Re: [Bug-libmatheval] derivatives vs new functions

[Aleksandar B. Samardzic]

On Fri, Mar 25, 2005 at 01:57:32PM +0100, Leander Seige wrote:
> Hi,
> 
> > Can I suggest, instead of removing this feature, to implement a sort
> > of "NAN" function that plays the role of the derivative of any non
> > differentiable function?
> 
> Yes, of course, that is another option, I forgot to mention it.

OK, here is my point of view:

Libmatheval is created for my very specific needs at the very specific
moment of time and thus it happened to support a bit of numeric and a
bit of symbolic calculations.  Libmatheval is not (yet) widely used
package, but from the feedback I received seems to me that other people
are much more using numeric part of functionality than symbolic part.
Also, feature requests are exclusively regarding adding new functions
and new operators and often even diverging from pure numerical
calculations towards extending libmatheval to support everything allowed
in C-like expressions.  Thus, symbolic part of the library is becoming
baggage, because some of these proposed features have no symbolic
counterparts at all and on the other side calculating expressions could
be implemented more efficiently without syntax tree, that is in turn
necessary for symbolic calculations (so adding kind of "not-implemented"
marker for functions without implemented derivatives would be short-term
solution only).  Thus I'm really inclined to start afresh to code 2.0
version along these lines, with derivatives functionality (and Fortran
77 interface) dropped.  However, at the moment I'm on hard pressure on
my daily job and I don't think I'll be able to start coding anything
before June.  But in the meantime, I'd really like to have some feedback
regarding above.  Little problem is that, as far as I can see, many
libmatheval users are not on this mailing list, but this list is
official place for discussing this kind of stuff so this is probably
best we can do.

In order to support minority of users still needing derivatives, I'm
thinking about spending some time before approaching forking new version
to make possible to add new functions (and eventually formulas for their
derivatives) as a kind of plug-in system in 1.x version.  I was already
explaining the procedure of adding new function to libmatheval to some
people (if there exist interest, I can post instructions here on the
list too), but I can further try to automate the procedure so that
everyone with basic C programming skills could add needed functions to
libmatheval without need to know either lex/yacc (that are used for
parsing) or libmatheval internals.  I'd like of course to discuss this
idea too.

Also, I am thinking about changing libmatheval so that it supports
arbitrary precision calculations, something like GMP.  Any feedback here
will be appreciated too.

Regards,
Alex