Migration 0039: harden non-comic cleanup, reorder ACL index, drop dead FTS dedupe

[ajslater]

Summary

Three review-driven cleanups to migration 0039_metron_age_rating_default_view_and_performance.py — no functional changes outside the migration, just safety and efficiency improvements to make it harder to fail or misbehave on real installs.

• Hardened _remove_non_comic_comics with two independent defenses against an unmounted comics volume turning the cleanup into a mass-delete:

  1. Sentinel stat probe — a new _comics_fs_reachable() helper probes up to 10 known comic-suffix paths via Path.is_file(). If none stat as files, the cleanup logs a warning and skips entirely.
  2. page_count == 0 predicate — the deletable queryset is now restricted to phantom-row signature only (the original importer bug always left page_count = 0). A real comic with an unusual filename has at least one page and is never a candidate.

• Reordered the composite ACL index codex_comic_lib_ari_idx to be built after _backfill_age_rating_metron_index so SQLite does one sorted index build instead of incrementally maintaining the index during the bulk UPDATE on codex_comic.

• Removed the FTS dedupe RunSQL (and the _FTS_DEDUPE_SQL constant). It ran immediately before DROP TABLE codex_comicfts, so it was dead code — pure overhead and one more statement that could fail mid-migration.

• Swapped print() for loguru.logger in the cleanup function — loguru is already imported at the top of the file and works fine at migration time.

Why

A review of the 0039 migration flagged three concerns: the FTS dedupe was unreachable code, the ACL index was being maintained row-by-row during a bulk UPDATE instead of built once at the end, and _remove_non_comic_comics relied entirely on Path.is_file() to protect legitimate comics — meaning a misconfigured volume mount could escalate into mass deletion. Page-count + sentinel together make it virtually impossible for the cleanup to delete a real comic regardless of FS state.

Test plan

• uv run python bin/manage.py migrate applies cleanly through 0039 on a fresh DB; sentinel correctly logs "no comic-suffix paths" and bails out instead of deleting.
• ruff check passes on the migration file.
• pytest tests/ — 26 passed.
• Apply on a v1.10-era DB with phantom rows present and confirm the cleanup deletes them and logs through loguru.
• Apply with the comics volume deliberately unmounted and confirm the migration completes without deleting anything.

🤖 Generated with Claude Code