Additional freeof function // avoiding rat /disrep

There is something more thorough that could be
done, perhaps causing good efficiency changes in many
places, but would have to be done thoughtfully.  Something
like this is done in Mathematica for its simplification/
evaluation.

The basic idea is to add to any MRAT form its totaldisrep'd form
as part of its header.  To make this work, it would be
necessary to also tag it somehow to say that it was produced
at timestamp XXX  and that if any of its constituents have
been changed since XXX, the disrep form is perhaps not valid.
Also other things that may have changed that affect the
form might be some global flags that affect disrep
though none of them come to mind at the moment.

Then disrep would be essentially free since the result
would be "memoized".

It would also be possible to memoize the reverse transformation
making rat  or even ratsimp free after the first time.

This facility would make less sense if memory were
very scarce, as was the case when macsyma was written.
With 512megabytes of RAM it doesn't seem like a bad idea though.

RJF




Dan Stanger wrote:
> I was not aware of the existance of FREEOFL,
> and I apologize to the list for that oversight.
> When I first looked at freeof, I thought it could benefit
> from changing it so that it would not convert to general form.
> I then started to look at the rat*.lisp packages, and saw that
> there was so little code documentation, that I was afraid that
> whatever I wrote for for nfreeof, would be reinventing the
> wheel, so I decided to wait on it.  I also looked at the
> definition for totaldisrep, and saw it first checks if its input
> is a atom and in that case just returns.
> Also I noticed that functions which take a list, instead of
> one argument, usually start with L.
> So I then posted to the list, which you quote,
> that I changed my mind, and was going to call the
> new function LFREEOF.  That would have been a
> good time to point out the existance of FREEOFL.
> I then wrote LFREEOF, so that it checked for
> incorrect usage, made a minor optimization
> which was to only do totaldisrep on one element
> of the list at a time, added a entry in the texi file,
> and added some tests to the rtest file which checked
> freeof.
> I am willing to either change the name of the function
> to FREEOFL or modify existing usage of FREEOFL to LFREEOF.
> Since I needed the functionality of LFREEOF for charsets,
> I think it needs to go in one of the files in the src directory.
> After the rat*lisp files are better documented,
> I will go back to LFREEOF and modify it then.
> Does this sound reasonable?
> 
> Dan Stanger
> 
> Wolfgang Jenkner wrote:
> 
> 
>>Dan Stanger <dan.stanger@ieee.org> writes:
>>
>>
>>>In conforming to maxima/macsyma naming conventions,
>>>I have decided to call the new freeof function
>>>LFREEOF
>>>with the usage LFREEOF(list, expression).
>>>It is consistant with the name and usage of LRATSUBST.
>>
>>Your function has been part of Maxima since the beginning under the
>>name $FREEOFL (its definition was not extant, though) and it is used
>>in facexp.mac and some other files.  You may find it in its new old
>>home genut.mac.
>>
>>I am somewhat puzzled, though, since the discussion on the mailing
>>list gave me the impression that your foremost concern was to avoid
>>conversion to general representation form.  What you committed (to
>>nrat4.lisp) doesn't show this concern anymore and your implementation
>>seems to have the same degree of efficiency as what is already in
>>genut.mac (by the way, I might replace my definition there by
>>
>>freeofl(varl,exp):=apply('freeof,endcons(exp,varl))$
>>
>>or some such thing).  Why do you think it's a good idea to add such a
>>function to the lisp core of Maxima?
>>
>>Wolfgang
>>--
>>wjenkner@inode.at
>>
>>_______________________________________________
>>Maxima mailing list
>>Maxima@www.math.utexas.edu
>>http://www.math.utexas.edu/mailman/listinfo/maxima
> 
> 
> _______________________________________________
> Maxima mailing list
> Maxima@www.math.utexas.edu
> http://www.math.utexas.edu/mailman/listinfo/maxima