Am Sonntag, den 14.03.2010, 22:19 -0700 schrieb Richard Fateman:
> I suggest that before anyone makes "great" slower, even a little, that
> you find out how much time is taken by great.
> It is considerable. Substantial time and effort were spent on making
> great as fast as possible within the constraints
> of its specifications.
>
> Speeding up great and more generally speeding up sorting in the
> simplifier can make Maxima's simplifier run
> many times faster on medium to large expressions. Barton and I have
> examples and patches for this.
> Unfortunately, the speedup depends on some tricks that do not work as
> well on some systems as others
> (depending on hash table speed), and there are some glitches that have
> to be ironed out before
> inserting these fixes into Maxima.
>
> RJF
>
>
>
>
> Robert Dodier wrote:
> > On 3/14/10, Dieter Kaiser <drdieterkaiser at web.de> wrote:
> >
> >
> > > 4. Modify great to handle special representations.
> > >
> >
> > This is my preference. I don't know how it should be accomplished.
Hello Robert, hello Richard,
thank you very much for the feedback.
I have listed the different possibilities because at this time I do not
know what is the best to do.
I tend to the opinion of Robert, because an implementation to handle
special representations in great is the most general approach. This
solves the problem at the origin.
But I know too how often great is called. On the other hand I think it
is possible to implement the handling of special representations in a
way which does not change anything for most of the calls to great
(perhaps more than 90 % of the calls with the testsuite). This can be
tested carefully.
Furthermore, any improvement of functionality needs more code and slows
things down.
On the other hand it is much easier to replace the call of specrepcheck
with a call to $totaldisrep in $orderlessp and $ordergreatp. At this
place it is something like a workaround. Now we get more overhead for
these functions, but this might be more acceptable. Furthermore, we
could ignore the problem that a Poisson series still does not work in
general. If a user opens a bug report about this problem we could close
it as "won't fix".
One further option I have not listed is to do nothing and to close the
bug report as "won't fix".
Dieter Kaiser