[SciPy-user] Installation from source on OS X: 'NoneType' object has no attribute 'link_shared_object'

Robert Kern robert.kern@gmail....
Fri Sep 26 18:14:21 CDT 2008


On Fri, Sep 26, 2008 at 17:55, Michael Williams <williams@astro.ox.ac.uk> wrote:
> On Tue, Sep 23, 2008 at 01:53:21PM -0500, Robert Kern wrote:
>> Unfortunately, now it looks like you are dealing with a bug in
>> gfortran. Can you downgrade to something earlier than 4.3.0?
>
> With difficulty (hence the delayed reply). I'm trying to install without
> user privileges and I'm not experienced with gcc installation.
>
> That being said, I seem to have succeeded. I installed gfortran 4.4.0
> from the HPC on OS X page into ~/usr/local (rather than /usr/local),
> added it to my shell's path and ran setup.py with the usual gnu95
> options. It seemed to build and install fine. (Does anyone less ignorant
> than I am about compilers and libraries know if I need to keep the
> contents of ~/usr/local around to use scipy?)

Yes. gfortran will link against the shared libraries there unless if
you do a fairly complicated dance to make it link against the static
libraries.

I very much recommend against using the binaries from HPC. They
release binaries for buggy bleeding-edge versions of gfortran, and
don't keep previous versions around. I have had *much* more luck with
the binaries over here:

  http://r.research.att.com/tools/

Since you don't have admin privileges, you need to do a little command
line work instead of just being able to use the installer. Mount
gfortran-4.2.3.dmg. At the terminal:

  # Assuming you want it in ~/usr/local/
  $ cd ~/
  $ pax -zr < /Volumes/GNU\ Fortran\ 4.2.3/gfortran.pkg/Contents/Archive.pax.gz

Now you should have everything unpacked into ~/usr/local/.

But if everything is currently working for you, I would probably leave
well enough alone.

> I can import cleanly (Deprecation Warning about NumpyTest
> notwithstanding). scipy.test runs 1848 tests with one failure
> (check_dot). Last time I installed scipy I think most people were not
> too worried by errors in scipy.test() on OS X. Is that still the case?

You're probably not using that function, so it's fine in that sense.
But it is a bona fide bug in 0.6.0 and has been fixed on the trunk.

-- 
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless
enigma that is made terrible by our own mad attempt to interpret it as
though it had an underlying truth."
  -- Umberto Eco


More information about the SciPy-user mailing list