On Thu, 15 Feb 2024 at 11:14, Jonathan Wakely <cxx_at_[hidden]> wrote:

>
>
> On Thu, 15 Feb 2024 at 11:13, Jonathan Wakely <cxx_at_[hidden]> wrote:
>
>>
>>
>> On Thu, 15 Feb 2024 at 11:12, Marcin Jaczewski <
>> marcinjaczewski86_at_[hidden]> wrote:
>>
>>> czw., 15 lut 2024 o 11:23 Jonathan Wakely via Std-Proposals
>>> <std-proposals_at_[hidden]> napisał(a):
>>> >
>>> >
>>> >
>>> > On Thu, 15 Feb 2024 at 09:54, Tiago Freire via Std-Proposals <
>>> std-proposals_at_[hidden]> wrote:
>>> >>
>>> >> And what exactly would this do?
>>> >
>>> >
>>> > Exactly what the standard does for subrange::size_ in
>>> [range.subrange.general]
>>> >
>>> >
>>> > make-unsigned-like-t<iter_difference_t<I>> size_ = 0; // exposition
>>> only; present only
>>> > // if StoreSize
>>> is true
>>> >
>>> >
>>> >
>>> >
>>> >>
>>> >> -----Original Message-----
>>> >> Subject: [std-proposals] Requires-clause for data members
>>> >>
>>> >> Has there been any discussion of requires-clauses for data members?
>>> >> Something like:
>>> >>
>>> >> > template<int N>
>>> >> > struct numbers {
>>> >> > int x requires (N >= 1) = 0; // leading
>>> >> > int y requires (N >= 2) = 0; // trailing };
>>> >>
>>> >> It seems grammatically possible if you mandate parentheses, and it
>>> would certainly help people reduce the use of partial specializations in
>>> these places.
>>> >
>>> >
>>> >
>>> https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2013/n3613.pdf#section.5
>>> has some discussion of an earlier idea like this, based on features in the
>>> D language.
>>> >
>>>
>>> Then what will be the behavior of:
>>>
>>> ```
>>> size_ = 13;
>>> ```
>>>
>>> If we consider this as "zombi" name, we could issue a warning like:
>>> "variable is deleted".
>>> But it will work fine in code like:
>>>
>>> ```
>>> void setSize(int i) require (StoreSize == true)
>>> {
>>> size_ = i;
>>> }
>>> void setSize(int) require (StoreSize == false)
>>> {
>>> // nothing
>>> }
>>> ```
>>>
>>> This will still be "viral" but at least we could easily reason about
>>> code as `size_` always means the same thing but
>>> in some cases is not accessible.
>>>
>>
>> Yes, as discussed in
>> https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2013/n3613.pdf#section.5
>> which is why I linked to it :-)
>>
>
> But it's much simpler to just do:
>
> if constexpr (StoreSize)
> size_ = n;
>
> in the middle of another function, rather than adding a pair of overloaded
> setters.
>

And I don't think a warning is appropriate. Using a non-existent member
should be ill-formed, just as it is today if you implement a
"not-always-present" member via indirection through a subobject of class
template type where some specializations don't have the member.