicon into frontend/index.html (SVG + ICO fallback);
frontend/public/ seeded so Vite copies assets on next build
- Revert display strings archivr back to Archivr in UI
(Topbar, LoginPage, SetupPage, index.html titles, flake.nix)
- Update .gitignore to allow docs/branding/
* Build frontend with favicon links and assets
* Fix static file serving: serve all of static/ with SPA index.html fallback
Previously only /assets/* was served from disk; all other paths
(including /favicon.ico, /favicon.svg) hit the index.html fallback.
Replace nest_service("/assets") + fallback_service(index.html) with
ServeDir on the full static dir and not_found_service(index.html),
so any file that exists in static/ is served directly.
* feat(core): add requires_auth to collections; include name in entry-collection memberships
- Add `requires_auth INTEGER NOT NULL DEFAULT 1` column to the
collections DDL and as an idempotent ALTER TABLE migration in
initialize_schema (archive DB), not initialize_auth_schema.
- CollectionRecord and CollectionSummary gain `requires_auth: bool`.
- create_collection() and update_collection() accept the new field.
- get_entry_collection_memberships() now returns collection name as the
third tuple element; EntryCollectionMembership gains a `name` field
so the sidebar can show human-readable names instead of raw UIDs.
* feat(server): conditional auth for public collections; add requires_auth + original_url to API
- CreateCollectionBody gains requires_auth (default true).
- PatchCollectionBody gains requires_auth: Option<bool>.
- get_collection_handler: load record first, then skip auth.require_auth()
when record.requires_auth == false so public collections are accessible
to unauthenticated callers; caller_bits falls back to ROLE_GUEST (1)
so only visibility_bits=3 entries are returned to guests.
- Collection JSON response includes requires_auth and each entry now
includes original_url for use by the public collection page.
- list_collections_handler keeps require_auth (management UI).
* feat(frontend): public collection page at /c/:archiveId/:collUid
- Detect PUBLIC_COLL_ROUTE at module load time (like PREVIEW_ROUTE) and
return <PublicCollectionPage> before any auth checks so unauthenticated
users can view public collections without hitting the login gate.
- PublicCollectionPage fetches via getCollection() and renders the
server-filtered entry list (no client-side bitmask filtering - the
server already applies caller_bits=GUEST for unauthenticated requests).
Entry titles link to original_url when present; fall back to plain
text when original_url is null.
- api.js createCollection() gains requiresAuth param (default true),
sent as requires_auth in the request body.
- Storybook story covers WithEntries, Empty, and LoadError states.
* feat(frontend): collections view improvements
- addVis in the 'Add entry' form now syncs to the selected collection's
default_visibility_bits via useEffect on collDetail, so the default
matches the collection's configured entry visibility.
- Rename 'Default visibility' label to 'Entries\' default visibility'
in both the detail pane and the create form to distinguish it from
the new collection-level access setting.
- Add 'Require authentication to view' checkbox in the detail pane
backed by a PATCH to requires_auth; reads collDetail?.requires_auth
with fallback to the list-level selected record.
- Create form gains a matching requires_auth checkbox (default: true),
passed as 5th arg to createCollection().
* feat(frontend): context rail collection improvements
- Show collection name (c.name) instead of raw UID in the sidebar
Collections section; names now come from the updated
EntryCollectionMembership API response.
- Fix horizontal overflow on long collection names: coll-name gains
overflow:hidden + text-overflow:ellipsis + white-space:nowrap +
min-width:0; coll-row gets overflow:hidden.
- Single-entry 'Add to collection' UI: dropdown + button inside the
Collections rail section lets users add the current entry to any
non-default collection without multi-selecting. After add, the
membership list refreshes automatically.
- Collections section now shows even when entryCollections is empty,
as long as non-default collections exist (so the add form is
accessible for un-membered entries).
- Bulk 'Add to collection' now uses the target collection's
default_visibility_bits instead of hardcoded 2 (Users only).
- Both bulk and single-entry dropdowns filter out slug='_default_'
to match the backend rejection in add_entry_to_collection_handler.
- Collections list is now fetched on archiveId change (not just on
bulk mode entry) so it is available for single-entry mode too.
* build(frontend): update static assets
* feat(frontend): public collection link UX + app-styled public page
CollectionsView:
- When a collection has requires_auth=false, show a read-only URL input
and Copy button below the auth checkbox so the public link is
immediately discoverable. The input auto-selects on focus so manual
copy always works. Copy button tries navigator.clipboard.writeText
first; falls back to execCommand('copy') for HTTP deployments where
the Clipboard API is unavailable in non-secure contexts.
PublicCollectionPage:
- Rewritten to use the app's CSS classes and variables instead of
bare inline styles, so it visually matches the main archive UI.
Dark topbar (.pub-coll-topbar) with brand + collection name, paper
background body, entry list via .coll-entries-list / .coll-entry-row /
.coll-entry-info / .coll-entry-kind — the same classes used in the
authenticated Collections view.
styles.css:
- .coll-public-link-row / -wrap / -input / .coll-copy-btn for the
new link field in CollectionsView detail pane.
- .pub-coll-* classes for the public page layout and typography.
* feat(core): add get_collection_by_slug; scope search to active collection
- get_collection_by_slug(): new function mirroring get_collection_by_uid
but matching on slug, used to resolve the _default_ collection when no
?collection param is supplied.
- SearchEntriesQuery gains collection_id: Option<i64>. When set, the
search SQL adds an EXISTS subquery that checks collection_entries cef
for both membership (cef.collection_id = ?) and visibility bits in
that specific collection — preventing cross-collection visibility
leaks where an entry is public in one collection but private in the
current one. Without collection_id the original cross-collection
visibility fallback is kept.
* feat(server): collection-scoped entries/search with uniform auth gate
All entry listing and search now route through the active collection:
list_entries (?collection=<uid>|main|<omitted>):
- Resolves the target collection; omitted or 'main' resolves to _default_.
- Checks requires_auth on that collection; gates auth conditionally.
- Returns list_entries_for_collection() — same EntrySummary shape.
search_entries_handler:
- Same collection resolution + conditional auth as list_entries.
- Sets search_query.collection_id so SQL scopes membership + visibility
to the specific collection, not cross-collection fallback.
list_collections_handler:
- Dropped require_auth() — collection summaries (name/slug/uid/
requires_auth/default_visibility_bits) are public metadata needed for
the guest collection-switcher dropdown.
Tests:
- list_collections_requires_auth → list_collections_is_public (200).
- list_entries_requires_auth and search coverage still pass.
* feat(frontend): integrate collection switching into main Archive view
Replaces the standalone /c/:archiveId/:collUid public page with a
unified main-view approach where all collection logic lives at /.
URL param:
- ?collection=<uid> selects a collection; omitted or 'main' = default.
- 'main' is normalized to null in parseLocation() so the dropdown shows
'All entries' and the URL stays clean.
Collection switcher (Topbar):
- Dropdown always visible (guests need it to navigate public collections).
- Non-default collections only (All entries = no param = _default_).
- Guest selecting an auth-required collection calls onSignInClick().
- handleCollectionChange checks both named and _default_ requires_auth
before proceeding, redirecting guests to login if needed.
listCollections fetched for all users (guests too) since the endpoint
is now public; used to populate the switcher without auth.
Public-session mode (authenticated state, no currentUser):
- Auth gate: fetchArchives() + fetchEntries() with collection param;
401 falls through to login, 200 proceeds as guest.
- auth:expired suppressed when !currentUser.
- fetchEntryDetail skipped; ContextRail shows entry summary + sign-in prompt.
- ContextRail selection effect skips tag/collection API calls.
- runs/tags not fetched in guest mode.
- Child row expansion disabled in EntryRow (hasChildren = false).
api.js:
- fetchEntries/searchEntries both thread ?collection=<uid> to server.
