Re: [Bug-gsl] gsl-1.8: "make check" fails in eigen with gcc 4.1.0

[Brian Gough]

Patrick B. Greene writes:
 > I've just compiled gsl-1.8 on my pentium 4 laptop and "make check" fails
 > for "eigen".
 > 
 > I'm running Fedora Core 5 with the default gcc-4.1.0  I did use some
 > optimization flags to compile: CFLAGS="-O3 -march pentium4"
 > Once I get a little sleep I'll try recompiling without the flags.
 > 
 > I've pasted my "log" file below.
 > 
 > make[2]: Entering directory `/tmp/gsl-1.8/eigen'
 > FAIL: herm(10), direct eigenvalue(7), unsorted (-0.0545196117924558465
 > observed vs -0.0545196117924556592 expected) [1744]

Thanks for the bug report.  

The tolerance of that particular test was overly strict and I had not
tested it with those options -- I've increased the tolerance to allow
for the expected level of rounding error (patch below).

I can confirm that it is safe to use the library.

-- 
best regards,

Brian Gough

Network Theory Ltd,
Publishing the GSL Manual - http://www.network-theory.co.uk/gsl/manual/

Index: test.c
===================================================================
RCS file: /home/gsl-cvs/gsl/eigen/test.c,v
retrieving revision 1.23
diff -u -r1.23 test.c
--- test.c      22 Aug 2005 13:50:47 -0000      1.23
+++ test.c      18 Apr 2006 19:42:07 -0000
@@ -145,7 +145,7 @@
     {
       double ei = gsl_vector_get (eval, i);
       double e2i = gsl_vector_get (eval2, i);
-      gsl_test_rel(ei, e2i, 10*GSL_DBL_EPSILON, 
+      gsl_test_rel(ei, e2i, 10*N*GSL_DBL_EPSILON, 
                    "%s, direct eigenvalue(%d), %s",
                    desc, i, desc2);
     }