[#857] Fix some regressions in JPQLNext parser

[Mobe91]

Fixes #857

[beikov]

Looks good, but you missed a few cases

• com.blazebit.persistence.view.impl.objectbuilder.mapper.AbstractCorrelationJoinTupleElementMapper#AbstractCorrelationJoinTupleElementMapper
• com.blazebit.persistence.view.impl.PrefixingQueryGenerator#prefix
• com.blazebit.persistence.view.impl.objectbuilder.ViewTypeObjectBuilderTemplate#getCorrelationBasisType

Also, please add a test for this

[Mobe91]

Not really, the cases that you mention did not change compared to before your parser rework commit so I did not touch them.
I would need to disable expression caching to create a test case that triggers this behavior - ok?

[beikov]

True but the cases I mentioned should also allow object results.
You shouldn't need to disable a cache to reproduce this. If you do, that would be a bug.

[Mobe91]

True but the cases I mentioned should also allow object results.

Ok, I will apply the changes then.

You shouldn't need a cache to reproduce this. If you do, that would be a bug.

Expression caching itself is a bug at the moment. The cache keys need to consider the various flags that influence behavior of parsing and the expression visitor which is currently not the case. If it was implemented that way, those regressions (and possibly the additional cases you mentioned) would have come up with the existing tests I think.

I can fix the expression caching in the course of this issue.

[beikov]

Please create constants for the bit masks

[beikov]

Checkstyle issues: https://travis-ci.org/Blazebit/blaze-persistence/jobs/592945474#L953