[Pythonmac-SIG] Problems building 2.3.5 source on MacOS X Tiger

[Bill Northcott]

On 24/07/2005, at 4:20 PM, Bob Ippolito wrote:
>
> I fixed this problem for 2.4.1 (CVS HEAD should be fine too).  I  
> don't really consider 2.3 important enough to spend my time  
> backporting.  If you notice, Apple didn't compile Python 2.3 under  
> gcc 4 either.

I got a chance to look at the 'fix' and it is still does not seem  
right to me.

I can't see $(LOCALMODLIBS) in the link or am I missing something on  
this?

The main problem, as I see it, is using libtool to build the  
libraries, which is inherently fragile and is going to keep breaking  
particularly as Apple introduce more changes for the x86 transition.

The documented way is to use the compiler driver gcc.  This will  
automatically call ld, libtool and even lipo repeatedly if necessary  
to construct fat libraries.  It will also automatically include the  
necessary system libraries, which it seems to me are very likely to  
change again.  Finally the command will work with any compiler/MacOS  
combination without modification.

A few small points.  If you are determined to use libtool, it would  
be good to use the full path.  libtool on my machine means /usr/local/ 
bin/libtool which is gnu libtool.  The same would probably happen  
with Fink.   /usr/bin/libtool is guaranteed to be Apple's libtool.

Also I notice that -lcc_dynamic is now removed for gcc-4.0, but I  
don't see why it is there at all.  Its purpose is to get rid of  
unresolved symbols restFP and saveFP.  This problem only happens when  
trying to link objects compiled with an Apple gcc with some compiled  
with the FSF gcc.  Since Apple provide C, C++, Objective-C and  
Objective-C++, this only arises when linking Fortran code which has  
been built with the FSF g77.   Since there is no Fortran in Python,  
it can be removed altogether.

The fix I suggested is much simpler, more robust and removes the need  
to test in configure for different OS/compiler versions.

Finally, I don't see the point of '-arch_only ppc' as far as I know,  
the default is to use the host architecture.  It seems to me that the  
option, which is not configured, just guarantees that things will  
break on x86.

Cheers
Bill