[Help-gsl] Euclidean geometry package for gsl

[Dimitris Papavasiliou]

Hello again,

I'm sorry but I'm going to repost my previous message since it got a
little off topic and also because the choice of subject was probably a
little bad and it might not have caught the attention of the authors.
The might perfectly well have ignored me as well :).

=-----

as I've said in my first post I'm mostly into computer graphics so my
main interest in the gsl was not finding eigenvalues or solving
differential equations but plain and simple matrix and vector
operations.  Nevertheless solving differential equations are often
useful in cg for simulation for example so it would be even better to
have all of these available under a consistent framework.

The bad news, and the reason I'm bothering you with all this, is that,
as I had expected, the blas routines seem to be overkill for simple 3x3
matrix a 3vector operations.  Doing them straightforwardly turns out to
be much faster.

So the question is: do you think that a "package" specifically aimed at
three dimensional geometry (and projective geometry maybe) would fit
into the gsl.  The main operations I would need would be matrix-matrix
and matrix-vector ops., inversion (maybe), special matrices, eg. for
rotation, scaling etc. and maybe some convenience functions like vector
rotation etc.  The package could use the usual gsl vector and matrix
structures but would implement some functions specially to save time.
This way interoperability with other gsl packages would be guaranteed.

Altogether this is no big deal to implement but it would be quite useful
as I have not found any alternative library for C and even if there was
one, I'd probably end up using the gsl as well later on for its other
features.  Therefore I would be willing to implement it if it's
considered to fit into the overall gsl framework.

Let me know what you think,
Dimitris P.
=-----

Some more thoughts:
If having two different sets of function to operate on matrices and
vectors, one generic and the other specialized to three dimensions
doesn't sound good, then maybe the package could define "separate"
structures called say points (for vectors) and transforms or operators
(for matrices) and provide functions to convert these to normal matrices
and vectors.

Also useful calculations like distance from {point, line, etc.}  to
{plane, circle, etc.} would fit nicely into this although I can't
promise to implement these completely straight away.  Still others may
join the effort I suppose.

I'm sure other nice things like intersection tests could be done as well
but let's not get ahead of ourselves.

Waiting for comments,

Dimitris P.