Polyfit crashes octave 3.0.0

[Ben Abbott]

On Apr 10, 2008, at 3:19 AM, Jaroslaw Pychowski wrote:
> --------
> Bug report for Octave 3.0.0 configured for i486-pc-linux-gnu
>
> Description:
> -----------
>
>  Function polyfit causes segmentation fault.
>  51 is minimal affected value of n (see code below)
>  Messages emitted before segmentation fault:
>  octave:5> wn=polyfit(x,y,n);
>  warning: matrix singular to machine precision, rcond = 1.83429e-25
>  warning: attempting to find minimum norm solution
>  panic: Segmentation fault -- stopping myself...
>  attempting to save variables to `octave-core'...
>  save to `octave-core' complete
>  Segmentation fault
>
> Repeat-By:
> ---------
> n=51
> step=2/n;
> x=-1:step:1;
> y=abs(5*x);
> wn=polyfit(x,y,n);

I'm replying to report the results on Mac OSX.

On Mac OSX 10.5.2 with gfortran 4.3 and Apple's vecLib the results for  
each of the mercurial branches are below.

branch: release-3-0-x

octave:6> n=51;
octave:7> step=2/n;
octave:8> x=-1:step:1;
octave:9> y=abs(5*x);
octave:10> wn=polyfit(x,y,n);
warning: matrix singular to machine precision, rcond = 1.83428e-25
warning: attempting to find minimum norm solution
warning: dgelsd: rank deficient 52x52 matrix, rank = 42

branch: default

octave:5> n=51;
octave:6> step=2/n;
octave:7> x=-1:step:1;
octave:8> y=abs(5*x);
octave:9> wn=polyfit(x,y,n);
octave:10>

The default branch uses QR decomposition. The 3-0-x release uses left  
divide, but neither gives a seg-fault for me.

Ben