Robert Schwebel wrote:
[adding ptxdist to Cc:]
On Wed, Dec 17, 2008 at 09:55:16AM +0100, Erwin Rol wrote:
I am not sure this is the correct fix. The problem seems that ptxdist
uses ac_cv_sizeof_long_double with a value of 8, and that is where the
problem is, because that should be 12 on x86 and 16 on x86_64. That
error results in double and long double to have the same size and
configure to conclude there is no long double. And since that is never
true on x86 with gcc, there was no need for the ifdef with
HAVE_LONGDOUBLE in the x86 case.
Hmm, sounds reasonable. Nevertheless, the fix should be correct anyway,
because if the test detects *no* long_double, it shouldn't fail
compiling.
Now it likely will fail on runtime when you use a long double, that
might be even worse than failing at compile time :-)
But the real bug is that ptxdist always sets ac_cv_sizeof_long_double to
8 no matter what the architecture is. That might also be a silent
problem for other programs that use long double's.
- Erwin
rsc
--
ptxdist mailing list
[email protected]
