On Tue, 3 Jun 2025 at 23:24, Jonathan Wakely <cxx_at_[hidden]> wrote:

>
>
> On Tue, 3 Jun 2025 at 21:58, Howard Hinnant via Std-Proposals <
> std-proposals_at_[hidden]> wrote:
>
>> I view this as a bug in the spec. It was not intended to outlaw unsigned
>> durations for hh_mm_ss.
>>
>> Upon experimentation I found that my example implementation does not
>> outlaw it, gcc does not, msvc does, and llvm does.
>>
>> A revised wording might look something like: let abs_d have the value -d
>> if is_neg is true, else d. Then replace abs(d) with abs_d.
>>
>> This change in wording will not break gcc. And it will turn cases that
>> don’t compile on MSVC and LLVM into cases that do compile, so I believe the
>> “breakage” would be acceptable.
>>
>
> It's no accident that it works with GCC:
>
> commit r13-5002-ge36e57b032b2d70eaa1294d5921e4fd8ce12a74d <https://gcc.gnu.org/cgit/gcc/commit/?id=e36e57b032b2d70eaa1294d5921e4fd8ce12a74d>
> Author: Jonathan Wakely
> Date: Wed Jan 4 16:43:51 2023 +0000
>
> libstdc++: Fix std::chrono::hh_mm_ss with unsigned rep [PR108265 <https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108265>]
>
> libstdc++-v3/ChangeLog:
>
> PR libstdc++/108265 <https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108265>
> * include/std/chrono (hh_mm_ss): Do not use chrono::abs if
> duration rep is unsigned.
> * testsuite/std/time/hh_mm_ss/1.cc: Check unsigned rep.
>
>
> In response to https://gcc.gnu.org/PR108265 reported by one Michael
> Duggan :-)
>
>
I think this is a conforming extension (no valid program can detect the
difference) and I agree with Howard that it doesn't cause any real
breakage. The fact that chrono::hh_mm_ss uses chrono::abs internally is not
detectable using SFINAE or concepts, because
std::is_constructible_v<hh_mm_ss<seconds>, duration<unsigned>> is true,
even for implementations where it doesn't actually compile if you try to
construct that.

Changing the spec to make this work seems like a defect fix, it wouldn't
break anything.