ranges::distance fails to emulate the Standard's two overload presentation

[CaseyCarter]

By not properly constraining itself with the union of the constraints of those two overloads. If we weren't trying to do something fancy just to avoid a bit of non-subsumption concept overloading, we wouldn't have written this bug in the first place. Let's split out into two overloads.

Test coverage is incoming in the LLVM update, against which I've verified the fix.

Merge conflict with #3024

[strega-nil-ms]

Is there a reason not to just do the two-overload version, given that the two-overload version is how the standard is written?

[CaseyCarter]

Is there a reason not to just do the two-overload version, given that the two-overload version is how the standard is written?

There are exactly two reasons why an STL implementation would take advantage of as-if to do something differently from how it's phrased in the Standard: performance and throughput. This is the latter.

We generally prefer if constexpr dispatch to concept overloading because it's quicker, even in cases like this where subsumption isn't a factor (normalization and subsumption checking are the really expensive bits). That said, I could be convinced that this bug wouldn't have happened if I hadn't tried to be fancy, and that I should fix the wording in LWG-3664 if I want to do something different.

[strega-nil-ms]

I would agree with the latter point. If they took the same parameters, it would be different IMO, but doing this weird decay_t stuff to allow taking either a const I& OR an I by value is a lot.

[CaseyCarter]

I would agree with the latter point. If they took the same parameters, it would be different IMO, but doing this weird decay_t stuff to allow taking either a const I& OR an I by value is a lot.

Split into two overloads, reverified against the incoming libc++ test, and rewrote the commit message.

[StephanTLavavej]

Looks good to me. FYI @strega-nil-ms, I pushed a couple of comment changes after you approved.

[CaseyCarter]

Thanks for finishing this up, @strega-nil-ms!

[StephanTLavavej]

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

[StephanTLavavej]

Thanks for going the distance here! 🏃 🥇 😹