[Bug-gsl] Strange problem with gsl_eigen_hermv

[Alexander Klein]

Hello,

I spent the morning chasing a strange bug:

I've written a wrapper class in Smalltalk, so I can use the GSL from 
Cincom VisualWorks.

Trying to compute the eigenvalues and -vectors of the 4*4 Hilbert 
matrix from the GSL-manual, the Smalltalk version always came up with a 
lot of NaNs instead of the correct result.

Checking with gdb, I found that the data passed to gsl_eigen_hermv was 
definitely correct, but strangely enough, a C program that passed 
exactly the same data ran without problems.

After some further experimentation, it turned out that the problem went 
away after recompiling GSL with

setenv CFLAGS -O0

I use gcc 3.4.0 on a Pentium-IV running Suse Linux 9.0
The GSL-version is 1.5

Unfortunately, a sample program in Smalltalk is not really what you 
would call a 'minimal example', so I won't send the whole class 
definition here, but if there's any other information I can provide to 
track the bug, I'll be glad to do so.

Thank you anyway for providing the GSL. It saves so much work!

Best regards,

        Alex Klein

--
Alexander Klein, Diplom-Mathematiker
Lehrstuhl für Numerische Mathematik
Heinrich-Buff-Ring 44
35392 Giessen

smime.p7s
Description: S/MIME cryptographic signature