dolfin team mailing list archive

[mspieg <mspieg@xxxxxxxxxxxxxxxxx>]

Okay,
   I figured it out...it's an odd MacOS issue.  a long story  
short... my mpi libraries were in a non-standard place and the - 
isysroot /Developers/SDKs/MacOSX10.4u.sdk compiler option only works  
properly if your  dynamic libraries for things like mpi  are in /usr/ 
lib or /usr/local/lib (even /usr/local/openmpi won't work).

Apparently MacOSX keeps several separate /usr depending on which  
software developer kit you're writing to.

That's probably not sufficient information for a clean fix, but if  
anyone is interested in what I did, let me know.
cheers
marc

On Feb 28, 2009, at 6:16 AM, Johan Hake wrote:

> On Friday 27 February 2009 21:46:14 mspieg wrote:
> > Hi All,
> >     I've been running into similar problems on macosx with the JIT
> > compiler for Dolfin Functions in PyDolfin
> >
> > info:
> > python-2.5 from the enthought distribution
> > OS is macosx-10.4.11 intel.
> > swig-1.3.36
> > Dolfin 0.9.1 (development version)
> > FFC 0.6.1 (development version)
> > instant (dev version etc...)
>
> ...and ufc dev version? Just to be sure ;)

yes

> > running something like the restriction-interpolation demo, I get
> >
> > Calling FFC just-in-time (JIT) compiler, this may take some  
> > time... done
> > Creating Python extension (compiling and linking), this may take some
> > time... done
> > Calling FFC just-in-time (JIT) compiler, this may take some  
> > time... done
> > Creating Python extension (compiling and linking), this may take some
> > time... done
> > Calling DOLFIN Function just-in-time (JIT) compiler, this may take
> > some time...In instant.recompile: The module did not compile, see '/
> > tmp/tmpVKDvoA2009-2-27-15-07_instant/
> > dolfin_compile_function_227f8d92341151df5a32d2c19fc7a041/compile.log'
> >
> > looking at the log file, the issue seems to be that the compiler is
> > using the options
> >
> > g++ -arch i386 -arch ppc -isysroot /Developer/SDKs/MacOSX10.4u.sdk -g
> > -bundle -undefined dynamic_lookup blah blah blah
>
> I do not think we add this flag. This is probably added by python  
> distutils.
> We do add this manually for both ufc and dolfin when compiling the  
> python
> extension module. Here we use scons as build system.

> > which appends /Developer/SDKs/MacOSX10.4u.sdk/  in front of some of
> > my dynamic library paths which generates missing symbols etc. etc (in
> > addition, I don't want to build for ppc but that doesn't seem as
> > serious)
> >
> > In particular, my openmpi libraries happen to be in  /Applications/
> > mpich/openmpi-1.2.8/lib
> >
> > but ld  complains:
> >
> > warning can't open dynamic library: /Developer/SDKs/MacOSX10.4u.sdk/
> > Applications/mpich/openmpi-1.2.8/lib/libmpi_cxx.0.dylib
> > referenced from: /Users/mspieg/mspieg/hg_repositories/FEniCS/
> > srcFEniCS/dolfin/local/lib/libdolfin.dylib (checking for undefined
> > symbols may be affected) (No such file or directory, errno = 2)
> >
> > and then generates a pile of missing MPI symbols.
> >
> > All the c++ demos work as does some of the mesh and plotting python
> > demos.  Unfortunately, projection from python is just what I need
> > (and I'm hoping there's an easy fix)
>
> I am not a power mac user so I do not know exactly whats going on  
> here.
>
>
> Johan

----------------------------------------------------
Marc Spiegelman
Lamont-Doherty Earth Observatory
Dept. of Applied Physics/Applied Math
Columbia University
http://www.ldeo.columbia.edu/~mspieg
tel: 845 704 2323 (SkypeIn)
----------------------------------------------------