[compiler-segfault]

SELECT -1 ^ 0 is parsed as -(1 ^ 0) by PostgreSqlDialect, so it yields -1 where PostgreSQL yields 1; SELECT +2 ^ 2 yields -4 instead of 4.

PostgreSQL 17:  SELECT -1 ^ 0;  ->  1
sqlparser:      UnaryOp { op: Minus, expr: BinaryOp { .., op: PGExp, .. } }

In gram.y UMINUS is declared above COLLATE, AT and ^, so the operand of a unary +/- may only take ., :: and []. The operand precedence is now a Dialect method that defaults to the previous MulDivModOp value, and PostgreSqlDialect returns the COLLATE level. That also makes -a AT TIME ZONE 'tz' parse as (-a) AT TIME ZONE 'tz', which is the same reading gram.y gives.

PostgreSQL precedence table

parse_unary_operand_binds_tighter_than_exponentiation is added to tests/sqlparser_postgres.rs; without the change it fails with the AST quoted above.

cargo test --all-features, cargo fmt --all and cargo clippy --all-targets --all-features -- -D warnings all pass.

Prepared with AI assistance; I reviewed the diff and ran the checks above.

Fixes #2819.