Exclude SVM test generators from test build

[trylek]

As the two test generators for TypeHierarchy and GenericContext
reside within the test source tree, we need to explicitly exclude
them from building as part of the test projects, otherwise test
project build fails due to asset path mismatches. An alternative
option would be to put the two test generator projects under
src\tests\Common where other test-related utilites reside and which
is globally excluded from test build.

Thanks

Tomas

P.S. I plan to create a no-merge PR against runtime main to get
initial CI results for the SVM test cases; locally I see just
a handful of failures in the negative tests.

[trylek]

I have added the runtime check that SVMs must be abstract; that fixes the MethodFlags negative test. For the negative VarianceSafety tests, I'm not fully clear on what exactly is supposed to fail in the tests. The code comment

runtime/src/tests/Loader/classloader/StaticVirtualMethods/NegativeTestCases/VarianceSafety.il

Line 12 in cc03608

// Default implementation not allowed

seems to indicate that the TypeLoadException is supposed to be thrown during construction of InterfaceScenario1/2 due to the presence of a SVM with default implementation but I see no such implementation in the code. I also don't fully understand to what extent this actually tests variance in SVM context considering there's no actual deriving happening in the test.

For the tests UnimplementedAbstractMethodOnConcreteClass, I'm worried that this is exactly an example of a "costly upfront check" that could be detrimental for the performance of initialization of primitive types and subsequently for startup performance in general. I'm interested in your thoughts on how to address this.

[trylek]

I have triggered a PR run in main for a no-merge RI including this change to validate the pass rates of the new tests on various OSes and architectures and to double-check that we haven't regressed pre-existing behaviors:

dotnet#52173

[trylek]

OK, the results are not terrible but it looks like we've broken something in default interface methods, I'm seeing the failures locally too. I'm going to focus on that next.

[trylek]

Ahh, OK, I was surprised that the MethodFlags negative tests continues failing even with my fix but now I see it's on Mono. We'll probably need to disable all the SVM tests on Mono for now until the counterpart runtime support gets merged in.

[davidwrighton]

The various tests beyond the autogenerated test were designed to cover code that hasn't been written yet, so it's not necessarily surprising that they don't pass.