[Bug c++/33930] New: [4.3 regression] ambiguous overload in std::stack::push() when using --std=cxx0x

[Bug c++/33930] New: [4.3 regression] ambiguous overload in std::stack::push() when using --std=cxx0x

[a dot chavasse at gmail dot com]

g++43 -v:

Using built-in specs.Target: x86_64-unknown-linux-gnuConfigured with:
../gcc/configure --program-suffix=43 --disable-multilib
--enable-languages=c,c++Thread model: posixgcc version 4.3.0 20071028
(experimental) (GCC)

(it was built from svn trunk, revision 129693)

Not sure whether it really counts as a regression since it happens only in
c++0x experimental mode, but it does prevent code that normally compiles to
fail to do so when enabling c++0x mode, like the following snippet:

==========================
#include <stack>

void foo()
{
  std::stack< const int* > somestack;
  int* someptr = 0;
  somestack.push( someptr );
}
==========================
It fails to build with --std=cxx0x with the following error:

rvalref-overload-stdstack.cpp: In function 'void foo()':
rvalref-overload-stdstack.cpp:7: error: call of overloaded 'push(int*&)' is
ambiguous
/usr/local/lib/gcc/x86_64-unknown-linux-gnu/4.3.0/../../../../include/c++/4.3.0/bits/stl_stack.h:188:
note: candidates are: void std::stack<_Tp, _Sequence>::push(const typename
_Sequence::value_type&) [with
_Tp = const int*, _Sequence = std::deque<const int*, std::allocator<const int*>
>]
/usr/local/lib/gcc/x86_64-unknown-linux-gnu/4.3.0/../../../../include/c++/4.3.0/bits/stl_stack.h:193:
note:                 void std::stack<_Tp, _Sequence>::push(typename
_Sequence::value_type&&) [with _Tp =
 const int*, _Sequence = std::deque<const int*, std::allocator<const int*> >]


Here is a c++0x only test case not involving any headers that triggers the same
error with similar overloads:

==========================
typedef const int* type;

void foo( const type& ggg ) {}
void foo( type&& ggg ) {}

void bar( int* someptr )
{
  foo( someptr );
}
==========================


-- 
           Summary: [4.3 regression] ambiguous overload in
                    std::stack::push() when using --std=cxx0x
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: a dot chavasse at gmail dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33930

[Bug c++/33930] [4.3 regression] ambiguous overload in std::stack::push() when using --std=cxx0x

[a dot chavasse at gmail dot com]

------- Comment #1 from a dot chavasse at gmail dot com  2007-10-28 15:22 -------
(by --std=cxx0x I of course meant --std=c++0x)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33930

[Bug c++/33930] [c++0x] ambiguous overload in std::stack::push() when using --std=cxx0x

[pcarlini at suse dot de]

------- Comment #2 from pcarlini at suse dot de  2007-10-28 15:39 -------
Certainly can't be a regression. And certainly both overloads are present in
the most recent specifications (n2369). Doug, any chance you can comment on
this? I tend to consider this code invalid...


-- 

pcarlini at suse dot de changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |doug dot gregor at gmail dot
                   |                            |com, pcarlini at suse dot de
            Summary|[4.3 regression] ambiguous  |[c++0x] ambiguous overload
                   |overload in                 |in std::stack::push() when
                   |std::stack::push() when     |using --std=cxx0x
                   |using --std=cxx0x           |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33930

[Bug c++/33930] [c++0x] ambiguous overload in std::stack::push() when using --std=cxx0x

[pcarlini at suse dot de]

------- Comment #3 from pcarlini at suse dot de  2007-10-30 19:31 -------
If something (I doubt so, could only be a defect in n2369), this is a C++
issue, because the implementation in the library is totally trivial, identical
to the specifications.


-- 

pcarlini at suse dot de changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|libstdc++                   |c++


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33930

[Bug libstdc++/33930] [c++0x] ambiguous overload in std::stack::push() when using --std=cxx0x

[pcarlini at suse dot de]

------- Comment #4 from pcarlini at suse dot de  2007-10-30 23:28 -------
Actually, I don't know. Definitely the library is conforming to n2369, but
maybe something is wrong in the specifications of the two push. I'm going to
investigate that, let's tentatively consider this as a library issue, because
seems unlikely that the G++ implementation of rvalue-reference overloading
rules is wrong.


