[Pkg-scicomp-devel] Bug#569436: Bug#569436: lapack: zgesvd seems to give incorrect results

[Sylvestre Ledru]

Le samedi 13 février 2010 à 08:57 -0600, Kumar Appaiah a écrit :
> On Sat, Feb 13, 2010 at 03:20:19PM +0100, Sylvestre Ledru wrote:
> > severity 569436 normal
> > thanks
> > 
> > Hello,
> > 
> > I installed a chroot (through debootstrap) and I don't have the problem.
> 
> Did you try rebuilding libitpp on amd64? Did it FTBFS for you? 
I did not. I only checked lapack.

> > With the intel code, I get:
> >  Left singular vectors (stored columnwise)
> >  ( -0.86, -0.00) (  0.40, -0.00) (  0.32,  0.00)
> >  ( -0.35,  0.13) ( -0.24, -0.21) ( -0.63,  0.60)
> >  (  0.15,  0.32) (  0.61,  0.61) ( -0.36,  0.10)
> > 
> > The code is working fine with Octave or Scilab.
> 
> Thanks for checking. Did you use amd64? Did you notice that the FTBFS
> is on amd64?
No, I just tried with i386. I did not see any reference that it was a amd64-only bug.


> > Sylvestre
> > PS: I changed the severity to "normal".
> > FYI, grave means:
> > "makes the package in question unusable or mostly so, or causes data
> > loss, or introduces a security hole allowing access to the accounts of
> > users who use the package."
> 
> Thanks for being pedantic without providing any alternate suggestions,
You should relax. I didn't mean to be pedantic. I was just direct.
And I am not providing suggestions, it is because I don't have any for
now.

> but without SVD, I'd say most of the math related packages are
> unusable, since they give absurdly wrong results. But I won't play a
> severity ping-pong although at least my work and my package(s) are
> affected.
> 
> And, I just created a fresh amd64 chroot, and Octave gives me this:
You confirm that you only get this bug while you are in a chroot ?

[...]

> Some help as to how to debug this package would be appreciated, since,
> even according to your "definition" above, several packages are
> unusable for me, and the bug is "grave" for me. I would really
> appreciate some further pointers in finding where the issue is.
You should contact upstream mailing list here.

Sylvestre