v3.10.1

Highlights

Torznab and /dmm/filtered searches with a trailing year in q= (e.g. q=Oppenheimer 2023 or q=Oppenheimer (2023)) now return results. Previously they returned zero because the trigram match was running against the cleaned title (which never contains a year) and failing the similarity threshold.

This was an inherited upstream issue dating to iPromKnight/zilean@5149cd1 (Nov 2024).

The trailing year is now extracted from q= and routed into the structured Year filter automatically. Explicit &year= still wins when both are provided. Bare-year queries like q=1923 (the show) and mid-string years like q=The Matrix 1999 Remastered are unaffected.

Closes #22.

Bug Fixes

• extract trailing year from q= in filtered search (6c8e0d1)

v3.10.0

3.10.0 (2026-05-10)

Features

• gate per-torrent log spam, add per-batch stage timings (#21) (93c9b63)

Highlights

• Parsing.VerboseLogging (default false): gates per-title Python loguru lines and demotes C# matcher per-torrent logs (TorrentUpdated / TorrentRetained / NoSuitableMatchFound) to LogLevel.Debug. Operators wanting full per-torrent visibility set the flag to true (Python side) and drop Serilog MinimumLevel to Debug (C# side).
• Per-batch stage timings: Batch N stage timings: parse=Xms populate=Yms match=Zms upsert=Wms emitted at Information. Spot bottlenecks per stage without --trace.
• asyncio.to_thread reverted: prototyped + measured at 5.74x regression on synthetic 5K parse (RTN's parse() holds the GIL through Python re, dispatch overhead dominates). True multicore parser parallelism parked for a future PR.

Addresses parser-side observability + log gating from #19; true multicore is separate.

v3.9.4

3.9.4 (2026-05-09)

Bug Fixes

• bound IMDb match cache and warn on stale snapshot during long ingests (#20) (50e01d6)

Highlights

• Imdb.MatchCacheSize (default 100000, ~25-50 MB): bounds the in-memory (title, year, category) to IMDb ID cache. Replaces an unbounded ConcurrentDictionary that grew across long ingests.
• Imdb.SnapshotMaxAgeHours (default 24, set 0 to disable): logs a one-shot warning recommending a restart so the in-memory snapshot is rebuilt from a refreshed ImdbFiles table.
• Synthetic Lucene-matcher throughput benchmark (~9.6x speedup on 2K-row test corpus). Tagged [Trait("Category", "Benchmark")] and excluded from default CI.

Addresses the IMDb-matcher half of #19.

Documentation

• add trigram search tip for short queries (#18) (bff9801)

v3.9.3

3.9.3 (2026-04-06)

Bug Fixes

• generic search (t=search) now respects category filter (#17) (6ab6406)

v3.9.2

3.9.2 (2026-04-06)

Bug Fixes

• add EF migration for book/audiobook search threshold (#15) (f285679), closes #13

Documentation

• update README with supported categories and streamline fork changes (#14) (c0c63af)

v3.9.1

3.9.1 (2026-04-06)

Bug Fixes

• lower trigram similarity threshold for book/audiobook searches (#13) (ad605b4)

v3.9.0

3.9.0 (2026-04-06)

Features

• v4.0.0 - book and audiobook category support (#12) (2e3d1a2)

v3.8.3

3.8.3 (2026-04-06)

Bug Fixes

• v3.9.0 reliability - remove abrupt termination, add Python graceful degradation (#11) (d778a77)

v3.8.2

Test Infrastructure

• Add dotnet test to CI pipeline (Python tests excluded via trait filter)
• Add 12 new tests: 6 unit (Torznab XML, query validation) + 6 integration (health checks, search endpoints)
• Add ZileanWebApplicationFactory, TestDataBuilder, and Testcontainers fixture
• Fix Testcontainers port binding (random ports instead of fixed 5432)
• Enable branch protection on main (require CI pass before merge)

No user-facing changes. Internal test infrastructure only.

v3.8.1

3.8.1 (2026-04-05)

Bug Fixes

• trailing comma, ElfHosted defaults, endpoint suffix config (4d8ded1)

Continuous Integration

• skip release/docker for docs-only changes (1319b0c)
• use commit changelog for release notes instead of PR-based (b8c6f2f)

Documentation

• fix changelog ordering (4c82a69)
• remove internal CI/CD note from changelog (4c9f14d)
• update changelog for v3.8.0 (8a1b6b4)