The ClassMetadataInfo#inverseMappings attribute is an unnecessary complication.
The problem comes down to: "Given an association $assoc, how do I know if the association is bidirectional?".
If $assoc is the inverse side ($assoc->isInverseSide()), its clear, its bidirectional. But if $assoc is the owning side, we dont know.
What we currently do to determine that in such a case looks as follows:
So we check whether the target class has an inverse mapping to the source class. This is quite cumbersome. Additionally, we're usually only interested in the sourceFieldName of the inverse association.
Therefore, it would simplify internal code a lot if we simply introduced a "inversedBy" attribute on an association that is used on the owning side and that corresponds to "mappedBy" on the inverse side. Example:
This breaks BC. However, I think this is really necessary as it simplifies and speeds up the code, together with removing the inverseMappings attribute altogether.
So this should happen before beta.