Disable signature transformation for NTSTATUS

[kennykerr]

Fixes: #2656

[riverar]

I don't think there are enough STATUS_PENDING returning functions to warrant degrading the developer experience here. Is there something preventing us from specializing the return transformation for NTSTATUS?

[kennykerr]

I think the issue is that there are many many more success codes when it comes to NTSTATUS compared with say HRESULT.

The alternative would be to return Result<NTSTATUS, Error> but that would be very different to the way we handle error transformations today and just seems like it would cause more confusion than anything else.

[ChrisDenton]

The status quo relies on CanReturnMultipleSuccessValues being accurately applied (or not) to all NTSTATUS returning functions. Otherwise the experience is not just degraded, it's unusable for certain scenarios. But I think we simply don't have the people to ensure that the metadata 100% accurately applies the attribute. Maybe that'll come in time but we're not there yet.

So, all else aside, I think something has to change.

[riverar]

Agree we'd just be applying CanReturnMultipleSuccessValues to functions as we find them, but in the absence of more data demonstrating a large swath of functions that will be better off, we may want to consider it over regressing the experience across the board. I recognize, though, getting that data is probably not possible in the short term and concede as a result.

[ChrisDenton]

I guess there are alternatives but they'll require more work to get started on. So maybe the opposite attribute for NTSTATUS would be more useful (e.g. CanReturnSingleSuccessValues)? Have the least destructive behaviour by default and only apply Result ergonomics for functions that have been reviewed as only returning one success value.

[kennykerr]

I just don't think any of these attribute-based approaches are scalable.

[kennykerr]

Unrelated, the build is failing on the diff check for libs. That's probably because GitHub updated their MSVC toolchain and now the libs have a different version stamp again. I really need to get that version stamp wiped out.

[riverar]

Unrelated, the build is failing on the diff check for libs. That's probably because GitHub updated their MSVC toolchain and now the libs have a different version stamp again. I really need to get that version stamp wiped out.

Yup, will pick this up (removing the compiler ID per last discussion).

[kennykerr]

Manually updated the libs to unblock this PR.