Deleted: PublicCollectionPage.jsx, PublicCollectionPage.stories.jsx,
copy-link UI from CollectionsView, pub-coll-*/copy-link CSS.
* build(frontend): update static assets
* feat(core): add is_entry_publicly_accessible; checks entry+parent vs public collections
* feat(server): allow guests to fetch detail/children/artifacts for public entries
* feat(frontend): guest collection dropdown filtering; public entry detail without auth wall
* build(frontend): update static assets
* test(server): public entry detail/artifact/children contract for guests
* feat(server): filter auth-required collections from guest list_collections response
* server: add scoped media-token endpoint for Cast/AirPlay auth bypass
Chromecast and Apple TV fetch media URLs as independent HTTP clients
with no session cookie. The existing serve_artifact handler requires
auth_user.require_auth(), so those devices always received 401.
Changes:
- MediaToken struct stored in AppState (Arc<Mutex<HashMap>>), scoped to
a single (archive_id, entry_uid, artifact_index) tuple with a 2-hour TTL
- POST /api/archives/:id/entries/:uid/artifacts/:idx/media-token
requires an authenticated session, verifies the artifact exists,
prunes expired tokens, mints a 43-char URL-safe token, and returns
{ url, expires_in_secs }
- serve_artifact now accepts an optional ?token= query param; a valid
scoped token bypasses require_auth() while a missing/invalid/expired
token falls through to the normal 401 path
- CSP script-src extended to include https://www.gstatic.com so the
Cast sender SDK script (injected lazily by VideoPreview) is not blocked
- 4 new tests: bare-URL still 401, tokenized fetch succeeds without
session cookie, bogus token 401, wrong-artifact-index 401
* frontend: Cast/AirPlay overlay in VideoPreview
When the user opens a video archive entry, VideoPreview now:
1. Issues a signed media token (POST .../artifacts/:idx/media-token) and
uses the returned signed URL as <video src>. This ensures the video
element's src is one that Cast devices and Apple TV can fetch without a
session cookie.
2. Lazily injects the Google Cast SDK script (cast_sender.js from
gstatic.com, now allowed by the updated CSP). Once the SDK reports
available, a <google-cast-launcher> web component appears as an overlay
button in the top-right corner of the video. Selecting a Cast device
triggers loadMedia() with the signed URL and the artifact's MIME type.
3. Detects AirPlay support (webkitShowPlaybackTargetPicker on
HTMLVideoElement) and shows an AirPlay icon button alongside Cast.
The <video> element carries x-webkit-airplay='allow', so Safari's native
controls also surface the AirPlay option. The explicit overlay button
calls webkitShowPlaybackTargetPicker() for consistent placement.
Both buttons are hidden when the respective APIs are unavailable (HTTP
pages, non-Safari for AirPlay, no Cast extension/devices), so there is no
UI regression for users who don't cast.
PreviewPanel now passes contentType (derived from artifact extension) to
VideoPreview so Cast receives a correct MIME type.
New CSS: .video-tv-controls (absolute overlay), .video-tv-btn (frosted
glass icon button), .video-tv-loading (placeholder during token fetch).
* server: fix serve_artifact auth OR logic — bogus token falls back to session
Previously a request carrying ?token=<expired> was immediately rejected
with 401, even if the user held a valid session cookie. This broke
logged-in browser playback after the 2-hour signed-URL window expired,
because VideoPreview uses the signed URL as <video src>.
Fix: compute token_valid first; if the token is absent or invalid, fall
through to auth_user.require_auth() instead of returning early.
Effect: valid token skips session check, invalid/missing token checks
session, both invalid → 401 as before.
Updated the bogus-token-no-session test docstring to clarify it tests
the no-auth path specifically. Added new test:
media_token_bogus_token_with_session_returns_200 — verifies a logged-in
user can still fetch the artifact via a URL carrying a stale token.
* frontend: guard token-fetch effect against stale async resolution
A slow issueMediaToken() response for video A could resolve after the
user selected video B and call setSignedSrc(urlA), making the
preview/Cast play the wrong file.
Add a cancelled flag set in the effect cleanup; both .then and .catch
check it before touching state, so only the most recent src wins.
* frontend: load Cast media immediately if session already exists
Previously the effect only sent video to the TV on SESSION_STARTED /
SESSION_RESUMED events. Two gaps:
1. If a Cast session was already active when signedSrc became ready
(e.g. the SDK resumed a session before the token fetch finished, or
the user switches videos while already casting), nothing was sent.
2. Same gap if castReady fired after an already-established session.
Fix: extract loadMedia(session) and call it against
ctx.getCurrentSession() immediately when castReady + signedSrc are both
truthy, in addition to keeping the event listener for future connects.
* server: staged file-upload endpoint
POST /api/archives/:id/uploads streams a multipart body to a temp file
under the archive's store/temp/ directory and returns a staged_path the
capture pipeline can move into place.
- Routes: /api/archives/:id/uploads (POST, requires auth)
- Body cap: 10 GiB; chunk-streamed to disk, never buffered in memory
- Path-traversal sanitised on the filename field
- Temp files are cleaned up on error paths (disk-leak fix)
- main.rs wires the new route into the server startup
- Cargo: adds the multipart dependency
* frontend: file upload in Capture dialog
Drag-and-drop or 'Upload file' button stages files for archiving:
- File items sit alongside URL rows in the same list; each shows the
original filename, a live progress bar during upload, and a check badge
when ready. The locator input is replaced entirely — no editable field.
- Archive button is disabled until all uploads finish; each file item
contributes to the Archive N count once its upload is done.
- File items are excluded from sessionStorage persistence (they are
transient — the staged server path would be invalid after a reload).
Staged-file cleanup is handled at every exit path so temp/uploads/ does
not accumulate:
• removeRow on an in-progress item aborts the XHR; removeRow on a done
item calls DELETE /archives/:id/uploads.
• Dialog cancel (Escape / Cancel button) aborts all in-flight XHRs and
DELETEs all completed staged files via the close-event handler.
• handleArchive sets isSubmittingRef=true before dialog.close() so the
close handler skips cleanup — the background capture job handles
staged-file removal on success instead.
• uploadFile() returns { promise, abort } so the component can cancel
the XHR without any visible fetch.
api.js additions: uploadFile (XHR with progress + abort), deleteUpload.
(Static assets rebuilt from combined source to include screensharing
changes from this branch.)
* fix: collection enrollment with default_visibility_bits
Two related fixes from feat-file-uploading:
core: fix collection enrollment using default_visibility_bits instead of
entry.visibility — entries were being enrolled with the entry-level
visibility rather than the collection's configured default.
server: allow changing default_visibility_bits on the default collection
— the PATCH handler was incorrectly blocking updates to the default
collection's visibility configuration.
* server: fix unbounded staged-upload disk growth
Two review findings:
P2 — delete staged file on capture failure (routes.rs)
When perform_capture returns Err, the job was marked failed but
staged_upload_path was never removed. With a 10 GiB body cap a few
failed imports could exhaust archive storage before the next restart.
Mirror the success-path cleanup into the Err arm so the file is removed
immediately regardless of outcome.
P1 — periodic staged-upload pruning (main.rs)
The startup prune of temp/uploads/ only ran once, so uploads abandoned
mid-session (browser crash, navigation away) accumulated forever on a
long-running server. Folded the pruning logic into the existing 24 h
maintenance task alongside session cleanup, so stale dirs are swept
continuously without requiring a restart.
* server+frontend: fix staged-upload disk-growth and prune safety
Server (main.rs + routes.rs):
- Extract prune_stale_upload_dirs() helper called by both startup and
the periodic 24h task, eliminating the duplicated loop.
- Sentinel (.uploading) created in the UUID dir before streaming begins;
removed on successful completion; error path uses remove_dir_all so
the partial file and sentinel are cleaned up together.
