This is an incomplete list of things still to be implemented to make ocmesh usable.
Here the missing features
- Octree balancing
- Hexahedral mesh generation in the correct format
- Portable build-system (probably CMake)
Optionals/to be discussed:
- Better command line options parsing
- Better format for description of CSG scenes.
- Scale and rotate should ignore translation?
- Maybe it's worth to simply provide a python binding to the library
- Support for specifying the precision in both percentage (as is now) and absolute values, and allow it to be specified from the command line and/or from the CSG file.
- Support for units of measure in the CSG file, including radians/degrees for the rotate function
Of course the first thing to do is to measure real cases and profile. A few possible performance improvements seem already plausible.
- Investigate the possibility of parallelizing the subdivision.
- Implement the parallel version of the balancing procedure
- Implement optimized neighborhood finding (by starting the binary search from the common ancestor instead of the root).
- Separate the
voxelclass used to represent voxels in the compact octree from the class used to manipulate single voxels offline (for example during the intersection testing), to avoid repeating a lot of bitwise operations. - Measure if it's worth to cache the distance results during the intersection testing