[minleejae]

CREATE falls back to UnsupportedStatement for forms it does not model, but only when no earlier alternative claims the statement. CREATE TABLE and CREATE INDEX accept any leading words as modifiers, so PostgreSQL statements such as CREATE LANGUAGE plperl, CREATE OPERATOR CLASS …, CREATE DEFAULT CONVERSION … or CREATE TRANSFORM … entered one of them and failed before reaching the fallback. With parseStatements, one such statement made the whole script fail. Forms whose object keyword is a reserved token, such as CREATE AGGREGATE or CREATE CAST, already fell back.

Create() now takes the fallback first when the statement contains none of TABLE, INDEX, VIEW and POLICY, which every remaining alternative needs. Statements that parse today keep their result. The fallback also keeps OR REPLACE, which Create() reads before choosing an alternative; CREATE OR REPLACE AGGREGATE … used to render without it.

Statements that failed this way now return UnsupportedStatement instead of throwing, as other unmodelled CREATE forms do. One existing test changes for this reason: CREATE DEFINER = x EVENT …, whose DEFINER the EVENT rule does not read yet, now falls back instead of failing. Public method signatures are retained.

Syntax reference: PostgreSQL 18 SQL commands.

Validation:

• UnsupportedStatementTest adds 7 PostgreSQL CREATE forms, parsed with and without the PostgreSQL dialect, and a script in which CREATE LANGUAGE sits between CREATE TABLE and CREATE INDEX. All 8 fail on master 5d07ba424.
• Java 17 Gradle check: 12,225 tests, 0 failures, 25 skipped; grammar ambiguity, formatting (Spotless ratchet against upstream/master), Checkstyle, PMD, SpotBugs and JaCoCo pass. JaCoCo instrumentation headroom of CCJSqlParser: 967 entries on master, 932 with this change.

AI assistance: prepared with Claude Code.

[manticore-projects]

Good one!!! Especially those DDLs come in all kind a variations vendor lock-ins, so having strong fallback is most welcome here.