The periodic prune skips any dir containing .uploading (active XHR).
- Startup prune passes cleanup_stale_sentinels=true: the server has not
started accepting connections yet so any sentinel is a crash remnant —
it is removed and the dir proceeds to the age check, preventing leaked
dirs from a previous crash accumulating forever.
- Staleness measured from the newest non-sentinel child file mtime so a
just-finished slow upload (dir mtime stale, file mtime fresh) is not
pruned before the user can submit it for capture. Empty dirs fall back
to dir mtime.
- Failed captures (Err branch in spawn_blocking) now also delete the
staged file and UUID dir immediately, matching the success path.
Frontend (api.js + CaptureDialog.jsx):
- submitCapture attaches err.status = res.status on non-2xx responses
so callers can distinguish a definite HTTP rejection from a network
error where the response may have been lost.
- submitBgJob catch deletes the staged file only when e.status is set
(server definitively rejected the POST /captures request). A network
error leaves the file in place because the server may have accepted
the job and the response was lost — deleting would race the capture.
* feat(core): YouTube playlist/channel/YTM-playlist capture with parent–child entries
- ytdlp: add fetch_playlist_info() using yt-dlp -J --flat-playlist for
reliable container title + shallow entry list; normalize item URLs via
webpage_url → absolute url → id fallback (domain inferred from container
URL so YTM stays on music.youtube.com)
- capture: add record_container_entry() (no blob, no primary_media artifact);
extend record_media_entry() with parent_entry_id/root_entry_id params (all
existing single-item call sites pass None, None)
- capture: implement YouTubePlaylist / YouTubeChannel / YouTubeMusicPlaylist
capture path replacing the two not-implemented stubs: fetch playlist info →
create container entry (reusing existing run + item) → per-child run items
(parent_item_id = container item) → download each video/track as a child
entry; per-child failures are non-fatal; perform_capture returns result.status
reflecting actual run outcome so capture_handler marks the job correctly
- archive: add child_count i64 to EntrySummary (col 12 in all listing queries);
add get_entry_summary() private helper; fix get_entry_detail() to use
get_entry_summary() so child entries are resolvable via the detail endpoint;
add list_child_entries(conn, uid, caller_bits) with the same
admin/collection visibility predicate as list_root_entries
archive_runs.requested_count stays 1 (one user locator); discovered/
completed/failed_count reflect container item + N video items via
refresh_run_counters.
* feat(server,frontend): expose children endpoint + expand UI for container entries
server:
- add GET /api/archives/:id/entries/:uid/children → list_entry_children,
calling list_child_entries with caller_bits so visibility model is enforced
- fix capture_handler: use result.status ("completed"/"failed") to set job
status rather than always "completed", mirroring rearchive_handler; this
surfaces partial playlist failures to the polling client
frontend:
- api.js: add fetchEntryChildren(archiveId, entryUid)
- EntryRow: one outer div.entry-row-outer (display:block) keeps nth-child
striping correct; inner div.entry-row-main is the flex row with all column
cells and event handling; .child-entries sits below inside the outer wrapper
- expand chevron appears when entry.child_count > 0; clicking fetches children
lazily and renders ChildRow components reusing .col-* flex widths
- child-count badge shown next to title on container entries
- styles.css: scoped CSS with #entries-body > .entry-row-outer selectors
(higher specificity than > div) to override flex on outer wrapper; inner row
and column rules replicated at correct depth; nth-child, is-selected,
is-multi-selected, url-cell hover all handled
* fix(frontend): make child entry rows interactive
ChildRow now receives onRowClick and selectedUids from EntryRow (which
receives selectedUids from EntriesView alongside the existing booleans).
Clicking a child row invokes onRowClick(child, e) so it flows through
handleRowClick → selectEntry → fetchEntryDetail exactly as a root entry
would. Shift-range selection gracefully degrades to single-select since
child entries are not in the root entries array.
Selected/multi-selected visual state is wired: .child-entry-row.is-selected
shows the same #eee2d2 background + accent outline as root rows; hover
restores full opacity. Frontend static assets rebuilt.
* feat(core): playlist per-item quality + incremental sync
ytdlp.rs:
- Add PlaylistItemProbe / PlaylistProbeResult (pub, serde::Serialize)
- Add private available_video_heights_from_value() helper for Value entries
- Add probe_playlist_qualities(): yt-dlp -J (full metadata, no flat flag)
returns per-video quality lists in one subprocess call
capture.rs:
- Add per_item_quality: HashMap<String,String> and sync: bool to CaptureConfig
(both Default; keyed by yt-dlp video ID, not URL)
- Add pub locator_to_playlist_url(): validates only the three playlist sources,
expands shorthands; keeps locator_to_ytdlp_url's no-playlists contract
- Playlist capture block: sync-aware container resolution
- sync + existing container → reuse it via complete_archive_run_item,
skip already-archived children (by canonical URL) before creating
run items so refresh_run_counters only counts new items
- sync + no container → create normally (first sync run)
- non-sync → always create fresh container (existing behaviour)
- Per-item quality: config.per_item_quality.get(id) falls back to child_quality
archive.rs:
- Add get_archived_playlist_child_urls(): returns HashSet of canonical URLs
of all children under any container matching the playlist canonical URL
- Add find_container_entry_id_by_canonical_url(): returns most-recent
container entry id (parent_entry_id IS NULL) for a given canonical URL
routes.rs: stub per_item_quality/sync on both CaptureConfig sites (server
agent will wire body fields in Phase 2)
* fix(core)+test: propagate sync query errors; cover new playlist/sync functions
archive.rs:
- get_archived_playlist_child_urls: collect() as rusqlite::Result<HashSet<_>>
instead of filter_map(ok) so row-level errors surface rather than silently
skipping and causing duplicate downloads
capture.rs:
- match get_archived_playlist_child_urls result and fail_run on error instead
of unwrap_or_default, preventing silent re-downloads on DB failure
Tests added to capture.rs:
- locator_to_playlist_url_accepts_playlist_shorthands (yt:playlist/, ytm:playlist/, full URL)
- locator_to_playlist_url_accepts_channel_shorthands (yt:@handle)
- locator_to_playlist_url_rejects_non_playlist_sources (single video, tweet, web page)
Tests added to archive.rs (all use in-memory DB via make_tag_test_db):
- find_container_entry_id_returns_none_when_absent
- find_container_entry_id_returns_root_entry
- find_container_entry_id_ignores_child_entries (child with parent_entry_id set)
- get_archived_playlist_child_urls_empty_when_no_playlist
- get_archived_playlist_child_urls_returns_children
- get_archived_playlist_child_urls_excludes_other_playlists
* feat(server,frontend): playlist quality selector + per-video overrides + sync UI
routes.rs:
- CaptureBody gains per_item_quality (HashMap<String,String>, serde(default))
and sync (bool, serde(default)); both validated before use
- per_item_quality values validated against same quality predicate as top-level
quality field ("best"|"audio"|"NNNp") so bad per-video values are rejected
at the API boundary rather than silently falling through to quality_format
- capture_handler threads body.per_item_quality + body.sync into CaptureConfig
(replaces hardcoded empty stubs); rearchive_handler keeps empty defaults
- New POST /api/archives/:id/captures/probe-playlist: calls
probe_playlist_qualities via spawn_blocking; 400 for non-playlist locator,
502 on yt-dlp failure, returns PlaylistProbeResult as JSON
api.js:
- probePlaylist(archiveId, locator): POST probe-playlist endpoint
- submitCapture: forwards per_item_quality (non-empty) and sync:true from
extraExtensions param added to submitBgJob
CaptureDialog.jsx:
- isPlaylistSource(): detects yt:/youtube: playlist/@/channel, ytm:playlist/,
YouTube/YTM HTTP(S) URLs with list= param or channel pathnames
- makeItem(): 6 new playlist state fields
- applyPlaylistQuality(): conflict logic — videos that can reach selected
quality get it set; videos that can't and have no prior selection are left
null (conflict); videos with a prior selection keep it when quality is raised
- hasConflict(): any playlistItems entry with quality===null
- updateLocator(): isPlaylistSource branch with 800ms debounce→probePlaylist;
existing isVideoSource path unchanged
- Archive button disabled when anyConflict or any probe in flight
- Per-video expand list with individual quality selects, conflict badges,
sync toggle (appears after probe completes)
styles.css: playlist expansion, conflict, sync toggle CSS
* fix(core): ignore per_item_quality for YTM playlist items
YouTube Music playlists force child_quality = Some("audio") because
yt-dlp can't download DRM-free audio-only tracks any other way. The
previous per_item_quality lookup could override this with e.g. "best",
defeating the invariant. Guard the lookup behind !is_audio so YTM items
are always downloaded as audio regardless of what the caller sends.
