…the context of 'or' scopes where the property wasn't actually required. Technically we could architect this in a way that allowed a slightly higher percentage of INNER JOINs to be used (i.e. if the chain of associations from the parent domain to the current association are all non-nullable, then it doesn't matter that they were used in an 'or' scope--they could still be joined using an INNER JOIN.) However, due to the the complexity of that (and especially the need for considering how that might interact with sub-query embedded scopes) for right now, I'll push that off into the future in favor of fixing this bug more immediately.
Relscope enforces that queries won't join across has-many associations anyway (opting to use subqueries) so there will never be a case where the id's counted shouldn't be distinct. This change allows better DB performance since the database can't really know what we know about never joining across has-many's and the uniqueness of ids.
Otherwise the generated SQL is accepted by MySQL (since it just ignores the incorrect part) but PostgreSQL blows up (correctly so) since the count(id) really forces the need for group by if we order. And order shouldn't affect the count anyway, so why do it.
…ed when loading the root objects.
…ed on non-nullable/nullable.)
… subquery so that later requests for the current alias at a lower association level work correctly.
… essentially the code that would be used to create that scope.)
…e walking is nullable: false.
…e change to the 'options' map flagging it as a subquery happens before the detached criteria is processed with its projections.
…w property checks with the associationDescriptorStack and non-association based subqueries.
…e same association property caused duplicate aliases to be created.
… end of a scenario to guarantee that all instances get deleted.
…ubquery. Also fix bug that caused the new property verification tests to fail when extending into an association within an exists subquery.
… report hibernate types of value expressions. Also add some missing null-safety around lhs.
…n since for an ancestor walking expression, the association descriptor stack in the options map is only valid during the RelScope#toCriterion() tree walk.
…ion returns null if no results are found. Fixes #4.
…elScope.propertyFor() as well. Fix passing the wrong variable in the association walking loop in RelScope.propertyFor().
…he correct parent domain class when entering association sub-scopes.
…ove tuple length check to the ScopeComparisonFactory since it is used by more than one method in the between comparator class (also, we want it to fail as early as possible.)