On Jul 23, 9:36am, Paul Beardsley wrote:
> 1. What is the most extensively implemented library for
> matrix and vector classes at the moment, which uses the STL
> generic approach?
>
> 2. Is there any sign of a library which is likely
> to be supported and developed for a significant length
> of time e.g. something which is being considered for
> the ANSI C++ standard?
Those are interesting questions. At face value, they would appear
to pose a concept that would be of interest to all of us. The idea
of the biggest, most extensive, most generic library would have
some merit. A library so all encompassing as to become the standard
for all work.
My experience is that such a beast is not really a good idea.
It can never be efficient for every practitioner.
I am not saying a standard library is a bad idea. I would think the
best, most extensive library for regression analysis would be useful.
The people who work in that area may have found a common implementation
that uses the STL generic approach that is efficient for them. But I can
guarantee it would not be the best approach for people doing computer graphics.
I do not believe there is a generic library that is the most efficient
implementation across all applications that use matrices and vectors.
Thus a "most extensively implemented" library sounds as though the
library became a goal in itself, and the efficiency requirements of
the application got lost.
As an example, consider the infamous matrix inverse so overworked
in the last three decades. Say that your application creates
matrices of a definite small size that are never stiff. (Rotations
for example.) Now say you get a generic library that works for
large matrices that are stiff (ie a library built for regression
analysis.) The algorithm would be very inefficient for inverting
a rotation matrix. Granted you could put in different inversion
methods, but in practice people would not know which one lurks
behind an operator. And who needs all of the extra baggage
when you never use it? (There are ample examples of superfluous
functionality with all of the inherent inefficiency in
the old Fortran libraries.)
In short, more appropriate questions would query what matrix and
vector libraries are available for application area X. Which one is most
commonly used and meets the requirements of application area X.
What is the reliability of that library and what future support
will be available? I would like to know the answer to those questions.
PS: Maybe we could call these minimally implemented libraries. Just
the functionality you need, tuned specifically for your needs,
and nothing else. There is certainly a cost benefit to creating
and maintaining minimal libraries if your goal is the application
and not the creation of a library. The source of contention is
of course defining the minimal set for all of your needs.
-- Richard Weidner
This archive was generated by hypermail 2b29 : Wed Feb 20 2002 - 03:20:05 EST