[Maxima] Quality of share vs. src (WAS matrixexp problem (bug ?))
Wed Oct 25 14:12:57 CDT 2006
On 10/25/06, Barton Willis <willisb at unk.edu> wrote:
> the location of a package (src, share, or share/contrib) doesn't
> really reveal anything about its quality or utility. High-quality
> useful code is well, useful and high-quality regardless of which
> directory it is in.
There is certainly some mediocre code in src and some excellent code
in contrib, but I think it is useful to keep in mind the *intent* of
the distinction. Here's my take on it -- comments?
The core Maxima system (in src) is supposed to have consistent,
good-quality, and widely useful notations and functionality that are
actively maintained by us. Every part of the core system should
interoperate nicely with every other part. Functions should cover a
large part of their domain and give clean errors or noun forms when
used outside their domain. Notations should be unambiguous. All Maxima
programs can depend on the core functionality being available;
incompatible changes should be avoided.
The share directory contains additional functionality which may be
special-purpose or unmaintained. The core system does not depend on
the share directory. Different packages may use different and
non-standard notations or conventions, which are not necessarily
recommended by the Maxima project. Packages may not only not
interoperate correctly, but might even cause problems if loaded into
the same Maxima system. Functions may only cover special cases, and
may even give incorrect results silently if applied outside their
design domain. Implementation mechanisms may be much more inefficient
than core algorithms (e.g. using pattern matching for simplification).
The contrib directory contains packages which may be suitable for the
core system or the share directories, but which have not been
evaluated fully enough to put there.
By these definitions, there are several parts of the core system which
probably belong in share. One is poisson, which interoperates with
almost nothing else. Another is cf, which uses list notation to
represent continued fractions -- no other part of the system
understands this. Then there are cases where I think we agree that
the lack of interoperation is a bug we should fix (e.g. core matrix
inversion doesn't support floats or bfloats).
More information about the Maxima