Support other point types

[nzjrs]

Possible options:

1. cython and fused types

2. built one backend (.so file) per type and hide behind an import hook

import pcl.pointxyz as pcl

http://xion.org.pl/2012/05/06/hacking-python-imports/

[colincsl]

I saw this new addition to cython that allows for C++ templates. I'm not sure of the details, but the first thing I thought of was that it could potentially help python-pcl.

https://github.com/cython/cython/blob/master/CHANGES.rst#020-2014-01-18

[monkeybutter]

I've seen this issue has been opened for more than one year now. I don't know if any progress has been made during this time or there are other options now to be considered apart from the two proposed.

I can work supporting other point types as I need this feature for my project, but any help or clarification about where to start would be really appreciated, as I don't have much experience with cython.

[colincsl]

I don't know of any progress that has been made.

While I can easily code stuff up with Cython, unfortunately I don't know how to go about wrapping a large library with it.

[monkeybutter]

I've created this stripped down branch, only statistical filtering, making use of PointXYZL points, if anyone is interested. We just need this functionality out from pcl so this is enough for us, but it would be really useful if the main project would provide support for other pcl point types:

https://github.com/monkeybutter/python-pcl

[nzjrs]

@monkeybutter Thanks for your branch - I agree that this functionality would be good to have, and it is a shame that this bug is not resolved yet.

However, as you appreciate, going from a proof of concept to a real implementation takes time, and while I don't need this functionality at work I have no need to work on it. Perhaps you could extend your branch with a proof of concept that supports both types?

The difficulty is not in creating or instantiating new point types, the difficulty is how best to dispatch (from python) to the specialized C++ implementations.

I outlined a few approaches in the first comment. I'm not sure of the best one yet.