ios_base constants should have bitmask types

[Igor Zhukov (fsb4000)]

Fixes #3401

If I understand correctly we can't change types of the variables but we can add operators.

[Alex Guteniev (AlexGuteniev)]

Have you considered _BITMASK_OPS macro?

[Igor Zhukov (fsb4000)]

Thanks. I knew about the macro but I forgot about it.

[A. Jiang (frederick-vs-ja)]

std::ios_base::openmode is still int. We should fix the type mismatch, see [ios.base.general].

[Alex Guteniev (AlexGuteniev)]

std::ios_base::openmode is still int. We should fix the type mismatch, see [ios.base.general].

Not sure if it wouldn't be an ABI break

[Igor Zhukov (fsb4000)]

Can we fix it?

What if a user used std::ios_base::openmode in his function. Before the type was int, and it is std::ios_base::_Openmode after.

It looks like ABI break.

I don't really understand all this about ABI break, but I remember that intmax_t was considered a very bad decision because this type is forever int64_t, because changing it to a hypothetical int128_t would be an ABI break.

[Stephan T. Lavavej (StephanTLavavej)]

The variables don't seem to appear on the DLL's export surface, so it might be possible to change their types. Yes, theoretically someone could have been depending on their types in their own signatures, but it would be fairly hard (especially pre-decltype). I am pretty scared to change anything in iostreams but it might be possible, and it would be simpler and better for throughput than emitting a bunch of bitmask ops (which are unfortunately widely visible).

[Stephan T. Lavavej (StephanTLavavej)]

Thanks, I pushed small changes to the test.

I double-checked for changes to the release and debug DLL export surfaces, and found none. I think this is ABI-safe from a "does this affect the redist" perspective, and I think that it's unlikely to be disruptive (as I mentioned earlier, theoretically someone could have used decltype to make their own signatures depend on the types of these constants, but that is totally different from normal usage).

There is now unused enum machinery but I would prefer not to mess with that at this time (as the TRANSITION, ABI comments indicate, I once thought this affected ABI - I can't remember why).

[Stephan T. Lavavej (StephanTLavavej)]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

[Stephan T. Lavavej (StephanTLavavej)]

Thanks for fixing this long-standing nonconformance! ✨ 😺 🎉