Re: tolerance in binopdf.m

[Jordi Gutiérrez Hermoso]

On 20 September 2011 22:50, Ben Abbott <address@hidden> wrote:

> Looking at these individually ...
>
> printf ("%.17f\n", log (0.5))
> -0.69314718055994529
>  printf ("%.17f\n", gammaln (3))
> 0.69314718055994540
>
> Does anyone have any insight regarding this?
>
> What does Linux give for these two?

Well, I'm not sure that Linux has anything to do with it, but the GNU
C library I'm using seems to define lgamma, which is a C99 function,
and is ultimately the function my system calls for Octave's gammaln.

Can you grep your config.log for the HAVE_LGAMMA macro? Did it get
defined to 1? If not, this may explain it, as you would have a
different implementation of lgamma that could account for the
difference. From my reading of the code, if the native C function
isn't found, Octave will instead call the SLATEC dlgams Fortran
function from libcruft/slatec-fn/dlgams.f.

HTH,
- Jordi G. H.