LWG-3090 What is [time.duration.cons]/4's "no overflow is induced in the conversion" intended to mean?

[StephanTLavavej]

LWG-3090 What is [time.duration.cons]/4's "no overflow is induced in the conversion" intended to mean?

[cpplearner]

I suspect that it already works because of the use of _Ratio_divide_sfinae. (Apparently the person who implemented _Ratio_divide_sfinae was aware of LWG-2094 and thus understood the intent of [time.duration.cons]/4.)

STL/stl/inc/__msvc_chrono.hpp

Lines 117 to 121 in ab6dec6

	template <class _Rep2, class _Period2,
	enable_if_t<treat_as_floating_point_v<_Rep>
	|| (_Ratio_divide_sfinae<_Period2, _Period>::den == 1 && !treat_as_floating_point_v<_Rep2>),
	int> = 0>
	constexpr duration(const duration<_Rep2, _Period2>& _Dur)

[cpplearner]

But we also need to check "is_convertible_v<const Rep2&, rep> is true".

[StephanTLavavej]

Apparently the person who implemented _Ratio_divide_sfinae was aware of LWG-2094 and thus understood the intent of [time.duration.cons]/4.

That was me, in 2015-05-18's "constexpr-ize the STL, implementing 9 features, 10 issues, and 6 bugfixes" where I indeed explicitly implemented LWG-2094. 😺

[frederick-vs-ja]

But we also need to check "is_convertible_v<const Rep2&, rep> is true".

This part seems hard to test since requirements for non-arithmetic rep types are underspecified (#1909)...

[StephanTLavavej]

I expect there to be virtually zero usage of non-arithmetic representation types so I mostly don't care about that.