Require complete types in some missing places

[llvm-beanz]

This change adds additional calls to Sema::RequireCompleteType, one when evaluating unary sizeof, which fixes a reported issue with templates.

The second area is in the DXR diagnostics where complete type should be required but not diagnosed because ordering in the compiler is a bit wonky when diagnosing DXR entry points.

The third is when checking the template argument for Load/Store, where again an incomplete type will be diagnosed later, but needs to be handled at this point.

Fixes #7510

[tex3d]

Thanks Chris!

[llvm-beanz]

@tex3d's additions look good to me. I can't approve since I'm the original author.

[pow2clk]

It looks sensible to me. The overall question I have is how do we determine where this is needed? Is it anywhere we check a type in Sema? Is there not somewhere we can insert this once and for all? I added a lot of these as part of the longvector diagnostics

[hekota]

LGTM!

[tex3d]

It looks sensible to me. The overall question I have is how do we determine where this is needed? Is it anywhere we check a type in Sema? Is there not somewhere we can insert this once and for all? I added a lot of these as part of the longvector diagnostics

Yeah, it's a bit more complicated than that. While it would seem convenient to call RequireCompleteType from IsHLSLCopyableAnnotatableRecord, the code that's calling this may need to handle an incomplete type scenario differently than a failing IsHLSLCopyableAnnotatableRecord call. Plus, some code paths should be able to assume the type is completed already, otherwise something else is wrong. For instance, there's a code path that uses IsHLSLCopyableAnnotatableRecord for a call argument type (payload) during DiagnoseTranslationUnit, where there should definitely not be an incomplete type, and if there was, this is not the right time to complete it, so it will hit the assert instead of blindly trying to complete the type now.