Why the dependencies on matplotlib and scipy?

[giohappy]

Hi Olivier, thanks for your plugin. Giving a look to the code I wonder if we could avoid those dependencies, given that convex hull and delaunay are already available from other parts of QGIS. Did you use them for future developments?

[olivierdalang]

Hi ! Yes avoiding those would be really nice... Do you know how it's possible to access the already available algorithms from python ? At worst it's still possible to include the algorithms with the plugin...

[giohappy]

convexHull can be built from a QgsGeometry [1]. For Delaunay maybe the voronoy python implementation [2] in ftools could be ok (it's the porting of the Fortune's sweepline algorithm). Given that ftools is bundled (core plugin) it will always be there. The problem is that the tools/ folder inside fTools is not a module, so to obtain voronoy.py you should have to do something like:

import os,sys
import fTools
sys.path.append(os.path.abspath(os.path.dirname(fTools.__file__)+'/tools'))
import voronoy

[1] http://qgis.org/api/classQgsGeometry.html#a0cf0d95366ae2a1fb305a59742c35e12 [2] https://github.com/qgis/QGIS/blob/master/python/plugins/fTools/tools/voronoi.py

[giohappy]

PS: voronoy.py implements both the duals [1]

[1] https://github.com/qgis/QGIS/blob/master/python/plugins/fTools/tools/voronoi.py#L759

[olivierdalang]

Great thanks a lot ! I'll do this ASAP

[olivierdalang]

Ok scipy was easy to replace with QgsGeometry.convexHull(), but matplotlib's tri_api -(http://matplotlib.org/1.3.1/api/tri_api.html) has those nice "trifinders" which do all the job of finding in which triangle the points lies...

I'm no expert at python modules/dependencies etc, isn't there a way to ship only the needed .py files with the plugin ?