Re: 1.8.5p1 build failure on Solaris 10

"Daniel Berger" <[e-mail]> wrote on 30.06.2006 16:18:26:
> 
> I don't totally understand how the build process works, which is part of 
my 
> problem.  For example, I see a missing.h, and I can see 
> AC_REPLACE_FUNCS in the 
> configure.in file, but I'm not sure if/how they're tied together. Within 

> missing.h I see this:
> 
> #ifndef HAVE_ISINF
> # if defined(HAVE_FINITE) && defined(HAVE_ISNAN)
> # define isinf(x) (!finite(x) && !isnan(x))
> # else
> extern int isinf _((double));
> # endif
> #endif
> 
> So, in simple terms I think the logic should be:
> 
> a) On the Solaris platform...
> 
> b) If a) is true and the sunmath lib is present, and it contains the 
isinf 
> function, then link against it.
> 
> c) If both a) and b) are true and the sunmath.h header file is 
> present, and it 
> contains the isinf function, then include it.
> 
> d) If a) or b) fail then resort to the implementation in missing.h. This 

> wouldn't require adding more source code from what I can gather, 
> though I'm not 
> positive.
> 
> Does that seem reasonable?  I would try to implement this myself, but 
I'm not 
> sure how.
> 
 Its all doable, but I don't understand what you are trying to achieve. 
Didn't my patch
 fix your problem? My patch should basically make the configure script do 
the steps you have listed
 above. The problem is that the isinf can be builtin function (inline 
macro) and this might confuse
 autoconf detection logic quite badly. There is some discussion avaiable 
in gcc archive about this
 issue. 

 - Ville