[SciPy-dev] Re: SciPy 0.1 and Numeric 21

[Tom Loredo]

There appears to be an imcompatibility between the scipy 0.1
installation script and recent Numeric install scripts.

I wrote:
> The build and install goes fine, but "import scipy" fails with:
> >>> import scipy
>
> Traceback (most recent call last):
>   File "<stdin>", line 1, in ?
>   File "/home/laplace/lib/python2.2/site-packages/scipy/__init__.py", line
> 41, in ? from handy import *
>   File "/home/laplace/lib/python2.2/site-packages/scipy/handy.py", line 1,
> in ? import Numeric
>   File "/home/laplace/lib/python2.2/site-packages/Numeric/Numeric.py", line
> 124, in ? arrayrange = multiarray.arange
> AttributeError: 'module' object has no attribute 'arange'

Travis responded:

> This is strange.  It appears to be completely a Numeric issue.  Perhaps you 
> have an older version of Numeric  as a subdirectory of SciPy.  
> 
> I don't understand how SciPy could cause this error. 

Thanks for this insight.  Indeed, "import Numeric" turns up the
problem.  What I did was build SciPy with the latest Numeric
dist'n (21.0b1) in place of the "Numerical" directory.  But
now what I've done is moved the resulting "Numeric" package
directory elsewhere, and installed Numeric-21.0b1 "manually"
by doing the normal "python setup.py install" in the 
Numeric-21.0b1 directory.  This installs Numeric in a way that
works fine.  This installation has several subdirectories that
somehow did not get created when I let the scipy installation
build Numeric.  So there must be something in the scipy install
script that leaves out some important stuff from recent versions
of Numeric.

BTW, now that I have a working Numeric 21 installation, I just
copied the fastumath module that scipy had created when it
installed (incompletely) Numeric 21 to the working installation.
Is there anything else scipy installs under the Numeric package
that I have to worry about?  All scipy tests now pass except
some finitude/infinitude/NaN tests.

Thanks,
Tom Loredo