Re: compiling 1.3.4 against python 2.6? -- August 18, 2009 - 11:02

Hi,

That's awesome!  Could you make a patch file that I can apply it (and
credit you for it)?  For this, just type:

$ svn diff > some_patch_file_name

and then send the resultant file as a response to this.  It should
only contain this line change, therefore you need the sconstruct file
to have only this change. Your other changes need to be incorporated
in another patch, and probably discussed as well.  I would like to
make everything automatic for you, but we also need to take into
account that a user may have python and relax installed outside of the
Fink system (i.e. the repository 1.3 branch and some python version
installed as a user in the home directory or desktop).

Cheers,

Edward


2009/8/18 Jack Howarth <howarth@xxxxxxxxxxxxxxxx>:
> Edward,
>   The fix appears to be...
>
> @@ -486,7 +485,7 @@
>
>         # Catch Mac OS X and send the correct command line options to the 
> linker (these may become redundant as SCons improves).
>         if env['PLATFORM'] == 'darwin':
> -            env.Append(LINKFLAGS = ['-bundle', '-bundle_loader', 
> sys.prefix + path.sep + 'bin' + path.sep + 'python', '-dynamic'])
> +            env.Append(LINKFLAGS = ['-bundle', '-bundle_loader', 
> sys.prefix + path.sep + 'bin' + path.sep + 'python', '-dynamic', 
> '-undefined', 'dynamic_lookup'])
>             env['SHLINKFLAGS'] = SCons.Util.CLVar('$LINKFLAGS')
>             env['SHLIBSUFFIX'] = '.so'
>
>
> This should work on Mac OS X 10.4 or later.
>                      Jack
>
> On Mon, Aug 17, 2009 at 07:34:54PM +0200, Edward d'Auvergne wrote:
> > Hi,
> >
> > If that is the case, we need to switch back to the dynamiclib
> > approach.  Can you delete the 3 'sconstruct' script lines after "#
> > Catch Mac OS X and send the correct command line options to the
> > linker" and see if that works.  Or try gcc with the '-dynamiclib' flag
> > (note that some of the other flags, although I'm not sure which, need
> > to be removed).  It could be quite possible that the Fink Python 2.6
> > installation has removed the need for the bundle-loader approach :S
> >
> > Cheers,
> >
> > Edward
> >
> >
> > 2009/8/17 Jack Howarth <howarth@xxxxxxxxxxxxxxxx>:
> > > Edward,
> > >    Actually, I don't think the -bundle-loader approach will work
> > > with python2.6 because it is linked against the libpython2.6 dylib
> > > and not the static lib like python2.5. If you compare the symbols
> > > returned fron "nm /sw/bin/python2.6" with "nm /sw/bin/python2.5",
> > > the first case is a very short list...
> > >
> > > 0000200c D _NXArgc
> > > 00002008 D _NXArgv
> > >         U _PyMac_Error
> > >         U _Py_Main
> > > 00002000 D ___progname
> > > 00001fa8 t __dyld_func_lookup
> > > 00001000 A __mh_execute_header
> > > 00002004 D _environ
> > >         U _exit
> > > 00001fd0 T _main
> > > 00002010 d dyld__mach_header
> > > 00001f78 t dyld_stub_binding_helper
> > > 00001f30 T start
> > >
> > > compared to the latter case. I really think we need to explicitly
> > > link in /sw/lib/python2.6/config/libpython2.6.dylib.
> > >              Jack
> > >
> > > On Mon, Aug 17, 2009 at 07:08:31PM +0200, Edward d'Auvergne wrote:
> > > > Hi,
> > > >
> > > > I don't think this is an issue, as the include directories should only
> > > > contain *.h C header files.  Can you find any *.h python files which
> > > > are not in /sw/include/python2.6?  The second include directory is
> > > > just for the numpy *.h files.
> > > >
> > > > Cheers,
> > > >
> > > > Edward