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.
Python operator subclasses and operator types each get their own SRNA, causing double ups for bpy.types.__dir__()
From the operator type - these share names.
* ot->ext.srna
* ot->srna
Note that this conflict is still there, this only disables 'ot->ext.srna' from being included in dir(bpy.types).
It was possible to allocate an array of size<2 which would then raise an error on vector creation without freeing.
Fix to ensure the behaviour of Vector.Range was the same as for builtin range() function. When specifying 3 arguments, the step argument wasn't being used to correctly calculate the vector size.
Minor formatting edits for error messages.
This issue is caused by texelFetch which does not take wrapping options into account as it uses unnormalized uv coordinates to access a texel. A hack was made to do manual wrapping in the shader. This is acceptable because texture wrap mode is always treated as "repeat" elsewhere in 3D view in blender.
Thanks to Morten Mikkelsen for pointing out the cause of the issue and providing the fix.
This was a request by Daniel Salazar. It adds a new frame offset variable to fluid sim settings, which can be used to display baked fluid sims at different times. Eventually this could be replaced by real NLA strips for cached data, but until then this is a simple way to have more flexible cache result usage.
Not strictly a BCon3 patch, but after IRC discussion with Genscher, ZanQdo and kaito decided to commit it anyway, since it's a small feature and makes fluid sim a bit more usable. Similar patch for point cache (particles, smoke, cloth) is being worked on as well.
Crash was caused by attempting to create strip text from a missed sound datablock.
Actual error happened when this sound datablock was missed, but crash isn't nice
anyway and there were no way to repair scene.
1) The width of columns was incorrectly determined on windows, fixed by increasing the size of the dummy buf.
2) Added additional brackets to string for consistent formatting
the curve modifier was applying the offet between the mesh and the curve object twice.
for positive axis it was (offset+offser), for neg axis (offset-offset) --- giving no offset,
in both cases not very useful.
and projection matrix, so operators which depends on such things can easily save
settings in operator properties in invoke and then reuse them in exec callback.
This will fix: #24767: Knife tool last operations panel doesn't cause changes even though F6 pop-up does.
#27129: Problem with knife cuts/midpoint type in quad view
Usage is pretty simple:
- From operator template declaration function call ED_view3d_operator_properties_viewmat()
to register all needed properties in operator.
- From invoke callback call ED_view3d_operator_properties_viewmat_set to
store all needed settings in operator properties().
- To access this settings from exec callback, use function
ED_view3d_operator_properties_viewmat_get().
Additional change:
added function apply_project_float() which does the same as
project_float() but accepts actual values for viewport width/height and
projection matrix.
It was incorrect behavior of contarget_get_mesh_mat in cases when
object's Y axis is co-linear to average vertex group normal.
Use object's X axis for plane definition in such cases.
would indirectly clear it when clearing the window from the context. This makes
some sense when we support multiple scenes properly, but currently there's still
many places assuming there is a single active scene, so keep it available to
avoid crashes.
