ClickHouse: map PostgreSQL regex-match operators (~, ~*, !~, !~*) to match()

[alexey-milovidov]

What

Maps PostgreSQL's regex-match operators to ClickHouse's match() when transpiling to the ClickHouse dialect:

PostgreSQL	ClickHouse
x ~ 'p'	match(x, 'p')
x ~* 'p'	match(x, '(?i)p')
x !~ 'p'	NOT (match(x, 'p'))
x !~* 'p'	NOT (match(x, '(?i)p'))

The mapping lives on the RegexpLike/RegexpILike AST nodes, so MySQL RLIKE/REGEXP sources also transpile to match().

Why

ClickHouse has no REGEXP_LIKE function and no case-insensitive regex operator. Previously these operators were emitted as REGEXP_LIKE(...) (for ~/!~) or an invalid infix x REGEXP_ILIKE y (for ~*/!~*), both of which ClickHouse rejects:

Function with name `REGEXP_LIKE` does not exist

This surfaced while transpiling PostgreSQL's regression test suite to ClickHouse; the ~-family operators are common in real queries (e.g. WHERE name ~ '^A').

How

Adds a ClickHouse branch to generate_regexp_like and generate_regexp_i_like. Case-insensitivity uses an inline (?i) flag — inlined into a string-literal pattern for readable output, and wrapped as concat('(?i)', pattern) for a non-literal pattern so the flag still applies. !~/!~* are already represented as NOT(RegexpLike/RegexpILike), so they follow automatically.

Scope: only the unambiguous operators

Deliberately excluded are PostgreSQL's polymorphic operators — @>, <@, &&, @@, >>, <<. Their meaning is type-dependent (array vs jsonb vs range vs inet vs bit-string containment / overlap / shift), which a purely syntactic transpiler cannot resolve without operand type information. Mapping @> to hasAll would be right for arrays but wrong for jsonb/ranges, and PostgreSQL's >>/<< are bit-string/inet operators (not integer shifts), so bitShiftRight would be incorrect. These continue to pass through unchanged — surfacing as a ClickHouse parse error rather than silently-wrong results. The regex-match operators are unambiguous (text-only), which is why they are safe to map here.

Tests

Adds crates/polyglot-sql/tests/postgres_regexp_clickhouse.rs. Output semantics were verified against a real ClickHouse instance (match('ABC','(?i)abc') = 1, match('ABC','abc') = 0, NOT (match('abc','x')) = 1). The full polyglot-sql suite passes; the only failing test, tpch_query2_transpile_postgres_to_fabric, is pre-existing on main and unrelated to this change.

---

Upstream PR: tobilg#316