Re: Octave 3.4.3-rc0 release candidate available for ftp

[Michael D Godfrey]

On 10/01/2011 04:24 PM, Jussi Lehtola wrote:
> On Sat, 01 Oct 2011 15:59:45 -0700
> Michael D Godfrey<address@hidden>  wrote:
> > >  1. System that fails is:               model name    : Dual Core AMD
> > >  Opteron(tm) Processor 275
> > >  2. System that passes is:        model name      : Intel(R) Core(TM)
> > >  i7 CPU         860  @ 2.80GHz
> > >  
> > >  But, the system which fails passed until recently.  And, of course,
> > >  there may be other
> > >  software differences despite the fact that both systems are
> > >  up-to-date fc15.
> > >  
> > >  Since the one make check failure is within a very small margin of
> > >  eps, I think this should
> > >  not be considered to be more that a curious anomaly.
> Could the svds tolerance be increased as well, so that my build would
> complete without errors?
>
> I ran the build on an Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz.
> -- Jussi Lehtola Fedora Project Contributor 
> address@hidden
This is very weird.  As far as I know, this test has not failed in a 
very long time.
My failures are all just like:
 450: maximum absolute error 2.22478e-16 exceeds tolerance 2.22045e-16
which is:
octave:1> 2.22478e-16 /eps
ans =  1.0020
============================================
So, the difference is trivial.  It could not do any real harm if the 
tolerance were
changed to 1.1*eps, for example.  But, it would be best if the cause of 
this failure
could somehow be discovered.

The fact the you (Jussi) see it on an Intel processor, just about 
eliminates the
possibility of some (very unlikely) chance that it is CPU-dependent.

To add to the confusion, I just did a new rebuild from hg on the second 
fc15 system that was
failing.  Nothing else changed, but now it passes.  This is an Intel(R) 
Atom(TM) CPU D525   @ 1.80GHz

There is a chance that there have been changes in gnulib, so I will go 
back to the last failing
machine and start over.


Michael