[Bf-committers] Numpy integration

[mgschwan at blensor.org]

Hi,

I was talking to Campbell about integrating this patch

http://projects.blender.org/tracker/?func=detail&atid=127&aid=30908&group_id=9

which would expose several matrix solvers from the Eigen3 library
(included with blender) to python.


The question came up if it would be better to integrate numpy istead.

First of all why would this be good/needed:

I can't speak for everyone but having the least squares solvers 
available within
Blenders python interface would be quite nice for plane fitting, camera 
calibration, ...,
and in my case the mesh alignment 
(http://www.youtube.com/watch?v=LvF3ydZTzvM and
http://www.blensor.org/misc/media/videos/align_vertex.mkv)

Campbell asked me to check on how much this would add to blender,
and thats where it gets bad:

Even if all testing modules are removed from numpy it still clocks in
at ~12MB (bzip2 compressed ~3.7MB) which is not what one would call 
lightweight :)

So I ask everyone that needs it to comment on one of the three possible 
solutions:

1. Include a stripped numpy version, which will add ~3.7MB to the 
Blender download
2. Use the above mentioned patch to expose some of the Eigen3 
functionality that is
    shipped with Blender anyway
3. Don't care, don't need

I also would call to those devs who need numpy for other reasons, since 
it is also
good for performance hungry computations.

Best regards
Michael