pkg/40103: problem with math/blas/files/Makefile.blas

[russell_mcmanus%yahoo.com@localhost]

>Number:         40103
>Category:       pkg
>Synopsis:       problem with math/blas/files/Makefile.blas
>Confidential:   no
>Severity:       serious
>Priority:       high
>Responsible:    pkg-manager
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Thu Dec 04 14:35:00 +0000 2008
>Originator:     Russell McManus
>Release:        rhel 5
>Organization:
>Environment:
Linux frln31 2.6.9-67.0.15.ELsmp #1 SMP Tue Apr 22 13:58:43 EDT 2008 x86_64 
x86_64 x86_64 GNU/Linux
>Description:
The python numpy package does play nice with pkgsrc-2008Q3:

    $ echo import numpy | python2.5
    Traceback (most recent call last):
      File "<stdin>", line 1, in <module>
      File 
"/glib/pkgsrc-2008Q3/x86_64/lib/python2.5/site-packages/numpy/__init__.py", 
line 43, in <module>
        import linalg
      File 
"/glib/pkgsrc-2008Q3/x86_64/lib/python2.5/site-packages/numpy/linalg/__init__.py",
 line 4, in <module>
        from linalg import *
      File 
"/glib/pkgsrc-2008Q3/x86_64/lib/python2.5/site-packages/numpy/linalg/linalg.py",
 line 25, in <module>
        from numpy.linalg import lapack_lite
    ImportError: /glib/pkgsrc-2008Q3/x86_64/lib/libF77.so.0: undefined symbol: 
MAIN__

This problem is similar to the one described here:

  http://mail-index.netbsd.org/pkgsrc-users/2008/10/21/msg008397.html

From the above error message, one might think that the problem is
related to libF77.so.0.  However the problem is actually in math/blas.

I happened to have a pkgsrc-2008Q2 tree lying about which did not have
the problem.  Using this, I was able to isolate the issue.

Note the following:

  $ nm /glib/pkgsrc-2008Q2/x86_64/lib/libblas.so.1.0.1 | grep MAIN__
  $ nm /glib/pkgsrc-2008Q3/x86_64/lib/libblas.so.1.0.1 | grep MAIN__
          U MAIN__

After reverting math/blas/files/Makefile.blas from version 1.2 down to
version 1.1, MAIN__ no longer appears undefined in libblas.so.1.0.1,
and the import of numpy succeeds.

The "cvs log" message related to the 1.2 checkin is as follows:

    ----------------------------
    revision 1.2
    date: 2008/08/12 21:29:11;  author: markd;  state: Exp;  lines: +1 -1
    Tell libtool that its linking fortran so the correct libraries get
    linked in.

So I don't think that we can simply revert the change.

>How-To-Repeat:

>Fix: