Currently the RI follows a "fail late" approach for handling illegal method constraints. As this is not defined in the spec, the TCK shouldn't rely on this behavior, though (see BVAL-427).

Therefore I experimentally changed to the RI to raise any exceptions due to illegal method constraints in an eager manner in order to see whether any TCK tests would fail. As it turned out, no changes in the TCK tests were required, i.e. tests for legal and illegal configurations use separate test model types. Some RI tests relied on the behavior, though.

Nevertheless I think we could bring this change forward, as Hardy Ferentschik seemed to like a fail-early approach. I'm also starting to like it because it makes the implementation a bit easier, since we don't have to keep the lazily thrown exception around.