Re: Same .m file: different results with different versions of Octave

[Jaroslav Hajek]

On Mon, Apr 19, 2010 at 9:38 AM, Francesco Potortì <address@hidden> wrote:
>>Numerical accuracy is simply just one of important factors. It's often
>>better to get an inaccurate result fast. I don't see why Octave should
>>aim for substantially better accuracy than libc.
>
> My expectations may be completely wrong, but I personally would expect a
> much higher attention towards accuracy from a specialised math program
> like Octave than I expect from a generic library like libc.
>
> In other words, I would be more confident in results obtained through
> Octave than those obtained through code relying exclusively on the libc.
>

... even though you know that Octave uses libc and libstdc++?

>
> I would also expect that, as one way of getting this higher math
> standard, Octave and any other piece of specialised math code would
> implement work arounds for known problems in standard libraries like the
> libc.  To make it even more clear, this is not specific of Octave, libc
> or even math software.  It is a general expectation of high quality
> specialised software versus generic implementations.
>

I suppose Octave is just not high quality specialized software, then.
Octave is just a collection of sources, it can be built against any C
library. Just like other components, such as BLAS or FFTW, we simply
rely on libc and its accuracy, because it's difficult to check for
everything. Nothing prevents you from employing a high quality version
of libc if you have one.
I've got also no problem with implementing a workaround, if it's done
conditionally after a check for the potential defect. But I think that
workarounds for standard functions should be conditional, like what
gnulib does.

If high quality specialized software means Octave will intentionally
not use a working, efficient library just because some other
implementations *may be* buggy, the please don't let your high quality
specialized software anywhere near me.

-- 
RNDr. Jaroslav Hajek, PhD
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz