[Bf-committers] Blender roadmap article on code blog

CoDEmanX codemanx at gmx.de
Tue Jul 2 19:45:58 CEST 2013


Am 29.06.2013 16:41, schrieb Ton Roosendaal:
> - Where Python is too slow (I/O), we can also improve the api a lot still. For our UI now it's more than fast enough.

There are two areas where it's notably slow: user preferences input and 
addon UI - due to the high number of layout elements I guess. However, 
it's acceptable in this area.

Where python performance really bugs me is I/O. Due to GIL, python can't 
make use of multi-core systems (runs with max. 25% of an i5 with two 
real cores / four virtual cores). And multiprocessing isn't really 
applicable, since blender doesn't allow multiple threads accessing the 
RNA system without crashes.

Looking at the OBJ importer, the real bottleneck is the mesh splitting 
code. It takes a serious amount of time for gigabyte-sized OBJs, and a 
huge amount of memory (like 500 MB OBJ, 6 GB mem). Not sure if one could 
optimize the python code, but a C/C++ importer would always be superior 
(see Meshlab speed!). Any plans on merging assimp support from Bratwurst 
into trunk?


More information about the Bf-committers mailing list