35 lines
1.6 KiB
SQL
35 lines
1.6 KiB
SQL
|
-- +goose Up
|
||
|
|
-- +goose StatementBegin
|
||
|
|
-- Truncate audiobook_series so the table can be repopulated cleanly by
|
||
|
|
-- the next library scan. The data shipped by migration 145's regex
|
||
|
|
-- backfill (and subsequently overwritten by the scanner) was 87%
|
||
|
|
-- singletons: a separate analysis traced the singleton flood to
|
||
|
|
-- internal/scanner/audiobook.go's parser, which until commit <fixed
|
||
|
|
-- alongside this migration> used `tags["album"]` as a Series fallback.
|
||
|
|
-- In audiobook tagging conventions `album` holds the book title, so
|
||
|
|
-- every book without an explicit `series` or `mvnm` tag produced
|
||
|
|
-- series_name = title, populating the table with one unique "series"
|
||
|
|
-- per book.
|
||
|
|
--
|
||
|
|
-- With the parser fixed (Series no longer falls back to album), this
|
||
|
|
-- TRUNCATE clears the polluted state so the next scan rebuilds the
|
||
|
|
-- table from real `series` / `mvnm` tag data only. Books without those
|
||
|
|
-- tags simply don't get an audiobook_series row, which is the correct
|
||
|
|
-- end state.
|
||
|
|
--
|
||
|
|
-- The Series filter dropdown will be empty between deploy and the
|
||
|
|
-- next library scan completing; acceptable because 87% of the prior
|
||
|
|
-- entries were fake anyway.
|
||
|
|
TRUNCATE TABLE audiobook_series;
|
||
|
|
-- +goose StatementEnd
|
||
|
|
|
||
|
|
-- +goose Down
|
||
|
|
-- +goose StatementBegin
|
||
|
|
-- TRUNCATE cannot be undone — the table contents were discarded
|
||
|
|
-- intentionally because they were 87% noise from a parser bug. Down
|
||
|
|
-- migration is a no-op; the next library scan after rolling back the
|
||
|
|
-- parser fix would repopulate the table with the same polluted data,
|
||
|
|
-- which is not a desirable rollback target.
|
||
|
|
SELECT 1 WHERE FALSE; -- no-op
|
||
|
|
-- +goose StatementEnd
|