C++ Logo


Advanced search

Subject: Re: [std-proposals] is_value_constructible
From: Arthur O'Dwyer (arthur.j.odwyer_at_[hidden])
Date: 2021-05-03 10:11:36

On Mon, May 3, 2021 at 10:04 AM Bjorn Reese <breese_at_[hidden]> wrote:

> On 2021-05-02 18:56, Arthur O'Dwyer wrote:
> > I emphasize "for a real problem" because the code you've showed above is
> > not real: `alpha` is an empty struct, and you didn't provide the body of
> > the "value constructor," so it's not clear to the reader what it's
> > planning to do with the T&& it receives.
> My draft contains four use-cases from the standard (std::any,
> std::thread, std::range::common_view, and std::not_fn) and one ongoing
> proposal (std::any_invocable.)

Good. Show those examples. "Naming some STL types" is not an example. Show
what those pieces of code look like, for real, before and after your change.

> However, I don't see any need to standardize is_same_uncvref; it seems
> > trivial to write by hand.
> > Notice I could also have done
> > requires (not same_as<remove_cvref_t<T>, AlphableRef>)
> > which is almost as short.
> But not the same.
> The full hand-written example is:
> requires (constructible_from<AlphableRef, T, Args...>
> and not same_as<remove_cvref_t<T>, AlphableRef)

No, that's obviously incorrect. AlphaRef can't have a constructor template
that's gated on whether AlphaRef is constructible from something.
Also, it seems that now you're inventing a parameter `Args...` that wasn't
there in my example? What is `Args...`?

In my specific example, if you wanted to concept-ify it, you'd do:

    // https://godbolt.org/z/e7vErWfPW

    template<class T>
        // "value-construct" from anything that's not cvref-qualified
        requires (not is_same_uncvref<T, AlphableRef>::value)
    AlphableRef(T&& t)
        // ...and that matches our ad-hoc "provides an alpha member" concept
        requires requires { t.alpha(); }

For AlphaRef, there was never any requirement that anything in particular
be constructible-from T itself.

This is why I say your paper (A) is confused, and (B) needs *concrete*
examples. You can't just wave your hands and say "well, let's postulate a
constraint `is_value_constructible_from<Me, T>` that just magically does
the thing I want in each different case." And when you're called on it,
you can't just say "well, how about if we add some `Args...` somewhere."


STD-PROPOSALS list run by std-proposals-owner@lists.isocpp.org

Standard Proposals Archives on Google Groups