Make "targetEntity must not be a mapped superclass" a lazy check

[mpdude]

#10473 relaxed the runtime check that mapped superclasses must not use one-to-many associations. It turned the check the other way round, so that no mapped superclass may be used as the targetEntity of an association.

To find out whether a class is a mapped superclass, the \Doctrine\ORM\Mapping\ClassMetadataFactory::peekIfIsMappedSuperclass method was used, which is directly using the mapping driver.

#10552 reports a case where mapping configuration is not collected through the driver alone, but also a metadata listener is contributing relevant information. Updating metadata with a listener is currently not supported by the check in that method.

Also, #10473 (comment) has a mapping configuration that cannot be loaded in the improvised way that \Doctrine\ORM\Mapping\ClassMetadataFactory::peekIfIsMappedSuperclass employs.

As I explained in #10552 (comment), we cannot use the full ClassMetadataFactory itself to obtain information about the targetEntity while we're performing runtime validation for another class currently being loaded.

So, I don't see any other way than to make this check a non-runtime (offline?) check in the SchemaValidator. There, we can first completely load metadata for all classes and then inspect it.

Many fixes to the test models that were made in #10473 can be reverted, since these files are no longer scrutinized at runtime, and we probably don't care about the extra validation checks that much.

[alexander-schranz]

I did not have a look at the code as I'm not unfamiliar with internals here but this PR fixes our problems inside @sulu: sulu/sulu#7026

Thank you @mpdude 👍

[BreyndotEchse]

MappingException gone and all tests passed!
ty @mpdude

[mpdude]

Doctrine core team reviewers:

Please confirm that using a SchemaValidator based check only instead of a runtime check is a valid approach. That's what the schema validator is made for, right? Not all problems can easily be spotted at runtime.

[mpdude]

@derrabus, @greg0ire or @SenseException – can we try to get this merged in the near future? I have the impression it would unblock people who are kind enough to try 2.15-dev in their CI pipelines.

[greg0ire]

Thanks @mpdude !