[SciPy-User] worst-case scenario installation

[DEMOLISHOR! the Demolishor]

Folks,
  There is mounting evidence to suggest that I cannot install SciPy/ Numpy
under these restrictions:

  - I do not have root access to my machine
  - I use a later version of Python than my distribution provides (2.6
instead of 2.4--I am using RHEL5), built locally and installed in $HOME/bin
-- this prevents me from asking my sysadmin to install
numpy and scipy from the YUM repositories.
  - both g77 and gfortran are installed in /usr/bin and thus I cannot modify
my path to exclude g77 without fear of crippling the rest of the build
system in mysterious ways.

So here's what happens:
  - Using hand-built versions of ATLAS and LAPACK (following the
instructions on the ATLAS website) result in undefined reference errors in
liblapack.so to function calls that live in libblas.so
  - Using RPM versions of ATLAS and LAPACK from a RPM search site (I just
extracted and copied the libraries and headers into the relevant
directories) breaks down in the same way, and no RPM exists with these
libraries
built with g77.
  - Building ATLAS by hand using the instructions on the SciPy website  (
http://www.scipy.org/Installing_SciPy/Linux#head-89e1f6afaa3314d98a22c79b063cceee2cc6313c)
also breaks down because I cannot avoid having g77 in my path. Also I can no
longer build LAPACK with g77 (
http://www.netlib.org/lapack/lapack-3.2.html#_7_install_procedure).

The most aggravating part of this entire process is that I am going through
all these gyrations to avoid the Numpy setup script's obsession with g77. I
do not understand why the `--fcompiler' option doesn't work , or even
putting gfortran ahead of g77 in my PATH. I realize I am not a typical user,
but these things seem really simple to fix for someone that knows where to
look.

And as I was writing this, I figured this out:
I ended up linking *everything* in /usr/bin to ~/bin, removing /usr/bin from
my PATH, and then removing the links to g77 and f77 from ~/bin, and
rebuilding. Now Numpy fails 1 test, and SciPy fails 4, and I guess if I am
lucky I'll never need the stuff that doesn't pass the tests. So it was *not*
impossible to install an at least partial version of Scipy/Numpy on my
system.

Bottom line: weird people like me shouldn't suffer because of unimplemented
command-line options. That said, weird people are like me are grateful that
stuff like Scipy and Numpy exist because they keep me from having to use
ROOT.

Thanks,
Craig
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.scipy.org/pipermail/scipy-user/attachments/20090819/f3339d65/attachment.html>