Problem

#1350 stops CREATE_TABLESPACE from being scheduled when tablespace testing is disabled, which is the default on every platform other than Linux. QPG reaches the generator through a different path, so with --qpg-enable true every worker still dies with a NullPointerException on Query.getLogString() in GlobalState.executePrologue.

Cause

ProviderAdapter.mutateTables() selects a mutation by reward and PostgresProvider.executeMutator() resolves it with Action.values()[index]. That path does not consult mapActions(), so the scheduling guard does not apply to it, PostgresTableSpaceGenerator.generate() returns null, and executeStatement() dereferences it.

CREATE_TABLESPACE is the only one of the 28 PostgreSQL actions whose generator can return null, so this is the only action affected.

Reproduction

On macOS or Windows:

java -jar target/sqlancer-2.0.0.jar --num-threads 1 --num-queries 20 postgres --oracle NOREC --qpg-enable true --qpg-max-interval 1

With the default interval of 1000 the same failure needs a long run, and CI does not see it because it runs on Linux.

Fix

Throw IgnoreMeException from the generator. mutateTables() already catches it, so the QPG framework needs no change. The scheduling guard from #1350 stays as it is and remains the reason the generator is never called during database generation.

Testing

mvn -Dtest=sqlancer.postgres.gen.TestPostgresTableSpaceGenerator test passes. Verified on Windows with the reproduction command above: QPG runs to completion.