add _HAVE_LONG_DOUBLE flag, math.h update

[Howland Craig D (Craig)]

Jeff and I had some off-list discussions about this.  The end result is
that the fixed-number method (e.g. #define NAN (0.0F/0.0F)) is the
only one that works correctly in all situations when the 3.3 and later
GCC builtins are not used in math.h.  No warnings are produced for old
GCC versions that use them (at least not 3.2.3 and 2.95, which I have
to test with).
 
g++ would warn for the divide-by-0 in the NAN definition, so a
workaround
to avoid such warnings is added for GCC C++.  This method errors out
with plain gcc, which is why it is a special case for only g++.
 
Other than the tweak in the math.h changes, the other file changes in
the
patch are identical to the prior patches.
 
Craig
-------------- next part --------------
A non-text attachment was scrubbed...
Name: math3.patch
Type: application/octet-stream
Size: 9485 bytes
Desc: math3.patch
URL: <http://sourceware.org/pipermail/newlib/attachments/20090515/ab4a1d50/attachment.obj>