[SciPy-user] installation on fedora core 4

Stephen Walton stephen.walton at csun.edu
Mon Oct 10 11:31:21 CDT 2005


Ryan Krauss wrote:

>Do I need to speficy a version of gcc other than 4.0?
>
This thread came up in scipy-dev, and I've posted here before, but to 
clarify again:

In an all-default configuration, it is not presently possible to build a 
working scipy on Fedora Core 4, either old or new.  There is a bug in 
gfortran which causes it to mis-compile the i1mach, r1mach, and d1mach 
functions which are at the bottom of much of the Fortran code in scipy.  
If you use g77 instead by using the appropriate compatibility RPMs 
(compat-gcc-32-g77 and compat-libf2c-32), you'll find that gcc4 cannot 
link to libg2c because it isn't in a directory which gcc4 searches.  I 
haven't found a way of forcing a scipy build to use gcc 3.2.3 (the 
compatibility version which is part of FC4).

The only feasible workaround I've found is to use g77 on FC4 to build 
scipy (and Numeric and numarray), and create symbolic links to libg2c.a 
and libg2c.so so that gcc4 can find them.  That is, do the following:

ln -s /usr/lib/gcc-lib/i386-redhat-linux/3.2.3/libg2c.so 
/usr/lib/gcc/i386-redhat-linux/4.0.1
ln -s /usr/lib/gcc-lib/i386-redhat-linux/3.2.3/libg2c.a 
/usr/lib/gcc/i386-redhat-linux/4.0.1

This is not supported, use at your own risk, your mileage may vary, 
results are not guaranteed, pressure RedHat to release a gcc 4.0.2 
update for FC4 in which the gfortran bug is fixed.  Personally, I'm 
using Absoft Fortran on gcc4;  if you're an educational institution, 
Intel will let you download and use their Fortran compiler for free.  
Both seem to work fine with newscipy and FC4.





More information about the SciPy-user mailing list