[SciPy-dev] newcore and gcc4/gfortran (SUCCESS)

[Ravikiran Rajagopal]

On Tuesday 22 November 2005 03:06, Arnd Baecker wrote:
> Many thanks -
>   python setup.py config_fc --fcompiler=gnu95 -lg2c build
> worked fine for me.
>
> And with
>
>  >>> scipy.__core_version__   '0.7.1.1526'
>  >>> scipy.__scipy_version__  '0.4.2_1456'
>
> the check_bdtrik (scipy.special.basic.test_basic.test_cephes)
> Segmentation fault is gone!

Well, I am happy to report that I can do one better. With gcc4/gfortran combo 
from SVN yesterday, along with the usual Redhat/Fedora patches (see 
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=173771 for the complete 
discussion regarding SciPy), I can now compile ATLAS and SciPy with no 
problems and all tests pass with the following patch:

Index: Lib/lib/blas/fblaswrap.f.src
===================================================================
--- Lib/lib/blas/fblaswrap.f.src	(revision 1452)
+++ Lib/lib/blas/fblaswrap.f.src	(working copy)
@@ -1,7 +1,7 @@
 c     <p2=c,z,c,z> <s2=u,u,c,c> <ftype2=complex,double complex,\0,\1>
       subroutine w<p2>dot<s2> (r, n, x, incx, y, incy)
       external <p2>dot<s2>
-      complex <p2>dot<s2>, r
+      <ftype2> <p2>dot<s2>, r
       integer n
       <ftype2> x (*)
       integer incx

For Fedora Core 4 users who would like to play with this, please use 
gcc-4.0.2-7 currently available in FC4 updates-testing repository. I am using 
SciPy core 1526 and scipy 1452.

Question: Scipy runs only 1370 tests on my machine. Is that the right number 
or am I missing some?

Another question for Arnd: Why do you need g2c when you build with gfortran?

Regards,
Ravi