*ARRgh v0.1.2 — .NET Migration + New Sources + Testing Layer

Released: 2026-05-30

Highlights

• Full .NET migration — Rust server replaced with C# / ASP.NET Core. The project has grown beyond a weekend experiment; the maintainer is professionally fluent in C# but cannot read or debug Rust without AI assistance, making self-recovery from incidents impossible. .NET removes that dependency.
• manga18fx source — new explicit manhwa plugin with correct search URL and sidebar-contamination fix
• E2e Playwright layer — full test suite against a real Docker Compose stack
• LOG_LEVEL env var — quiet the docker console without rebuilding
• Source priority PATCH — change source ordering via API
• manga18fx lazy-load fix — chapter pages that use data-src placeholder pattern now return correct page URLs instead of 0 pages
• MangaFire removed — plugin, seeded source, and all related tests/docs deleted; was CF-protected and unreliable without CloakBrowser
• Sync warning cleanup — amber warning badge no longer fires per-miss; only fires when zero sources linked after full sync cycle

What's new

.NET server (replaces Rust)

The Rust/Axum server has been removed. server/ is now the .NET 9 / ASP.NET Core Minimal API implementation — feature-complete, same SQLite schema, same API contract.

Why: ARRgh started as a weekend project. It has grown into something maintained long-term. The maintainer writes C# professionally and can read, debug, and recover incidents in C# without AI assistance. Rust required AI for every server change. That is not sustainable for a solo project.

See ADR 0030.

manga18fx plugin

New explicit manhwa source (default_explicit: true). Covers the gap left by CF-protected sources (Toonily, MangaFire, AsuraScans) that frequently return 502 without CloakBrowser.

• Search: /search?q= endpoint (not WordPress /?s= fallback)
• Chapters: scoped to a[href^="/manga/{slug}/chapter-"] — fixes sidebar contamination where "Most Popular Manga" chapter links from other series were being scraped into unrelated titles
• Pages: img[src*="img01.manga18fx.com/uploads"]

Source priority PATCH

PATCH /api/sources/:id now accepts a priority field. GET /api/sources returns priority in each row. Allows reordering download fallback preference without recreating sources.

LOG_LEVEL env var

Set LOG_LEVEL=warn (or debug/info/error) in docker-compose.yml to control both the Docker console output and the in-app log viewer simultaneously. Default: info.

environment:
  LOG_LEVEL: warn   # quiets per-chapter download lines and other Info noise

Source priority tuning

• AsuraScans demoted to last priority (110) among all sources — CF-protected, frequently returns no results without CloakBrowser
• MangaFire marked CF-protected in plugin-index.json — requires CloakBrowser
• MangaDex removed from manhwa ContentTypes — title name formats differ, was producing false "no match" errors on manhwa titles

E2e test layer (Playwright)

Full end-to-end test suite against a real Docker Compose stack.

• plugins/fixture/ — deterministic standalone Express fixture server (4 modes: happy path, no-match, 502, empty pages)
• docker-compose.test.yml — isolated test stack (fixture replaces plugin-host, no CloakBrowser, isolated volume)
• e2e/ — Playwright + allure-playwright with globalSetup (seeds admin → storageState)
• .github/workflows/e2e.yml — runs on push to main, results feed shared Allure report at /test-reports/

Allure dark theme

Test reports at /test-reports/ use the Obsidian Immersive palette. Applied via CSS injection after allure generate.

Setup guard + UI polish

• Setup wizard guard prevents re-registration on already-configured instances
• Reader zoom control (50–150% via popover)
• Tags rendered as pills instead of plain text
• E-Hentai gate — explicit titles require allow_explicit flag

manga18fx lazy-load support

Some manga18fx chapters serve placeholder GIF images in src and put the real CDN URL in data-src (lazy-loading pattern). The pages scraper previously only checked src, returning 0 pages for these chapters.

Fixed selectors now match both src*= and data-src*= attributes, with data-src taking precedence:

$([
  'img[src*="manga18fx.com/uploads"]',
  'img[src*="img01.manga18fx.com"]',
  'img[data-src*="manga18fx.com/uploads"]',
  'img[data-src*="img01.manga18fx.com"]',
].join(', ')).each((_, el) => {
  const src = $(el).attr('data-src') || $(el).attr('src')
  if (src?.startsWith('http')) urls.push(src)
})

MangaFire plugin removed

MangaFire is Cloudflare-protected and requires CloakBrowser to return results. Without it, every sync attempt produces false "no results" warnings and never links chapters. Removed entirely:

• plugins/mangafire/ directory deleted
• Seeded ExternalSource row removed from Program.cs
• Entry removed from plugin-index/index.json
• All contract/behavior tests removed
• Docker build steps removed

Sync warning accuracy

MatchSourcesAsync previously appended a SyncWarnings row for every source that returned no results or no title match. This created spurious amber badges on titles that were successfully linked via other sources.

New behaviour: warning only appended when zero TitleSources rows exist after the full source loop completes.

AniList synonym storage

When adding a title via the AniList source, all synonyms returned by the AniList API (including title.romaji and title.english) are now stored as TitleAliases rows. These aliases are used during MatchSourcesAsync to improve fuzzy matching across plugins that use translated or romanized titles.

Bug fixes

• Download status stuck — DownloaderService was setting status to "in_progress" instead of "downloading". Frontend polling and progress indicators were all checking "downloading" — none detected the active state.
• Manhwa titles getting no chapters — fixture plugin only handled its own source key; extended to wildcard /:source/* routes.
• manga18fx chapter contamination — a[href*="/chapter-"] matched sidebar "Most Popular Manga" links, pulling in chapters from other series (e.g. Secret Class ch.307) into unrelated titles. Selector narrowed to a[href^="/manga/${seriesId}/chapter-"].
• manga18fx search URL — plugin was using WordPress /?s= fallback; correct URL is /search?q=.
• 502 on chapter sync fails silently — sync_chapters now returns Err on 502 so a sync warning row is created instead of silently leaving 0 chapters.
• GET /api/version required auth — moved to public router; used as server-ready check in e2e globalSetup.
• GHCR build cache SAS token expiry — Docker layer cache switched from type=gha to type=registry (no SAS token expiry).
• 0 pages on manga18fx chapters using lazy-load — data-src attribute was ignored; now preferred over src placeholder
• Amber warning badge on clean syncs — per-source miss warnings fired even when another source succeeded; now only fires on total miss
• MangaFire false misses — removed source eliminates all CF-related sync noise

Removed

• Rust server — server/ is now the .NET implementation. See ADR 0030.
• Royal Road plugin — no meaningful catalogue; frequent false matches on novel titles.
• Comick plugin — stub chapter listings (no real page data), unreliable.

New API behaviour

| Method | Path | Change |
|--------|------|--------|
| PATCH | /api/sources/:id | Now accepts priority field |
| GET | /api/sources | Now returns priority per source |
| GET | /api/version | No longer requires auth |
| GET | /api/titles/:id/sync-log | Sync progress log entries |

DB migrations

No new migrations in this release. All migrations from v0.1.1 are compatible with the .NET server.

Upgrading

docker compose pull
docker compose up -d

No breaking API changes. No new DB migrations required.

Existing installs — update AsuraScans priority (optional, demotes it to last):

# Get source ID
curl -H "Authorization: Bearer <token>" http://localhost:8080/api/sources \
  | jq '.[] | select(.name=="AsuraScans") | .id'

# Update priority
curl -X PATCH -H "Authorization: Bearer <token>" \
  -H "Content-Type: application/json" \
  -d '{"enabled":true,"priority":110}' \
  http://localhost:8080/api/sources/<id>