The way how ClassMetadata goes to cache seems to be inconsistent. The problem is affected in #12302

When some code adds to ClassMetadata some constraints to ParentClass and then SubClass's ClassMetadata is built the constraints will appear in that metadata and will have a chance to be cached. Then they could appear in SubClass's ClassMetadata in other area of the code where ParentClass's ClassMetadata doesn't have the additional runtime constraints.

And vice versa - ParentClass's ClassMetadata runtime constraints won't affect SubClass's ClassMetadata if it is cached already. Another case is when constraints added to ParentClass after SubClass is generated won't affect its metadata.

IMO this leads to cases that are hard to debug and hard to bypass. Could it be counted as issue? Are the acceptable ways to resolve the cases?