The necessary magic to use Plex’s SQLite is here:
I believe this is associated with improvements in how full-text searching is implemented.
There’s a new index, index_title_sort_icu. And two new virtual full-text-search tables, fts4_metadata_titles_icu and fts4_tag_titles_icu - that’s where the collating add-on tokenizer is being used.
Things I haven’t tested:
I wonder if dropping index_title_sort_icu and the fts4*icu virtual tables and associated triggers would work, if done correctly. Similar to how the Repair a Corrupt Database instructions have you drop index_title_sort_naturalsort.
There’s a big group of schema migrations associated with these. They all start with 500000000000 so it might be possible to do.
