[result_of] Inconsistent results between function references and function pointers

[Michel Morin]

For function references, decltype-based boost::result_of uses decltype to determine the type. But, for function pointers, decltype-based boost::result_of falls back to boost::tr1_result_of and does not use decltype.

This inconsistency makes different results for function references and function pointers. On clang trunk and gcc-4.7 in a C++11 mode, for "typedef const int F();",

• decltype-based boost::result_of<F&()> is int, but
• decltype-based boost::result_of<F*()> is const int.

To remove the inconsistency, decltype-based boost::result_of should use decltype for function pointers.

A patch attached.

[Michel Morin]

This patch fixes the inconsistency. This patch also fixes the test so that the patched result_of passes the test. BOOST_STATIC_ASSERT((is_same<result_of<func_ptr(void)>::type, int>::value)); is removed, since this is uncallable. (BOOST_STATIC_ASSERT((is_same<tr1_result_of<func_ptr(void)>::type, int>::value)); is kept untouched.)

[Daniel Walker]

(In [77702]) Applied patch from Michel Morin to fix #6755. Updated copyright to recognize contributors from the past several years.

[Daniel Walker]

This is somewhat related to #6731. Really, neither pointers nor member pointers should be dispatched to TR1 result_of. This patch takes us part of the way there.

[Eric Niebler]

(In [80359]) merge [77702] to release, fixes #6755