[Numpy-discussion] numpy install on mac os x 10.4

belinda thom bthom at cs.hmc.edu
Sun Dec 31 01:16:18 CST 2006

Not sure if this helps but I stumbled upon the following trick.

Do the following:

1) install g77 via instructions at:


in particular download:


and then do:

   sudo tar -xvf g77v3.4-bin.tar -C /

which installs everything in /usr/local, most importantly creating:

   lrwxrwxrwx    1 root  wheel       18 Dec 30 15:29 libg2c.0.dylib@ - 
 > libg2c.0.0.0.dylib
   lrwxrwxrwx    1 root  wheel       18 Dec 30 15:29 libg2c.dylib@ ->  
   lrwxrwxrwx    1 root  wheel       18 Dec 30 15:29 libgcc_s.dylib@ - 
 > libgcc_s.1.0.dylib

in /usr/local/lib/.

2) once this was done, magically scipy could work. in particular,  
these libg2c dylibs need to be on my machine.

Not sure if this is equivalent to "getting the fortran stuff to  
work", but it at least allowed me to run all the spicy.test()

On Dec 30, 2006, at 11:02 PM, Christopher Barker wrote:

> belinda thom wrote:
>> Advice on a painless way to install scipy on my G5 OS X 10.4.8 mac
>> greatly appreciated.
> Sorry, there isn't one at this point -- I think numpy, SciPy, MPL, and
> wx are all fairly stable right now, so it's a pretty good time to  
> do it,
> but it's a challenge because:
> 1) SciPy, MPL, numpy and wx all have to be compatible -- so it's  
> best if
> the same person does them all, or at least communicates enough to make
> sure the packages at pythonmac all match.
> 2) Building MPL requires the Universal version of a few libs (though
> libpng may be the only one now (or is it libjpeg? -- don't have my Mac
> handy), as there MAY be a version of libfreetype that works  
> provided by
> Apple now.
> This is the hardest one:
> 3) SciPy (or at least parts of it) requires Fortran. Apple has not
> released a gcc Fortran, and the ones that do exist are not Universal,
> and require libs in inconvenient places. This makes it hard to build a
> Universal, easy to install, binary of SciPy -- It's still hard to  
> build
> one yourself, but if you don't need it universal, it is doable by mere
> mortals. I'd love to see a Universal one in the pythonmac repository

I think more on this point could really be helpful.

I'm working off of Python2.4 because some AI-related code crashes on  
2.5, so I hope this can still remain a priority.

For me, the following combo seems to work good enough:

1) get matplotlib from www.macpython.org/packages/py24-fat

(The superpack at scipy fails b/c a TkAgg library can't be found)

2) use the superpack for installing numpy and scipy.

(These will only work for me if I've already done the g77 trick.)

3) use ipython gotten via easy_install

(The superpack version is broken b/c it doesn't provide an executable)

> (and I think with the right incantations of lipo , it should be  
> doable),
> but in the meantime, maybe we should at least have separate PPC and
> Intel versions-- and is there any chance of either statically  
> linking or
> putting the libs in the Python tree somewhere?

A need for a "scipy" package that actually works w/ipython,  
matplotlib, numpy, and scipy is seriously needed. Most of my friends  
think I'm crazy to have wasted all my time on this stuff. Its  

> So, I'd love to have someone:
> Start with Python2.5 from pythonmac (2.4 would be nice too -- but  
> let's
> focus on 2.5)
> 1) Get the latest wxPython for OS-X (2.8.*)
> 2) Build the latest numpy
> 3) Build MPL against the above (and Numeric and numarray, if possible)
> 4) Build SciPy for both Intel and PPC (probably separately)
> 5) Put all that up on pythonmac.
> I'd like to do it, but I'm not the least bit sure when I'll be able  
> to--
> someone please beat me to it!

I would like to say I could help, but: i) I'm pretty new to all this,  
ii) I've wasted so much time getting something running on my machine  
that I'm running out of time (have a class I need to teach to prepare  
for). Hopefully my comments here are at least helpful :-)


p.s. I'd like to thank you for all the work you've done on this  
regard. I'm beginning to realize how time consuming this open source  
stuff can be...

More information about the Numpy-discussion mailing list