* fix(frontend): exclude /watch from isPlaylistSource
youtube.com/watch?v=...&list=... and music.youtube.com/watch are single
videos in the backend (Source::YouTubeVideo / YouTubeMusicTrack) regardless
of the list param. Previously isPlaylistSource returned true for these,
which would have triggered the playlist probe path while the video probe
was already running, and the render would attempt to show playlist UI on
an item whose playlistProbeState stays idle.
Guard: if pathname === '/watch', return false before the list-param check.
* fix(frontend): tighten isPlaylistSource to mirror backend routing exactly
Previous fix excluded /watch but still returned true for any youtube.com
URL with a ?list= param (e.g. /shorts/xxx?list=yyy). Backend determine_source
only routes to YouTubePlaylist on /playlist?list=... and to YouTubeChannel on
/@handle, /channel/, /c/, /user/ paths — everything else is a single item.
Rewrite the HTTP block to match:
- youtube.com: pathname==='/playlist' && list param → playlist
: /@, /channel/, /c/, /user/ → channel
: anything else (incl. /watch&list=, /shorts?list=) → false
- music.youtube.com: pathname==='/playlist' && list param → YTM playlist
: /watch → single track (falls through to false)
* fix(frontend): guard handleArchive against Enter-key bypass of disabled state
The Archive button is disabled when anyConflict || anyProbing, but
onKeyDown on the locator input calls onSubmit() → handleArchive()
directly, bypassing the button's disabled check entirely.
Add the same conditions as early returns inside handleArchive itself,
operating on toSubmit (the items that would actually be submitted) so
the guard is tight — items with no locator are already excluded by the
toSubmit filter.
* fix(frontend): drop m.youtube.com from isPlaylistSource
Backend determine_source playlist/channel regex only matches
(?:www\.)?youtube\.com — mobile URLs hitting m.youtube.com would be
probed as playlist in the UI but captured as Source::Url server-side.
Remove m.youtube.com from the detector to keep frontend and backend
in exact agreement. Add backend support when needed.
* fix(frontend): audio-only conflict handling in playlist quality selector
applyPlaylistQuality('audio'):
- Only sets quality='audio' on items where has_audio=true
- Items with has_audio=false: keep prior selection if set, else null
(conflict) — same rule as unsupported height, blocks archive until
user explicitly picks a quality for those items
Playlist-level 'Audio only' option:
- Changed hasAnyAudio → allHaveAudio (every item must have audio)
- When any item lacks audio, the option is hidden entirely so the
selector can never create immediate conflicts just by appearing
* fix(frontend): add yt:user/ to isPlaylistSource shorthand detection
Backend determine_source routes yt:user/... (and youtube:user/...) to
YouTubeChannel — already covered by the yt: shorthand block for
playlist/, @, channel/, c/ but missing user/. Old-style user channel
URLs would capture correctly server-side but never show the playlist
quality/sync UI.
* fix(frontend): block playlist submission unless probe is done
Previous guard only blocked while playlistProbeState==='probing'.
Two remaining bypass paths:
- idle: 800ms debounce not yet fired after URL typed
- error: probe failed — no per-video quality data available
Change anyProbing and handleArchive guard to:
isPlaylistSource(locator) && playlistProbeState !== 'done'
This means idle/probing/error all block submission for playlist items.
error is intentionally blocking — without quality data the per-video
requirement can't be satisfied; user must retry or remove the URL.
* fix(frontend): accurate error message when playlist probe fails
Previous text said 'using best quality' implying the capture would
proceed, but probe error now blocks submission. Replace with 'Probe
failed — edit URL to retry' in orange (capture-quality-hint--error)
so the disabled button and the message are consistent.
* fix(frontend): exact quality match in applyPlaylistQuality
Replace maxHeight >= newHeight (cap check) with item.qualities.includes(newQ)
(exact match). A video with [2160p, 1080p] does not support 1440p; the
previous logic would mark it as supporting any quality up to 2160p and
submit '1440p' which yt-dlp silently downloads as 1080p — misrepresenting
the selected quality.
With exact match, unsupported qualities correctly fall through to the
conflict path (keep prior selection or null), enforcing the same manual-
choice requirement as any other unsupported quality.
Per-row selects are unaffected: they already render only pi.qualities
(the video's actual available formats), no maxHeight logic involved.
* fix(frontend): move playlist expand chevron to left of input
User asked for the chevron to be on the left of the playlist input,
not tucked after the quality selector on the right.
- Remove chevron from qualityEl (it was between the quality select and
the remove button)
- Add it as the first child of capture-row-main, before the <input>,
when isPlaylistSource && playlistProbeState === 'done'
- Show a same-width placeholder span while probing/idle/error so the
input does not jump left when the chevron appears after probe completes
- Add capture-playlist-toggle--left modifier (flex-shrink:0, tighter
padding) and .capture-playlist-toggle-placeholder (fixed 22px width)
* doc(server): scope per_item_quality guarantee to the UI
The server validates per_item_quality value shapes but does not enforce
that every playlist item has an entry. Items without an override get
the global quality as a yt-dlp cap with graceful fallback.
The 'must choose quality for unsupported videos' invariant is a UI
constraint enforced by the frontend before submission. A direct API
caller bypassing the UI accepts yt-dlp's standard cap-and-fallback
behavior. Document this scope explicitly so the gap is intentional,
not accidental.
* fix(frontend): prevent 'reading some of undefined' crash from stale sessionStorage
Old captureItems entries saved before the playlist fields were added
have playlistItems=undefined (missing key). The hasConflict guard
checked !== null, which undefined passes, then called .some() on
undefined → TypeError.
Two-part fix:
1. sessionStorage restore: merge each saved item over makeItem() defaults
so any missing fields (playlistItems, playlistProbeState, etc.) are
filled with their correct initial values before the item is used
2. hasConflict: use Array.isArray() instead of !== null so undefined
is also safely rejected — defence-in-depth for any future field gap
* fix(core): playlist total size includes children; per_item_quality as include-set
archive.rs: total_artifact_bytes for root entries now adds a correlated
subquery summing children's blob bytes so playlist/channel containers
show the real download size instead of 0.
capture.rs: non-empty per_item_quality map now acts as an include-set —
items whose yt-dlp ID is absent are skipped entirely. This wires the
UI's per-video delete button to actual capture exclusion. Empty map
preserves the existing behaviour (download everything).
routes.rs + capture.rs doc: comments updated to reflect both semantics
(empty = all / non-empty = only listed IDs) accurately.
