On Fri, Mar 27, 2009 at 4:31 AM, Patrick Mihelich <patrick.mihelich@gmail.com> wrote:
I hate to rain on your enthusiasm, but I agree with Luke that your high-performance low-dimensional linear algebra library is not very compelling. You will only be reinventing the wheel. This problem is already solved beautifully by Eigen2, which you should take a second look at - unlike uBLAS, it has first-class support for fixed-size vectors/matrices and does very clever SIMD optimizations. There is also Sony's Vector Math library in Bullet, I believe. These have liberal licenses - not so liberal as Boost, but good enough.
It's not much reinventing the wheel -- a 3d vector/matrix/quaternion library without optimizations can be written in under a week, the main work of the proposal would be data structures and algorithms that would be built over it.
A selection of 3-dimensional geometry algorithms could be interesting, though. That would already be meaty enough for a GSOC project. My suggestion would be to work with Barend & Bruno and simply write your algorithms using their primitives, Point etc. (...) and I think you will have much more of an impact integrating with an existing geometry library than writing your own from scratch.
There are some problems with naming conventions ( as to the standard used in CG3D ), and also the fact that only the point class (and maybe segment) would be of any use to me. However this is compelling, as I would be working on the things I planned from almost scratch anyway, but from the outside it would look like expanding something existing ;). -- regards, Kornel Kisielewicz