Re: [PFR][review] Andrzej's review

From: Andrzej Krzemienski (akrzemi1_at_[hidden])
Date: 2020-10-06 11:42:05

wt., 6 paź 2020 o 13:29 Peter Dimov via Boost <boost_at_[hidden]>
napisał(a):

> Andrzej Krzemienski wrote:
>
> > 3. The library buds on a basic concept of "Base-less aggregate".
>
> That's not really true though, is it? E.g.
>
> #include <boost/pfr.hpp>
> #include <string>
> #include <iostream>
>
> struct X
> {
>     std::string a;
> };
>
> struct Y
> {
>     std::string b;
> };
>
> struct Z: X, Y
> {
>     std::string c;
> };
>
> int main()
> {
>     std::cout << boost::pfr::tuple_size_v<Z> << std::endl;
> }
>
> compiles and prints "3". Bases are treated as members by the precise API.
>

But this is only for tuple_size. If you do a for_each, the compilation
breaks:

```
Z z {{}, {}, {}};
boost::pfr::for_each_field(z, [](auto const&){});
```

This is because brace initialization of aggregates is "incompatible" with
structured binding for aggregates now in C++20.

I still acknowledge that I may have not understood the requirement that PFR
puts on types. This makes my request even stronger: I have to know what it
expects of my types to work. I expect that this would work when I switch
between C++17 and C++20. (C++20 changed the initialization of aggregates:
base is treated as yet another member. But it did not change the structured
binding.)

Regards,
&rzej;