* fix(frontend): QA fixes — playlist UX, selection stroke, URL expand
CaptureDialog.jsx:
- Placeholder no longer appears on idle playlist rows (only during
probing); chevron shows only after probe completes — no left-padding
while the user is still typing
- Per-video delete button added to expanded playlist list; removes item
from playlistItems so its ID is absent from per_item_quality on submit
- anyEmptyPlaylist guard: Archive button disabled + handleArchive early-
return when all videos have been deleted (empty map would otherwise
silently download everything)
styles.css:
- Selection stroke switched from outline to box-shadow:inset everywhere
(entry-row-outer, child-entry-row, legacy flat-div selector) —
guaranteed inside element bounds, no layout interference
- URL cell overflow only on :hover; removed is-selected .url-cell rule
that was expanding child URLs when their parent was selected
- Playlist items redesign: separator lines instead of background fills,
amber left-border for conflicts, thin scrollbar, tighter padding
- Remove button: opacity 0.3 always-visible baseline; full opacity on
hover/:focus-visible; forced to 1 on coarse-pointer (touch) devices
* fix(frontend): no left gap on playlist rows until chevron exists
Remove the probing-state placeholder span entirely. The chevron renders
only when playlistProbeState === 'done'; all other states (idle, probing,
error) render null. The small layout shift when the chevron appears after
probe is acceptable; blank padding while there is no chevron is not.
* fix(frontend): clear box-shadow on entry-row-outer to prevent double stroke
The flat selector '#entries-body > div.is-selected' already applies
box-shadow to .entry-row-outer (it IS a direct child div). The outer
override rule only cleared 'outline', so the box-shadow leaked through,
wrapping the entire parent+children block with a second stroke.
Add box-shadow: none to both .is-selected and .is-multi-selected on
.entry-row-outer so the stroke sits only on .entry-row-main.
* docs: document per-video exclude in README YouTube playlists section
* fix(frontend): scope selection background to entry-row-main only
Moving background:#eee2d2 off .entry-row-outer onto > .entry-row-main
so that expanded child entries don't inherit the selection highlight.
Outer wrapper now explicitly unsets background (cancelling the flat-div
cascade rule) and clears outline+box-shadow. Both the selection colour
and the inset stroke live on .entry-row-main only.
* fix(frontend): alternating stripe backgrounds on child entry rows
Child rows were inheriting the parent entry's stripe color, making the
expanded list look like one flat block. Apply the same odd/even palette
as root entries (var(--paper-3) / #f2ede5) so each video row is visually
distinct within the expanded group.
* fix(core): delete_entry correctly nulls FK for child entries
archive_run_items.produced_entry_id has no ON DELETE action so it must
be manually nulled before the entry row is deleted. The old query used
WHERE root_entry_id = entry_id, which finds descendants of a root but
returns nothing when entry_id IS a child (children have no sub-children,
so no row has root_entry_id = child_id). The DELETE then failed under
foreign_keys=ON.
Fix: add OR produced_entry_id = ?1 so the entry's own run_item FK is
always cleared before deletion, regardless of whether it is a root or a
child. The subtree subquery is kept for the root-deletion case where all
child run_items also need nulling.
* fix(core+frontend): child entry selection and delete correctness
database.rs — delete_entry:
- subtree_ids now uses WHERE id = ?1 OR root_entry_id = ?1 so the entry
itself is always included; previously a child deletion passed an empty
vec to cascade_cached_bytes_after_subtree_delete (no grandchildren
exist), leaving cached_bytes stale on entries sharing that child's blobs
App.jsx — handleRowClick:
- Shift-range now queries DOM order (#entries-body [data-entry-uid])
instead of entries.findIndex(); child rows are in the DOM but not in
the root entries array, so findIndex always returned -1 for them
- Ctrl/meta branch computes next set before the state update so
selectEntry can fire synchronously for child rows; auto-snap only
restores root entries, so ctrl-clicking a lone child never loaded its
detail panel — now calls selectEntry(entry) when the child ends up as
the sole selection, selectEntry(null) on multi or deselect
- selectedUids added to handleRowClick's useCallback dep array
* fix(frontend): resolve detail entry for any remaining child after ctrl-deselect
Add entryCacheRef (uid→entry Map) populated on every row click. When
ctrl/meta-deselecting leaves exactly one other entry selected, look up
the remaining UID in the cache before falling back to the root entries
array. Without this, deselecting from a multi-selection where the
remaining entry is a child row left selectedEntry null (auto-snap only
searches root entries).
* fix(frontend): child row stripes, deleted-child visibility, selection completeness
EntryRow.jsx / ChildRow:
- Index-based light/dark classes (child-entry-row--light/dark) replace
nth-child rules; parity comes from children.map idx so no sibling —
including the loading div — can shift the stripe order
- Loading div moved outside .child-entries so it never affects child
row ordering at all
- Accept deletedUids prop; filter expanded children array before render
so deleted children disappear immediately without waiting for reload
EntriesView.jsx: thread deletedUids through to EntryRow
App.jsx:
- Add deletedUids state; handleEntryDeleted/handleBulkDeleted populate it
- isRoot/hasChildDelete computed from entries before setEntries (safe in
StrictMode — no side-effects inside updater functions)
- Child delete triggers loadEntries to refresh stale parent child_count
and total_artifact_bytes
- handleRowClick ctrl/meta cache-miss branch uses det.summary (not det)
from fetchEntryDetail; archiveId added to dep array
- handleRowClick dep array includes archiveId
* fix(frontend): add inset stroke to child-entry-row.is-multi-selected
* fix(frontend): suppress mouse-click focus ring on entry expand button
* fix(frontend): index-based stripes for root and child rows
Root rows: EntriesView passes rowIndex from entries.map to EntryRow,
which applies entry-row-outer--light/dark. Retires both nth-child stripe
blocks so skeleton rows can never shift the first real entry to dark.
Skeleton rows keep a :not(.entry-row-outer) nth-child fallback.
Child rows: colours changed from the warm root palette (paper-3/#f2ede5)
to cooler near-whites (#fafaf8/#f2f0ec) so children are visually distinct
from their parent row regardless of which stripe the parent sits on.
* feat(core+frontend): bare yt:ID resolves to YouTube video
determine_source: yt:ID / youtube:ID with no prefix and exactly 11
chars [A-Za-z0-9_-] → YouTubeVideo via is_youtube_video_id helper.
Reserved prefixes (playlist/, channel/, c/, user/, @) still fire first
so they are unaffected by the fallback.
expand_shorthand_to_url: bare yt:ID expands to watch?v=ID using the
same predicate, consistent with how ytm:ID → music.youtube.com/watch.
isVideoSource (frontend): same 11-char /^[A-Za-z0-9_-]{11}$/ regex so
yt:ID triggers the quality probe and capture-row guards identically to
yt:video/ID.
Tests: test_is_youtube_video_id covers valid IDs (alphanumeric, with _
and -), too-short, too-long, and invalid-char cases using genuinely
invalid fixtures. test_youtube_sources adds bare-ID cases and confirms
reserved prefixes (playlist/, @) are not affected.
* fix(core+frontend): exclude avatar blobs from tweet % cached display
tweet/tweet_thread entries have avatar artifacts that are always
deduplicated from the first capture of each author. Counting them in
the cached-bytes percentage makes it artificially high (or incorrect
when the real media content is new but avatars are cached).
database.rs:
- refresh_entry_cached_bytes: AND ea.artifact_role != 'avatar'
- cascade_cached_bytes_after_delete: same filter
- cascade_cached_bytes_after_subtree_delete: same filter
- Initial cached_bytes migration: same filter
- Re-migration (else branch): recomputes cached_bytes for existing
entries that have avatar artifacts, scoped to only those entries
archive.rs:
- EntrySummary gains cacheable_bytes: i64 — non-avatar total bytes,
computed inline in every SQL query as the denominator for % cached
- ENTRY_SELECT_COLS adds cacheable_bytes at index 13 (with children
subquery, same as total_artifact_bytes)
- list_root_entries adds same expression
- All 6 row-mapping closures include cacheable_bytes: row.get(13)?
EntryRow.jsx:
- SIZE display stays: formatBytes(entry.total_artifact_bytes)
- % cached badge uses cacheable_bytes as denominator:
cached_bytes / cacheable_bytes * 100
* feat(frontend): j/k keyboard navigation for entries; test avatar cached_bytes
App.jsx — j/k handler:
- Fires on keydown when not focused on INPUT/TEXTAREA/SELECT/contenteditable
- Ignores meta/ctrl/alt modifier combos
- Uses DOM order (#entries-body [data-entry-uid]) so expanded child rows
participate, matching the shift-range selection logic
- Resolves target entry via entryCacheRef → root entries array →
fetchEntryDetail(..).summary on cache miss
- Scrolls target into view (block: nearest); updates lastAnchorIndexRef
so subsequent shift-click ranges start from the keyboard-navigated row
archive.rs — cached_bytes_excludes_avatar_blobs test:
- Two tweet entries sharing an avatar blob (100B) and a media blob (900B)
- Asserts refresh_entry_cached_bytes sets cached_bytes = 900 (not 1000)
- Asserts list_root_entries summary: cached_bytes=900, cacheable_bytes=900,
total_artifact_bytes=1000 — SIZE includes avatar, % cached denominator
and numerator both exclude it
* fix(frontend): guard j/k uncached-child fetch with monotonic token
Rapid j/k over child rows that aren't in entryCacheRef triggers
fetchEntryDetail calls in parallel. Without a guard the last one to
settle wins, desyncing selectedUids (highlight) from selectedEntry
(detail panel/URL).
Fix:
- jkSeqRef (monotonic counter) incremented before each server fetch;
the .then() guard tok === jkSeqRef.current drops results from
superseded navigations
- handleRowClick increments jkSeqRef.current so any click also cancels
an in-flight j/k fetch
* fix(frontend): guard ctrl/meta cache-miss fetch; add / search shortcut
App.jsx ctrl/meta branch: cache-miss fetchEntryDetail now captures
tok = ++jkSeqRef.current before the fetch and gates selectEntry on
tok === jkSeqRef.current — same pattern as the j/k handler — so a
slow response after a later click/navigate can't overwrite selection.
/ key: reuses the Cmd+K/Ctrl+K handler path (focus+select search input,
or pendingSearchFocus + archive view switch). Guards: editable targets
(INPUT/TEXTAREA/SELECT/contentEditable) and modifier keys are checked
before preventDefault() so typing / in inputs is untouched.
* feat(core): attempt SpotifyTrack via yt-dlp instead of hard-failing
Previously all Spotify sources returned an error claiming yt-dlp cannot
download DRM-protected audio. yt-dlp does have a Spotify extractor
(experimental, content-dependent), so refusing upfront is worse than
trying and letting it report the real failure.
SpotifyTrack now goes through the same yt-dlp metadata + audio-quality
download path as YouTubeMusicTrack. SpotifyAlbum and SpotifyPlaylist
still return an explicit error — fetch_playlist_info has YouTube-specific
URL fallback logic that would produce bogus child URLs for Spotify flat
entries; those sources need dedicated container handling first.
* chore: remove docs/superpowers from repo and gitignore whitelist
* feat(core+frontend): Spotify album/playlist capture via yt-dlp container path
Previously SpotifyAlbum and SpotifyPlaylist hard-failed with a DRM error.
yt-dlp does support Spotify (experimentally), so they now go through the
same probe/container/child path as YouTube playlists.
ytdlp.rs:
- Extract normalize_item_url() helper used by both fetch_playlist_info
and probe_playlist_qualities; eliminates the duplicate URL-normalization
blocks and the YouTube-specific fallback_host variable
- Fallback logic is now platform-aware: YouTube/YTM bare IDs → watch URL,
Spotify → open.spotify.com/track/{id}, unknown → skip with warning
capture.rs:
- locator_to_playlist_url: accept SpotifyAlbum | SpotifyPlaylist so the
probe-playlist endpoint accepts Spotify album/playlist URLs
- Container branch: add SpotifyAlbum | SpotifyPlaylist to the matches!
- is_audio: true for SpotifyAlbum | SpotifyPlaylist (audio-only, like YTM)
- child_source: SpotifyTrack for Spotify containers (not YouTubeMusicTrack)
- generate_entry_title and record_media_entry both use the correct child
source so entity_kind, source_kind, and representation_kind are right
CaptureDialog.jsx:
- isPlaylistSource: recognise open.spotify.com/album/ and /playlist/,
and spotify:album:ID / spotify:playlist:ID shorthands, so Spotify
containers get the probe UI, per-track excludes, and sync toggle
* fix(core+server): partial playlist refresh + child visibility inheritance
database.rs:
- finish_archive_run: reverted 'partial' status (violates CHECK constraint);
restored binary completed/failed
- get_run_completed_count(): new helper for callers that need to distinguish
partial success without touching the DB status enum
capture.rs:
- CaptureResult gains completed_count: i64 (0 for single-item captures;
populated from get_run_completed_count for container/playlist captures)
- All CaptureResult construction sites updated
routes.rs:
- Playlist job_status mapping: mark job 'completed' when completed_count > 0
(even if status == 'failed'), so partially-successful playlist captures
trigger onCaptured and show the archived entries without a manual reload.
Truly zero-success runs (completed_count == 0, status == 'failed') stay
failed as expected.
- probe_playlist_handler error message updated to mention Spotify
archive.rs (list_child_entries):
- Add third OR arm: children are visible when their parent container is in a
collection visible to the caller. Fixes empty child list for non-admin
users with access to a playlist container but not its newly-created
children (which are all in the default 'private' collection).
* fix(server): use completed_count to distinguish partial from total failure
finish_archive_run is binary (completed/failed) — a playlist where some
tracks succeed and some fail returns status='failed'. Without this fix,
the job mapping treated any failed status as a job failure, causing
onCaptured to never fire and leaving successfully archived entries hidden
until a manual reload.
Now: job is 'failed' only when status='failed' && completed_count==0.
Partial runs (completed_count > 0) map to a completed job so the UI
refreshes and shows the entries that were successfully captured.
* fix(core+server): exclude container from child success count; rename field; tests
database.rs:
- get_run_completed_child_count(): renamed from get_run_completed_count and
scoped to child items only (parent_item_id IS NOT NULL). The container
run item is always completed first, so the old function inflated the count
by at least 1 for every playlist, making all-video-failed runs appear as
partial successes.
- New regression test: completed_child_count_excludes_container — completes
both a root and a child item, asserts DB completed_count==2 while
get_run_completed_child_count==1.
capture.rs:
- CaptureResult.completed_count renamed to completed_child_count to match
the function and make the semantics unambiguous at the call site.
routes.rs:
- job_status decision now uses result.completed_child_count == 0 so that a
playlist where every video fails (child_count==0) is correctly reported
as a failed job, not a completed one.
archive.rs:
- New regression test: list_child_entries_inherits_parent_visibility —
enrolls only the container in a USER-visible collection, asserts the
child is visible to a USER caller and invisible to a GUEST caller.
* feat(frontend): add SkeletonEntryRow with shimmer animation
Adds a new SkeletonEntryRow component that renders animated shimmer
placeholder cells matching the exact column layout of EntryRow
(col-added, col-title with icon circle, col-type pill, col-size,
col-url). CSS appended to styles.css using existing design tokens
(--paper-2, --line-soft, --paper-3) for the warm-toned shimmer.
* feat(frontend): async capture UX — reset dialog on submit, show skeleton rows
When the user presses Archive, CaptureDialog now immediately resets its
form to a fresh empty state and closes. Background captures continue
polling via intervals that survive the dialog close.
Skeleton rows appear at the top of EntriesView (filtered to the active
archive) for each in-flight job, giving visual feedback that something
is being processed. On completion the skeleton is removed and the entry
list refreshes; on failure the skeleton is removed and an error toast
fires — matching the existing toast behaviour.
Architecture:
- App owns pending-capture state (pendingCaptures) as the single source
of truth, persisted to sessionStorage['pendingCaptures']. On page
refresh, App seeds the list and passes it to CaptureDialog as
activeJobs so polling reconnects without a second sessionStorage read.
- CaptureDialog emits onJobStarted({id,jobUid,locator,archiveId}) only
after submitCapture() returns a job_uid — never before, never on
failure — ensuring no orphan skeletons can persist after a refresh.
- onJobSettled(id) is called by startPolling on any terminal state
(completed, failed, or network error), removing the skeleton.
- EntriesView filters pendingCaptures by archiveId so a capture in
archive A never shows a skeleton in archive B.
Removed from CaptureDialog: submitItem, resetRow, hasActiveJobs,
anyActive, CapStatusDot. CaptureRow is simplified to idle-only display
with no disabled state, no retry button, no status dot.
* fix(frontend): skeleton replacement ordering and col-check alignment
- Await the entry list refresh before removing the skeleton so the
real row arrives before the placeholder disappears. handleCaptured
now returns its Promise.all; startPolling awaits it before calling
onJobSettled on the success path.
- Add col-check as the first child of SkeletonEntryRow to match
EntryRow's DOM structure; without it, columns misalign on touch
devices where col-check becomes display:flex.
* fix(frontend): use Promise.allSettled in handleCaptured to prevent runs-fetch failure from poisoning successful captures
Promise.all rejects on the first failure. If the /runs refresh threw,
the rejection would propagate through startPolling's await and land in
the outer catch block, firing an error toast for a capture that had
already succeeded. Promise.allSettled settles unconditionally so a
transient runs fetch error is silently absorbed while the entry list
still refreshes.
Replace the broad previousElementSibling/HEADER removal with a
precise selector that matches only the .flex.justify-end wrapper
containing the 'Download article' button. The old heuristic was
removing article headers on NYT/WaPo captures.
* feat(core): add via_freedium to CaptureConfig; route WebPage captures through Freedium mirror
When via_freedium is true and the locator is not already a Freedium URL,
perform_capture passes https://freedium-mirror.cfd/<original-url> to
singlefile::save() so paywalled articles are fetched via the mirror.
The original locator is kept for requested_locator and canonical_locator
in the DB entry. An empty cookie map is used for the mirror fetch to
prevent original-domain credentials from being sent to freedium-mirror.cfd.
* feat(server): expose via_freedium in capture API (default on)
CaptureBody gains via_freedium: Option<bool>; absent defaults to true.
The rearchive handler sets it to false — existing entries should not be
silently re-fetched through a mirror.
* feat(frontend): add Freedium mirror toggle to capture advanced options
- freediumEnabled state defaults to true (on by default)
- via_freedium forwarded through submitCapture to the capture API
- Toggle rendered last in the advanced panel, matching existing rows
- Built frontend static assets included
* fix(core): Freedium capture fixes — title, notifications, reader images
- extract_html_title: take().read_to_end() up to 256 KiB (single read() can
short-read, leaving title at byte ~106 K undiscovered); regression test added
- Strip " - Freedium" suffix before storing entry title
- is_freedium_fetch keys off actual fetch_url host
- Remove Freedium toast overlay ([data-sonner-toaster]) before capture
- Resolve lazy images (data-zoom-src etc.) before Readability in reader mode
* fix(core): extract HTML title after font stripping, not before
SingleFile embeds fonts as base64 data URIs in <style> blocks in the
<head>, pushing the <title> tag to ~1.2 MB in the raw temp file.
The 256 KiB read window in extract_html_title missed it.
Font extraction rewrites the HTML in-place (2.4 MB → 1.26 MB) before
hashing, so the title is accessible at byte ~106 KB in that content.
Fix: add extract_html_title_str() that operates on a &str; call it on
the in-memory rewritten string after font extraction (server path).
CLI path (no font extraction) falls back to result.title as before.
* fix(core): inline reader-mode images via Rust post-processing
SingleFile cannot inline resources added to the DOM at before-capture
time — only resources tracked during the page's initial load cycle get
embedded. Article images in Readability output fall into this gap when
the page framework has already resolved their lazy src to a CDN URL
that isn't in SingleFile's resource cache for the new DOM elements.
Fix in three parts:
- Browser script: after body.innerHTML = article.content, stamp
data-archivr-src=<absolute-proxy-url> on any image whose src is
not an already-inlined large data URI. Remove loading attr. Don't
touch src (let SingleFile try; Rust handles the rest).
- Rust (save_with): after SingleFile writes the file and before hashing,
call inline_archivr_img_srcs() to scan for data-archivr-src markers.
- inline_archivr_img_srcs(): fetches each marked URL with blocking
reqwest (10 s timeout, 5 redirects, image/* Content-Type guard,
20 MiB cap), base64-encodes, replaces src with the data URI, and
removes the marker attr. Non-fatal — fetch failures are logged.
Also: Freedium UI cleanup now removes footer, #progress, and empty
data-nosnippet wrappers in addition to the existing nav/toaster removal.
* fix(core): clean up Freedium chrome and drop reader-mode debug counters
Two fixes:
1. freedium_cleanup: remove remaining Freedium article chrome that
survived the existing nav/footer/toaster pass:
- <header class="p-6 bg-gray-50 ..."> — author/metadata bar with
profile pictures and byline, a Freedium wrapper around the article
- <section> containing [data-slot="dropdown-menu-trigger"] or
[aria-haspopup="menu"] — the "Download article" dropdown
Selectors target stable Tailwind utility class prefixes (p-6, bg-gray-50,
bg-zinc-800) for the header and the WAI-ARIA menu role for the button,
both resilient to minor Freedium UI updates.
2. Reader-mode meta tag: strip per-capture debug counters.
_archivrReaderMark now writes 'applied' instead of
'applied:pre_s=N,...:post_s=N,...' — the counters were useful during
development but have no place as permanent archive content.
* fix(core): prevent lazy-resolver double-processing on Freedium reader images
_archivrResolveLazyImgs is called twice: before Readability (_pre) and
after the post-body stamp pass (_post). The stamp pass sets data-archivr-src
but previously left data-src/data-zoom-src/etc intact, so _post's
'lazySrc && _isPlaceholder' condition fired on the same images and
rewrote src to the CDN URL — which SingleFile then tried and failed to
fetch from the Freedium proxy context, redundantly.
Fix: strip all lazy attrs (data-src, data-lazy-src, data-zoom-src,
data-original, data-lazy) when stamping data-archivr-src. _post then
finds no lazySrc on those images and skips them cleanly. Rust owns
them via data-archivr-src. _post continues to handle any remaining
placeholder images not covered by the stamp pass (non-Freedium reader
captures).
* fix(core): address Codex review findings in reader image post-processor
P1 — Enforce size cap before buffering (singlefile.rs fetch_image_as_data_uri):
resp.bytes() buffered the entire response before checking MAX_BYTES, enabling
OOM on oversized or attacker-controlled images. Fix: reject via Content-Length
header when present, then stream at most max_bytes+1 bytes with Read::take so
the cap is enforced without materialising the full body first.
P2 — Decode HTML entities in extracted image URLs (inline_archivr_img_srcs):
The browser's HTML serialiser encodes & as & in attribute values, so CDN
signed URLs with query parameters (e.g. ?a=1&b=2) arrived as &-escaped
strings. reqwest sent the wrong URL, breaking signed or transformed CDN
images. Fix: unescape & < > " before passing to the fetcher.
P2 — Preserve authentication for same-origin lazy images (inline_archivr_img_srcs):
The post-processor created a bare reqwest client with no cookies, causing 401/403
for reader-mode captures of auth-gated sites whose lazy images are same-origin.
Fix: pass capture_url and cookies into inline_archivr_img_srcs; attach a Cookie
header only when domain_from_url(img_url) == domain_from_url(capture_url) and
cookies is non-empty. Third-party image hosts and Freedium fetches (which receive
empty cookies in capture.rs) are unaffected.
* test(core): extract helpers and add unit tests for Codex review fixes
Extract two testable pure functions from inline_archivr_img_srcs:
- html_attr_decode: decodes HTML character references in attribute values.
Fixes decode order: & runs LAST so &lt; → < (one layer
removed), not < (two layers). Previous order caused double-decoding.
- same_origin_cookie_header: returns a Cookie header value only when
img_url's domain matches capture_url's domain.
Add 11 unit tests covering:
- html_attr_decode: plain URL no-op, & in CDN query params, single-layer
decode (&lt; → < not <), double-encoded amp (&amp; → &),
direct </>/" decode
- same_origin_cookie_header: same host attaches cookies, third-party returns
None, empty cookies returns None, Freedium empty-cookies no-op
- bounded_read: Read::take stops at max+1 bytes (guard fires), allows
exactly-at-limit payloads (guard does not fire)
* docs: refresh AGENTS.md and mental model; drop NEXT.md
AGENTS.md: mention Freedium mirror in the overview and capture flow,
add vendor/readability/ to Key Directories, drop the NEXT.md pointer.
ARCHIVR-MENTAL-MODEL.md: fix stale references to legacy static/ paths
in Where To Edit (frontend now lives in frontend/src/); add capture.rs
and auth.rs rows; replace the outdated 'Current Limitations' section
(which claimed no capture and no auth) with a Server Capabilities
section reflecting async capture jobs, the auth model, search, and
admin scope; add a Web Capture Pipeline section covering the Freedium
mirror, SingleFile+Chromium, vendored Readability, cleanup, and Rust
post-processing.
NEXT.md: remove; the roadmap tracking is stale and unused.
AGENTS.md: mention Freedium mirror in the overview and capture flow,
add vendor/readability/ to Key Directories, drop the NEXT.md pointer.
ARCHIVR-MENTAL-MODEL.md: fix stale references to legacy static/ paths
in Where To Edit (frontend now lives in frontend/src/); add capture.rs
and auth.rs rows; replace the outdated 'Current Limitations' section
(which claimed no capture and no auth) with a Server Capabilities
section reflecting async capture jobs, the auth model, search, and
admin scope; add a Web Capture Pipeline section covering the Freedium
mirror, SingleFile+Chromium, vendored Readability, cleanup, and Rust
post-processing.
NEXT.md: remove; the roadmap tracking is stale and unused.
Extract two testable pure functions from inline_archivr_img_srcs:
- html_attr_decode: decodes HTML character references in attribute values.
Fixes decode order: & runs LAST so &lt; → < (one layer
removed), not < (two layers). Previous order caused double-decoding.
- same_origin_cookie_header: returns a Cookie header value only when
img_url's domain matches capture_url's domain.
Add 11 unit tests covering:
- html_attr_decode: plain URL no-op, & in CDN query params, single-layer
decode (&lt; → < not <), double-encoded amp (&amp; → &),
direct </>/" decode
- same_origin_cookie_header: same host attaches cookies, third-party returns
None, empty cookies returns None, Freedium empty-cookies no-op
- bounded_read: Read::take stops at max+1 bytes (guard fires), allows
exactly-at-limit payloads (guard does not fire)
P1 — Enforce size cap before buffering (singlefile.rs fetch_image_as_data_uri):
resp.bytes() buffered the entire response before checking MAX_BYTES, enabling
OOM on oversized or attacker-controlled images. Fix: reject via Content-Length
header when present, then stream at most max_bytes+1 bytes with Read::take so
the cap is enforced without materialising the full body first.
P2 — Decode HTML entities in extracted image URLs (inline_archivr_img_srcs):
The browser's HTML serialiser encodes & as & in attribute values, so CDN
signed URLs with query parameters (e.g. ?a=1&b=2) arrived as &-escaped
strings. reqwest sent the wrong URL, breaking signed or transformed CDN
images. Fix: unescape & < > " before passing to the fetcher.
P2 — Preserve authentication for same-origin lazy images (inline_archivr_img_srcs):
The post-processor created a bare reqwest client with no cookies, causing 401/403
for reader-mode captures of auth-gated sites whose lazy images are same-origin.
Fix: pass capture_url and cookies into inline_archivr_img_srcs; attach a Cookie
header only when domain_from_url(img_url) == domain_from_url(capture_url) and
cookies is non-empty. Third-party image hosts and Freedium fetches (which receive
empty cookies in capture.rs) are unaffected.
_archivrResolveLazyImgs is called twice: before Readability (_pre) and
after the post-body stamp pass (_post). The stamp pass sets data-archivr-src
but previously left data-src/data-zoom-src/etc intact, so _post's
'lazySrc && _isPlaceholder' condition fired on the same images and
rewrote src to the CDN URL — which SingleFile then tried and failed to
fetch from the Freedium proxy context, redundantly.
Fix: strip all lazy attrs (data-src, data-lazy-src, data-zoom-src,
data-original, data-lazy) when stamping data-archivr-src. _post then
finds no lazySrc on those images and skips them cleanly. Rust owns
them via data-archivr-src. _post continues to handle any remaining
placeholder images not covered by the stamp pass (non-Freedium reader
captures).
Two fixes:
1. freedium_cleanup: remove remaining Freedium article chrome that
survived the existing nav/footer/toaster pass:
- <header class="p-6 bg-gray-50 ..."> — author/metadata bar with
profile pictures and byline, a Freedium wrapper around the article
- <section> containing [data-slot="dropdown-menu-trigger"] or
[aria-haspopup="menu"] — the "Download article" dropdown
Selectors target stable Tailwind utility class prefixes (p-6, bg-gray-50,
bg-zinc-800) for the header and the WAI-ARIA menu role for the button,
both resilient to minor Freedium UI updates.
2. Reader-mode meta tag: strip per-capture debug counters.
_archivrReaderMark now writes 'applied' instead of
'applied:pre_s=N,...:post_s=N,...' — the counters were useful during
development but have no place as permanent archive content.
SingleFile cannot inline resources added to the DOM at before-capture
time — only resources tracked during the page's initial load cycle get
embedded. Article images in Readability output fall into this gap when
the page framework has already resolved their lazy src to a CDN URL
that isn't in SingleFile's resource cache for the new DOM elements.
Fix in three parts:
- Browser script: after body.innerHTML = article.content, stamp
data-archivr-src=<absolute-proxy-url> on any image whose src is
not an already-inlined large data URI. Remove loading attr. Don't
touch src (let SingleFile try; Rust handles the rest).
- Rust (save_with): after SingleFile writes the file and before hashing,
call inline_archivr_img_srcs() to scan for data-archivr-src markers.
- inline_archivr_img_srcs(): fetches each marked URL with blocking
reqwest (10 s timeout, 5 redirects, image/* Content-Type guard,
20 MiB cap), base64-encodes, replaces src with the data URI, and
removes the marker attr. Non-fatal — fetch failures are logged.
Also: Freedium UI cleanup now removes footer, #progress, and empty
data-nosnippet wrappers in addition to the existing nav/toaster removal.
SingleFile embeds fonts as base64 data URIs in <style> blocks in the
<head>, pushing the <title> tag to ~1.2 MB in the raw temp file.
The 256 KiB read window in extract_html_title missed it.
Font extraction rewrites the HTML in-place (2.4 MB → 1.26 MB) before
hashing, so the title is accessible at byte ~106 KB in that content.
Fix: add extract_html_title_str() that operates on a &str; call it on
the in-memory rewritten string after font extraction (server path).
CLI path (no font extraction) falls back to result.title as before.