-- 

pcarlini at suse dot de changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |pcarlini at suse dot de
                   |dot org                     |
             Status|UNCONFIRMED                 |ASSIGNED
          Component|c++                         |libstdc++
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2007-10-30 23:28:01
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33930

[Bug libstdc++/33930] [c++0x] ambiguous overload in std::stack::push() when using --std=cxx0x

[a dot chavasse at gmail dot com]

------- Comment #5 from a dot chavasse at gmail dot com  2007-10-30 23:51 -------
Well, I did notice that the library was conforming to n2369, but I have really
no idea otherwise about this - except that it did break the compilation of some
really straightforward and innocent looking code of mine that was pushing const
pointers to objects on a stack so it did feel like something was wrong.

Perhaps an explicit cast to a const pointer is supposed to be necessary to
remove the ambiguity, but it feels counter-intuitive since it's my
understanding that the usage of rvalue references (and move semantics) instead
of lvalue references is meant to be a transparent optimization in most cases.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33930

[Bug libstdc++/33930] [c++0x] ambiguous overload in std::stack::push() when using --std=cxx0x

[pcarlini at suse dot de]

------- Comment #6 from pcarlini at suse dot de  2007-10-30 23:56 -------
You are right. I'm investigating the issue with the help of knowledgeable
people in this area, we should be able to resolve it rather quickly. A possible
"solution", which would essentially render the issue moot, is replacing the two
overloads with a single "emplace" overload, as per n2369 in the sequence
containers. We'll see...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33930

[Bug libstdc++/33930] [c++0x] ambiguous overload in std::stack::push() when using --std=cxx0x

[pcarlini at suse dot de]

------- Comment #7 from pcarlini at suse dot de  2007-10-31 00:07 -------
For concreteness would be something like the below (you can try it out, but
consider that the underlying sequence container "emplace" push_back are not
implemented yet, therefore expect only backward compatibility with C++03):

Index: stl_stack.h
===================================================================
--- stl_stack.h (revision 129768)
+++ stl_stack.h (working copy)
@@ -184,14 +184,16 @@
        *  to it.  The time complexity of the operation depends on the
        *  underlying sequence.
        */
+#ifndef __GXX_EXPERIMENTAL_CXX0X__
       void
       push(const value_type& __x)
       { c.push_back(__x); }

-#ifdef __GXX_EXPERIMENTAL_CXX0X__
-      void
-      push(value_type&& __x)
-      { c.push_back(std::move(__x)); }
+#else
+      template<typename... _Args>
+        void
+        push(_Args&&... __args)
+       { c.push_back(std::forward<_Args>(__args)...); }
 #endif

       /**


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33930

[Bug libstdc++/33930] [c++0x] ambiguous overload in std::stack::push() when using --std=cxx0x

[pcarlini at suse dot de]

------- Comment #8 from pcarlini at suse dot de  2007-10-31 22:30 -------
Update: on the LWG reflector Howard Hinnant replied that likely this is a bug
in the G++ implementation of rvalue references (I could not believe that!): the
conversion from int* to const int* leads to a temporary which then should bind
safely to the rvalue reference overload. Orthogonally, people apparently like
the idea of changing the container adaptors per Comment #7, therefore short
term our library can avoid the issue.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33930

[Bug libstdc++/33930] C++0x overloading problem with const lvalref and rvalref

[pcarlini at suse dot de]

------- Comment #9 from pcarlini at suse dot de  2007-11-01 01:24 -------
Ok, so I'm committing momentarily the single push changes to the library (per
my brand new DR 756) and at the same time fixing the Summary and recategorizing
to C++.


-- 

pcarlini at suse dot de changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|pcarlini at suse dot de     |unassigned at gcc dot gnu
                   |                            |dot org
             Status|ASSIGNED                    |NEW
            Summary|[c++0x] ambiguous overload  |C++0x overloading problem
                   |in std::stack::push() when  |with const lvalref and
                   |using --std=cxx0x           |rvalref


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33930

[Bug c++/33930] C++0x overloading problem with const lvalref and rvalref

[dgregor at gcc dot gnu dot org]

------- Comment #10 from dgregor at gcc dot gnu dot org  2007-11-05 18:45 -------
Fixed as part of PR33235


-- 

dgregor at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33930