reported as [#29376] BMESH_TODO: remove tessface CD_ORIGINDEX layer
for a single mesh there could be 3 origindex mappings stored, one on the polygons and 2 on the tessfaces.
(CD_POLYINDEX and CD_ORIGINDEX).
as Andrew suggests, now tessfaces (which are really a cache of polygons), using origindex to point to polygons on
the same derived mesh, and polygons only store the original index values.
In fact there were several issues fixed (all of them regressions since bmesh merge):
- Creating navmesh crashed because creating new faces for mesh was trying to set
default values for all customdata layers in this face. This requires memory
pool created for this datablock.
Usually this pool is creating on creating datablock if there're some elements
to be stored in this block. In cases of regular primitive creating it wasn't
an issue because they doesn't create customdata layers, they only creates
geometry.
Navigation mesh creates geometry and customdata layers (CD_RECAST layer)
which used to confuse a bit custom data functions. Solved by ensuring there's
memory pool created for polygons datablock after adding new custom data layer.
Most probably it's better to be resolved on CD level (like smarter track on
changed amount of stored data and so) but prefer not to make such global changes
so close to the release.
- Toggling edit mode lead to loosing recast datalayer. Solved by adding recast
layer to bmesh mask so it'll be copied to/from edit mesh.
- Some part of code assumed raycast layer is in face datablock, some that it's in
polygon datablock. Made it to be in polygons datablock.
Kind of temporary solution to make navmesh working, probably it'll fail if one
will want to edit navmesh by hand after it was generated.
Proper way would be to ensure the whole navmesh things are using ngons.
There were two issues discovered:
- Triangles mapping didn't free in buildNavMeshData if there's no recast data for an object
- KX_NavMeshObject used not-guarded allocation for polygons storage, but used guarded
freeing stuff to free used memory, producing error messages in the console and leading to
memory leak.
Wasn't actually harmful for users -- there was no memory corruptions and error happens only
when object was set up in a way when navmesh can't work in theory.
- Upgrade Recast library to latest portable version
- Implement recast_qsort based on FreeBSD qsort.c to have
portable thread safe quick sort for use in conversion routine.
- Better default value for the Build Navigation Mesh operator
- Some declarations after statement left.
- Do not use static inline functions in MOD_navmesh. It produces errors
with msvc and not sure it's actually helps -- optimizer should
make it inlined itself.
