On 02/19/2016 12:45 PM, Paul Eggert wrote:
> On 02/19/2016 11:32 AM, Orion Poplawski wrote:
>> g++ -DHAVE_CONFIG_H -I. -I.. -I../lib -g -O2 -MT foobar.o -MD -MP -MF
>> .deps/foobar.Tpo -c -o foobar.o foobar.cpp
>> In file included from foobar.cpp:3:0:
>> ../lib/math.h: In function ‘int signbit(float)’:
>> ../lib/math.h:2576:1: error: ‘int signbit(float)’ conflicts with a previous
>> declaration
> This is a C++ thing, plus I don't have GCC 6, so I'm afraid you'll have to do
> most of the legwork.
>
> Why was Gnulib trying to define its own signbit function? Why didn't the
> signbit function from GCC 6 pass Gnulib's test? Look in config.log to find
> out. On my platform (Fedora 23, GCC 5.3.1) signbit passes all the tests so
> there's no reason for Gnulib to try to substitute its own.
It appears that if gnulib detects the "signbit compiler built-ins" it insists
on using them.
I end up with:
GNULIB_SIGNBIT='1'
REPLACE_SIGNBIT='0'
REPLACE_SIGNBIT_USING_GCC='1'
configure:7721: checking for signbit macro
configure:7826: gcc -o conftest -g -O2 conftest.c >&5
configure:7826: $? = 0
configure:7826: ./conftest
configure:7826: $? = 0
configure:7837: result: yes
configure:7839: checking for signbit compiler built-ins
configure:7941: gcc -o conftest -g -O2 conftest.c >&5
configure:7941: $? = 0
configure:7941: ./conftest
configure:7941: $? = 0
configure:7952: result: yes
and math.h has:
#if 1
# if 1
# undef signbit
/* GCC 4.0 and newer provides three built-ins for signbit. */
# define signbit(x) \
(sizeof (x) == sizeof (long double) ? __builtin_signbitl (x) : \
sizeof (x) == sizeof (double) ? __builtin_signbit (x) : \
__builtin_signbitf (x))
# endif
# ifdef __cplusplus
# ifdef signbit
_GL_MATH_CXX_REAL_FLOATING_DECL_1 (signbit)
# undef signbit
_GL_MATH_CXX_REAL_FLOATING_DECL_2 (signbit)
# endif
# endif
In the preprocessed output I get:
constexpr bool
signbit(float __x)
{ return __builtin_signbit(__x); }
constexpr bool
signbit(double __x)
{ return __builtin_signbit(__x); }
constexpr bool
signbit(long double __x)
{ return __builtin_signbit(__x); }
template<typename _Tp>
constexpr typename __gnu_cxx::__enable_if<__is_integer<_Tp>::__value,
bool>::__type
signbit(_Tp __x)
{ return __x < 0 ? true : false; }
from /usr/include/c++/6.0.0/cmath
and:
inline int signbit (float f) { return _gl_cxx_signbitf (f); } inline int
signbit (double d) { return _gl_cxx_signbitd (d); } inline int signbit (long
double l) { return _gl_cxx_signbitl (l); }
from lib/math.h (gnulib).
But this is incompatible with the new g++ 6 definitions.
--
Orion Poplawski
Technical Manager 303-415-9701 x222
NWRA, Boulder/CoRA Office FAX: 303-415-9702
3380 Mitchell Lane or...@nwra.com
Boulder, CO 80301 http://www.nwra.com
