[PATCH] fix-includes for darwin's math.h testing __APPLE_CC__ instead of __GNUC__

[Andrew Pinski <pinskia at physics dot uc dot edu>]

For the following testcase:


#include <math.h>
const volatile static float arg2_float_23 = NAN;

NAN should be allowed in this situation as allowed by C99
but since darwin's headers do:
#if defined(__APPLE_CC__) && (__APPLE_CC__ >= 1345)
#define NAN             __builtin_nanf("0x7fc00000") /* Constant 
expression, can be used as initializer. */
#else
#define NAN             __nan( )
#endif

we get NAN as a function call to __nan which is not a constant value as 
we reject the code.
Since we can fixincludes the check of __APPLE_CC__ to be instead a 
check on what version
of GCC we are using, it fixes the bug.  But that does not work for 
powerpc-darwin7.9 as we
already fix-included math.h by making a new one so we change just 
define __APPLE_CC__ as 1345 in that new header to get around the 
problem here (undefine it afterwards).


OK? Bootstrapped and tested on powerpc-darwin with no regressions (and 
tested fix-includes
also with no regressions).

Thanks,
Andrew Pinski

ChangeLog:

	PR target/15642
	* inclhack.def (AAB_darwin7_9_long_double_funcs [replace}): Define 
__APPLE_CC_
	as 1345.
	(broken_nan): New.
	* fixincl.x: Regenerate.

Attachment: t.diff.txt
Description: Text document