Merge the Alternate Number sort into the Alternate Series sort

[ajslater]

What

The two half-working sorts for alternate series are now one key: Alternate Series (reprints). Comics group by their alternate series identity (name, volume, language) and order by the parsed AlternateNumber inside the group — #2 before #10 — falling back to the comic's own series and issue so untagged comics interleave instead of clumping. alternate_number is removed from the order-by choices with a full stored-settings remap. Matches the reader, which already reads alternate series in parsed-number order.

Why

Previously each sort did half the job: Alternate Series ordered by the display label, so "#10" sorted before "#2" and fallback comics had no issue tail; Alternate Number sorted numerically but only resolved which reprint via the reprints filter, silently degraded to the plain issue sort without it, and never considered the series name. The merge also fixes two latent bugs: the label sort compared case-sensitively (SQLite collation doesn't survive Coalesce/Concat/Min), and the filtered number sort aggregated number and suffix independently, so a comic with reprints "#2" and "#10a" sorted by a fabricated "#2a".

How

• Comic rows (columns.py): a correlated subquery elects one reprint per comic by natural tuple order; ORDER BY is a five-field Coalesce list reading from that election. All five subqueries share one election ordering, so tied rows carry identical values — the tuple stays atomic without a composed string key. An active reprints filter narrows the election (pk__in); no more FilteredRelation. Benchmarked at parity with the join+aggregate shape (164 vs 167 ms sorting 120k comics; the forced GROUP BY pk keeps the correlated subqueries out of GROUP BY). The fallback series segment is lower(name) rather than the article-moved sort_name, so an untagged "The Batman" interleaves with its "The Batman" alternate namesake.
• Collection rows (intersections.py): the intersection RawSQL emits a matching fixed-width key with an in-SQL fallback to the collection's own name (a Django Coalesce wrapper would re-trigger the quadratic Folder GROUP BY regression). The key reads only reprint columns — a per-comic fallback under GROUP BY target_id returned an arbitrary child's issue — and its printf formats are %%-escaped so the DEBUG query logger (bin/dev.sh default) doesn't consume bound params. Both fixes have regression tests.
• Settings: 0053's choices tuple edited in place (unreleased); new pure-RunPython 0054 remaps alternate_number → reprints in order_by, order_extra_keys, and collection_order_memory with dedupe (forward-only; the merge is lossy). Sidecar restore gets a separate sort-key rename map covering the same three homes — kept apart from _LEGACY_KEY_RENAMES, which also drives filter-column resolution.
• Frontend: the alternate_number caption branch is gone; formatReprints and the order_value display contract are unchanged.

Reviewer notes

• Accepted behavior changes (called out in NEWS): saved "Alternate Number" sorts become "Alternate Series" with series-grouped rather than purely numeric ordering; a collection's cover pick no longer follows the alternate numbering (the merged key normalizes to sort_name in cover subqueries like every other M2M sort); the reprints filter no longer influences collection-row ordering.
• reprints stays a legal multi-sort extra; Comic.alternative_issue_number is an unrelated display-only field and is untouched.
• Tests: 23 in test_browser_reprints_column.py including cross-series election atomicity, case folding, article interleave, and a DEBUG-cursor regression; 0054 remap and restore-rename suites. The subtle guarantees were mutation-tested (each fails a dedicated test when broken). Full suite 1032 passed; lint/ty/complexity clean.

🤖 Generated with Claude Code