[Gcl-devel] R. Waters XP, multi-precision floating point

[Camm Maguire]

Greetings, and thanks!

"Paul F. Dietz" <address@hidden> writes:

> Camm Maguire wrote:
> > Greetings!  Currently, we just use the integer functions from gmp, so
> > the separation of the multi-precision floating point ('mpf') routines
> > into a separate mpfr lib will not affect us.  But this brought to mind
> > the question -- can common lisp make use of multi-precision floating
> > point?
> 
> There might be problems with the values of the *-epsilon and
> (most/least)-(positive/negative)-*-float constants, if long floats
> (say) were defined to have variable precision.
> 
> If you just mean: a large but constant precision, there shouldn't
> be any problem with that as long as the precision is at least as
> great as required by the standard for that particular float type.
> 
> It's also legal to add additional float types beyond the four
> standardized ones, and to make these be subtypes of float (disjoint
> from the standardized ones.)
> 

OK.  It would appear that this might be of interest only if the
precision was user settable.  I'm thinking about calculations in
maxima and similar programs where a result is desired with a given
number of decimal places.  Were this in lisp itself, it might make use
of gmp/mfpr and be quite efficient.  Can you conceive of a
user-specified constant precision that would fit in with the general
standard? 

We obviously have bigger fish to fry.

Separately, what is your opinion on the xp pretty printer?  Can we
save time by absorbing this?  My preference is to avoid replacing
functions already in place, usually in C, and well tested 'in the
field'.  Would this entail extensive modification to xp in your
opinion?  

In general, as we go forward on the ansi front, I'd like to merely
mimic the most compliant lisp functions (written in lisp) available in
the other free lisp systems.  Where I'd like to be unique is in the
compiler, the facilities for handling native code, interfaces to
external libs, etc.  While I feel comfortable enough rolling our own
new lisp functions, it just seems like a waste of effort given the
work others are doing.  For example, we can follow cmucl modifications
to pcl almost verbatim, noting of course that the cmucl license is
compatible with the LGPL.  So in your estimation, which subsystems of
which implementations would you like to see GCL follow in this manner,
i.e. which are doing the best job at compliance?

If I'm mistaken and you feel that we have some likelihood of writing
significantly better routines from scratch, please so state.

Take care,

>       Paul
> 
> 
> _______________________________________________
> Gcl-devel mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/gcl-devel
> 
> 
> 

-- 
Camm Maguire                                            address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah