Tests for P1831 and workaround for DevCom-10948908

[YexuanXiao]

Even with /W3, MSVC doesn't respect the [[deprecated]] attribute of template specializations (DevCom-10948908).
Therefore, the macro _CXX20_DEPRECATE_VOLATILE has no effect with tuple_size, tuple_elements, variant_size and variant_alternatives.

Due to another bug of MSVC, applying the [[deprecated]] attribute directly to type still has no effect, so it can only be applied to _Mybase. Since my attempt to reproduce it failed, I cannot report it as another bug. When DevCom-10948908 is fixed, this bug may still persist, but all two workaround can be removed.

[YexuanXiao]

This issue also exists in Clang19 and Clang20, but it will be fixed in Clang21. See commit.

[YexuanXiao]

@StephanTLavavej What is the difference between checks numbered 3, 4, and others in Build and Test x64 x64? They unexpectedly passed some new tests.

[MattStephanson]

What is the difference between checks numbered 3, 4, and others in Build and Test x64 x64? They unexpectedly passed some new tests.

#1414 originally introduced this feature, but #2794 concisely explains it:

We use LLVM lit's "sharding" feature to split our large test suite across 8 VMs per architecture

If you go to the pipeline run webpage and click on the results tab, you'll see a list of the failed tests.

[YexuanXiao]

What is the difference between checks numbered 3, 4, and others in Build and Test x64 x64? They unexpectedly passed some new tests.

#1414 originally introduced this feature, but #2794 concisely explains it:

We use LLVM lit's "sharding" feature to split our large test suite across 8 VMs per architecture

If you go to the pipeline run webpage and click on the results tab, you'll see a list of the failed tests.

Based on your explanation, I believe I have found a clue to reproduce the issue. Thanks!

[YexuanXiao]

The workaround for tuple_element and variant_alternative does not work for Clang 19, even when applying the [[deprecated]] attribute directly to type instead of _Mybase. It's as tricky to simplify to reproduce as the MSVC bug. I can't figure out how to implement a workaround for it.

[YexuanXiao]

The MSVC team believes a CWG issue is needed, so it may not be fixed in the short term. How should STL adapt to this situation? @StephanTLavavej

[StephanTLavavej]

@YexuanXiao Thanks for looking into this. Because MSVC's behavior is unlikely to change in the moderate future and the impact of missing deprecation warnings for volatile machinery is highly limited, I believe I would prefer to abandon this attempt. I appreciate the effort you've put into it, but at this point I don't think making changes here is worth the complexity, risk, and time.