[LucaCappelletti94]

PostgreSQL accepts an index written without a name, such as CREATE INDEX ON t (s), and names it itself. pg2sqlite copied the missing name into its output, so SQLite saw CREATE INDEX ON t(s) and refused it as a syntax error, and a later DROP INDEX t_s_idx written against the server's name failed before translation because the catalog held the index without one.

Each unnamed index now gets the name PostgreSQL 17 gives it, asked of sql-traits against the catalog as it stands just before the statement, so counters, clashes with other relations and names freed by an earlier DROP INDEX come out as on the server. The name is written into the statement before anything else reads it, which also lets an unnamed NULLS NOT DISTINCT index name its companion where it used to be refused. An index whose name sql-traits cannot predict is refused with a request to name it, since emitting it unnamed is what SQLite rejects. A gauntlet test compares the index names SQLite ends up with against pg_indexes on a live PostgreSQL.

This needs connetto-rs/sql-traits#146 for ParserDBIngestor::create_index_name, so CI here stays red until that merges.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!