Re: Octave-maintainers Digest, Vol 79, Issue 14

[c.]

On 10 Oct 2012, at 19:00, address@hidden wrote:

> Message: 2
> Date: Wed, 10 Oct 2012 09:52:54 -0700
> From: Ed Meyer <address@hidden>
> To: "c." <address@hidden>
> Cc: address@hidden,   octave maintainers mailing list
>       <address@hidden>
> Subject: Re: [OctDev] Test failures due to tolerance in fftfilt.m
> Message-ID:
>       <address@hidden>
> Content-Type: text/plain; charset="iso-8859-1"
> 
> On Wed, Oct 10, 2012 at 1:11 AM, c. <address@hidden> wrote:
> 
>> 
>> On 10 Oct 2012, at 09:16, Daniel J Sebald wrote:
>> 
>>>> I propose fixing this test by replacing the tolerance eps with something
>>>> like 2*eps*norm(z)
>> 
>> FYI this could be expressed as
>> 
>> 2 * eps (z)
>> 
>> from the help text for eps () :
>> 
>> "Given a single argument X, return the distance between X and the next
>> largest value"
>> 
>> c.
> 
> 
> Thanks, Carlos, I wasn't aware of this capability.  I thought it was just
> what I needed until I
> tried it on a vector, expecting something like eps(z) = eps*norm(z) but
> what I get is eps(z(1)).
> Is that the intended behavior?

I don't think so, it is not compatible with Matlab:

>> z = [1 2 3]
z =
     1     2     3
>> eps (z)
ans =
   1.0e-15 *
    0.2220    0.4441    0.4441
>> 


I think this is a bug, could you please post a report?
It should be quite easy to fix, but I am in a hurry now, I'll look at it 
tomorrow if no one does before ...

> -- 
> Ed Meyer
c.