potential bug regarding rvalue refs?

potential bug regarding rvalue refs?

[Brian Budge]

Hi all -

I have been playing around with universal references and wanted to see
what the difference is between something like

template <typename T>
void f(T t) {
  f_impl(std::forward<foo>(t));
}

and

template <typename T>
void f(T &&t) {
  f_impl(std::forward<foo>(t));
}


I wrote a quick program to start to try and tease out any differences,
but found some totally unintuitive behaviour that I think may be a
bug?

#include <iostream>

#if 0
struct foo {
  int a;

  foo(int b) : a(b) {}
};

#else

typedef int foo;

#endif

void f_impl(foo const &a) {
  std::cerr << "f by const ref" << std::endl;
}

void f_impl(foo &&a) {
  std::cerr << "f by rvalue ref" << std::endl;
}

template <typename T>
void f(T t) {
  f_impl(std::forward<foo>(t));
}

void g_impl(foo const &a) {
  std::cerr << "g by const ref" << std::endl;
}

void g_impl(foo &&a) {
  std::cerr << "g by rvalue ref" << std::endl;
}

template <typename T>
void g(T &&t) {
  g_impl(std::forward<foo>(t));
}

int main(int argc, char **args) {

  foo lvalue = 5;

  f(lvalue);
  f(5);

  g(lvalue);
  g(5);

  return 0;
}


When I compile this via g++ 4.8.2, I get the following output:

:./a.out
f by rvalue ref
f by rvalue ref
g by rvalue ref
g by rvalue ref

Although I'm far from an expert with rvalue refs and universal refs,
this seems like a bug to me.  Thoughts?

Thanks,
  Brian

Re: potential bug regarding rvalue refs?

[Florian Weimer]

On 01/29/2014 07:58 PM, Brian Budge wrote:

>    f_impl(std::forward<foo>(t));

You need to write std::forward<T> instead of std::forward<foo> to get 
the expected forwarding behavior.

> Although I'm far from an expert with rvalue refs and universal refs,
> this seems like a bug to me.  Thoughts?

What output do you expected instead?


-- 
Florian Weimer / Red Hat Product Security Team

Re: potential bug regarding rvalue refs?

[Jonathan Wakely]

This is a question about C++ and not specific to G++ so not really
on-topic on this list.

On 29 January 2014 18:58, Brian Budge wrote:
>
> Although I'm far from an expert with rvalue refs and universal refs,
> this seems like a bug to me.  Thoughts?

G++ is correct, you're forwarding 't' as an rvalue, so overload
resolution chooses the function taking an rvalue reference.

Re: potential bug regarding rvalue refs?

[Brian Budge]

On Wed, Jan 29, 2014 at 12:09 PM, Jonathan Wakely <jwakely.gcc@gmail.com> wrote:
> This is a question about C++ and not specific to G++ so not really
> on-topic on this list.
>
> On 29 January 2014 18:58, Brian Budge wrote:
>>
>> Although I'm far from an expert with rvalue refs and universal refs,
>> this seems like a bug to me.  Thoughts?
>
> G++ is correct, you're forwarding 't' as an rvalue, so overload
> resolution chooses the function taking an rvalue reference.

Florian, Jonathan, thanks for the responses.

  Brian