Note 1: this was added in 4.2.16.

To genereate correct SQL we need to substitute comma ", " with "cross join". In other words - use org.hibernate.dialect.Dialect#getCrossJoinSeparator method while generating SQL.

Note 2: some dialects return just a comma (", ") from their getCrossJoinSeparator() implementations. E.g. DB2Dialect, TimesTenDialect. This is not a problem for me (I'm not using any of them), but this probably means this regression is still present on certain platforms.