[v3] fix libstdc++/48521

[Jonathan Wakely]

2011-04-19  Jonathan Wakely  <jwakely.gcc@gmail.com>

        PR libstdc++/48521
        * include/std/type_traits (result_of): Handle pointer to member.
        * include/std/functional (__invoke): Likewise.
        (_Function_to_function_pointer): Remove.
        (_Reference_wrapper_base): Provide nested types independent of
        unary_function and binary_function.
        (reference_wrapper::operator()): DR 2017.
        (ref(const A&&), cref(const A&&): Define as deleted.
        * include/std/future (async): Simplify SFINAE and use result_of to
        support pointer to member.
        * testsuite/20_util/reference_wrapper/invoke.cc: Test pointer to
        member.
        * testsuite/20_util/reference_wrapper/24803.cc: Likewise.
        * testsuite/20_util/reference_wrapper/typedefs.cc: Test for types
        instead of derivation from unary_function and binary_function.
        * testsuite/20_util/declval/requirements/1_neg.cc: Adjust.
        * testsuite/20_util/reference_wrapper/invoke-2.cc: New.
        * testsuite/20_util/reference_wrapper/ref_neg.c: New.
        * testsuite/20_util/reference_wrapper/typedefs-3.c: New.

This resolves a few issues in <functional> which I first noticed
because our std::async didn't work with pointers to member. To fix
that I needed to make result_of support pointer to member, then use
result_of in async, which required changing the enable_if otherwise we
form an invalid result_of expression outside the immediate context of
the type deduction, where SFINAE doesn't apply.

I implemented DR 2017 (reference wrapper makes incorrect use of
result_of) and because we now use a reference to the wrapped type in
the result_of expression we don't need Function_to_function_pointer to
prevent forming an invalid expression.

I made reference_wrapper define the nested
{first_,second_,}argument_type typedefs correctly if the wrapped type
defines them. Previously we only did so if the nested type derived
from {unary,binary}_function but the C++0x draft was changed to stop
depending on those types.

Finally, I've also added missing ref/cref overloads to prevent
creating reference_wrapper to rvalues.

mem_fn doesn't yet support varargs member functions, I'll fix that
when PR c++/48424 is fixed.

Tested x86_64-linux, committed to trunk.  Will wait a day or two and
commit to 4.6 too.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 48521.diff
Type: text/x-patch
Size: 27429 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20110419/76ffa4ef/attachment.bin>