[LucaCappelletti94]

A unique constraint or unique index declared NULLS NOT DISTINCT was refused, because a SQLite unique index always keeps NULLs apart and dropping the clause would let the replica accept rows PostgreSQL rejects. A schema could therefore not declare a constraint its backend supports.

The plain unique key is still emitted, so foreign keys and ON CONFLICT targets find it, and a companion unique index named after it with _nnd appended encodes each key that can be NULL as (key IS NULL, ifnull(key, 0)), keeping the column's collation and a partial index's predicate. NULL keys then collide as an ordinary unique violation, which INSERT OR IGNORE and ON CONFLICT DO NOTHING skip as PostgreSQL does. A trigger raising on an IS match would have aborted those and surfaced as a trigger error. DROP INDEX drops the companion as well, read from the schema as it stood before the drop, and an unnamed NULLS NOT DISTINCT index is refused since its companion would have no name to follow.

ON CONFLICT (key) DO UPDATE still aborts on a repeated NULL key where PostgreSQL updates the row, since SQLite resolves that target to the plain index. The README lists it under semantic differences.

[codecov]

Codecov Report

❌ Patch coverage is 98.00000% with 5 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/impls/translator_impls/nulls_not_distinct.rs	98.19%	3 Missing ⚠️
src/impls/translator_impls/create_index.rs	95.45%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!