[Bug c++/50929] New: [C++0x] Wrong function selected for overload with template and rvalue reference

[Bug c++/50929] New: [C++0x] Wrong function selected for overload with template and rvalue reference

[jarrydb at cse dot unsw.edu.au]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50929

             Bug #: 50929
           Summary: [C++0x] Wrong function selected for overload with
                    template and rvalue reference
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: jarrydb@cse.unsw.edu.au


When choosing between a constructor

  A(const A& a)

and

  template <typename T>
  A(T&& t)

given an object of type A&, the compiler chooses the second function. For the
following code:

class A
{
  public:
  A() : value(0) {}

  A(const A& rhs) : value(1) { }

  A(A&& rhs) : value(2) { }

  template <typename T>
  A(T&& t) : value(3) { }

  int value;
};

int main()
{
  A a;
  A b = a;
  return b.value;
}

compiled with:
  g++ -std=gnu++0x rvalue_deduction.cpp

then running:
  ./a.out
  echo $?

the output is 
  3

I expect the output to be 1, not 3.


Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/home/jarrydb/current/soft/install-latest/libexec/gcc/x86_64-unknown-linux-gnu/4.7.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: /home/jarrydb/current/soft/src/gcc-git/configure
--prefix=/home/jarrydb/current/soft/install-latest --disable-multilib
--enable-languages=c,c++
Thread model: posix
gcc version 4.7.0 20111027 (experimental) (GCC)

[Bug c++/50929] [C++0x] Wrong function selected for overload with template and rvalue reference

[daniel.kruegler at googlemail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50929

Daniel Krügler <daniel.kruegler at googlemail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |daniel.kruegler at
                   |                            |googlemail dot com

--- Comment #1 from Daniel Krügler <daniel.kruegler at googlemail dot com> 2011-10-31 07:15:48 UTC ---
Your expectations are in violation to the standard ;-). The constructor is
selected by overload resolution. The non-template constructor A(const A& a)
would be preferred, if there would be a perfect match (e.g. if you had an
lvalue of type const A), but in your example you have an lvalue of non-const A.
The "perfect-forwarding" template constructor deduces to

  template <> A(A& t)

via reference-collapsing, which is a better match for the argument expression
a.
So, this looks like an invalid issue to me.

[Bug c++/50929] [C++0x] Wrong function selected for overload with template and rvalue reference

[jarrydb at cse dot unsw.edu.au]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50929

--- Comment #2 from Jarryd Beck <jarrydb at cse dot unsw.edu.au> 2011-10-31 07:42:54 UTC ---
That's a shame, and rather annoying. I read pages and pages of the standard to
try to understand this one. Do you know where this is explained?

I added a non-const constructor and also tried it with a const object, they
worked as you said.

[Bug c++/50929] [C++0x] Wrong function selected for overload with template and rvalue reference

[daniel.kruegler at googlemail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50929

--- Comment #3 from Daniel Krügler <daniel.kruegler at googlemail dot com> 2011-10-31 08:24:58 UTC ---
(In reply to comment #2)
> That's a shame, and rather annoying. I read pages and pages of the standard to
> try to understand this one. Do you know where this is explained?

Actually this is already required in C++03 (modulo perfect forwarding), see
[class.copy], footnote 106:

"Because a template constructor is never a copy constructor, the presence of
such a template does not suppress the implicit declaration of a copy
constructor. Template constructors participate in overload resolution with
other constructors, including copy constructors, and a template constructor may
be used to copy an object if it provides a better match than other
constructors."

In C++11 this is just part of the general description of initialization
semantics in N3290, [dcl.init] p16.

[Bug c++/50929] [C++0x] Wrong function selected for overload with template and rvalue reference

[jarrydb at cse dot unsw.edu.au]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50929

--- Comment #4 from Jarryd Beck <jarrydb at cse dot unsw.edu.au> 2011-10-31 10:07:04 UTC ---
I didn't realise that A(A&) was a better match. I was thinking of C++ code
where you might write:

  template <typename T>
  A(const T& t);

  A(const A& rhs);

in which case A(const A&) would be chosen. I didn't realise that the rvalue
reference puts a spanner in the works. I was reading the part about a template
function and a non-template function being candidates, and if the non-template
function is no worse than the template function then it is chosen. But I didn't
realise that in this case the non-template one is actually worse.

I suppose that the bug can probably be marked as invalid then.

[Bug c++/50929] [C++0x] Wrong function selected for overload with template and rvalue reference

[daniel.kruegler at googlemail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50929

--- Comment #5 from Daniel Krügler <daniel.kruegler at googlemail dot com> 2011-10-31 10:18:04 UTC ---
(In reply to comment #4)
> I didn't realise that A(A&) was a better match. I was thinking of C++ code
> where you might write:
> 
>   template <typename T>
>   A(const T& t);
> 
>   A(const A& rhs);
> 
> in which case A(const A&) would be chosen. 

Sure. But if you had provided

    template <typename T>
    A(T& t);

instead that would again be a better match for a non-const A lvalue.
Non-template functions only win over templates, if they are otherwise equal in
matching.

> I didn't realise that the rvalue reference puts a spanner in the works.

Note that "rvalue-reference" is a red herring for the perfect-forwarding
signature

    template <typename T>
    A(T&& t);

It depends on whether the argument is an lvalue or not to decide whether the
finally deduced type is effectively A& or A&& (or whatever argument type had
been provided). In your example the argument was an lvalue of type A, therefore
the deduced signature was effectively

    template <> A(A& t);

[Bug c++/50929] [C++0x] Wrong function selected for overload with template and rvalue reference

[jarrydb at cse dot unsw.edu.au]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50929

--- Comment #6 from Jarryd Beck <jarrydb at cse dot unsw.edu.au> 2011-10-31 10:34:16 UTC ---
Yeah I understand now. Basically having a perfect forwarding constructor means
that I need a non-const constructor if I want my example to work as I initially
said.

[Bug c++/50929] [C++0x] Wrong function selected for overload with template and rvalue reference

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50929

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID

--- Comment #7 from Paolo Carlini <paolo.carlini at oracle dot com> 2011-10-31 10:40:41 UTC ---
Thanks Daniel, let's close this then.

[Bug c++/50929] [C++0x] Wrong function selected for overload with template and rvalue reference

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50929

--- Comment #8 from Jonathan Wakely <redi at gcc dot gnu.org> 2011-10-31 11:45:32 UTC ---
this is a duplicate of PR 46004

[Bug c++/50929] [C++0x] Wrong function selected for overload with template and rvalue reference

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50929

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|INVALID                     |DUPLICATE

--- Comment #9 from Paolo Carlini <paolo.carlini at oracle dot com> 2011-10-31 12:19:51 UTC ---


*** This bug has been marked as a duplicate of bug 46004 ***