[std-proposals] Template Argument Deduction for call to a base class constructor

[m***@gmail.com]

Consider


template<class T>
struct A
{
A(T) {}
};

template<class T>
struct B : A<T>
{
B(T t, int i)
: A(t) //< nope
{}
};

int main()
{
A c(12); //< yep
}


This case is not considered in p1021
<http://www.open-std.org/JTC1/SC22/WG21/docs/papers/2018/p1021r1.html>. Any
reason what this might be? An omission, a technical issue?
Because sooner or later people will wonder why it works in one context and
not in other.

[Barry Revzin]

Post by m***@gmail.com
This case is not considered in p1021
<http://www.open-std.org/JTC1/SC22/WG21/docs/papers/2018/p1021r1.html>.
Any reason what this might be? An omission, a technical issue?
Because sooner or later people will wonder why it works in one context and
not in other.

Those are pretty different things, despite having the same syntax. In the
mem-initializer list, we already know what the thing we're constructing is
- it would very strange if at that point we stopped and redid overload
resolution to re-figure out what it is we're already in the process of
constructing. What if it happens to give you a different answer?

You can already directly use A<T> there, access the injected-class-name via
B::A.

[m***@gmail.com]

Post by Barry Revzin

Post by m***@gmail.com
This case is not considered in p1021
<http://www.open-std.org/JTC1/SC22/WG21/docs/papers/2018/p1021r1.html>.
Any reason what this might be? An omission, a technical issue?
Because sooner or later people will wonder why it works in one context
and not in other.

Those are pretty different things, despite having the same syntax. In the
mem-initializer list, we already know what the thing we're constructing is
- it would very strange if at that point we stopped and redid overload
resolution to re-figure out what it is we're already in the process of
constructing. What if it happens to give you a different answer?
You can already directly use A<T> there, access the injected-class-name
via B::A.

Yeah, they are different but both are a construction call and this
situation will rise stackoverflow questions for sure.

Is there a problem to resolve to B::A, if it wold fail otherwise? This will
make calls consistent with the expectations, though behind the scenes the
technicalities might be different.