Two issues here:
- Evaluated object data is to only be updated for selection only after modifier
stack is done its job. Otherwise it's possible to have selection batch update
called on an input data, at the same time as original object data is being
evaluated.
- If object's modifier stack did not create its own evaluated mesh (in case
when there is no effective modifiers, for example) can not update selection
on object's data, as it might cause threading issues between objects sharing
same data.
When we create the cage procedurally, we need to remove any edge split
modifiers. Since the new depsgraph in 2.80 we were removing the
modifiers straight from the evaluated object (it is a copy anyways).
On top of that we need to reset its eval data state (BKE_object_eval_reset)
to make sure the call to BKE_object_to_mesh to generate the cage would take the
new modifier stack state into account.
However doing so was freeing the low poly mesh we use later to convert
the normal space.
The solution (and this patch in fact ;) ) as suggested by Sergey Sharybin is to
use BKE_mesh_new_from_object() directly as well as force the modifiers to be
recalculated when any edge split modifier is removed.
The root cause seems to be an assumption in
[generate_video_frame()](https://developer.blender.org/diffusion/B/browse/master/source/blender/blenkernel/intern/writeffmpeg.c)
that we're always using 4 bytes per pixel. This is not true when using
QTRLE in RGB mode, because that uses the RGB24 pixel format (so 3 bytes
per pixel). Just updating the `linesize` property doesn't fix it though,
but just creates a crash somewhere else.
This at least fixes the crash by always forcing RGBA to be written, even
when the user selects RGB.
There are several aspects to the fix:
- Always calculate bounding box for meshes and curves from dependency
graph evaluation function.
There is a reason why mesh was tagged for geometry update, so can not
be spare here in attempts to avoid calculation.
- Copy bounding box and texture space to original object for active
dependency graph.
This matches object-level bounding box and allows to remove bounding
box evaluation from RNA.
As Usual...
Note that we have to edit each C-defined of those callbacks that set
Node->id pointer, instead of just moving the generic call to
`id_us_plus()` in `node_init()` below the call to
`ntype->initfunc_api()`, since that one manipulates RNA pointers, which
are supposed to already handle usercounts themselves on assignment.
The C callbacks are moving back to manipulating directly DNA, so it's
their responsability to handle user count then (Py code should do that
automatically through RNA assignement).
... instead of overiding it (previous behavior).
In practice it's not really noticeable.
This means an object with alpha will never be more opaque when enabling
xray.
The wireframe drawing for face masks is intrusive as selected wires
were solid white and always drawn. This made it hard for users to see
the exact color near edges.
This patch draws only the border of the selected faces,
edges between two selected faces are not drawn at all.
Reviewed By: brecht, fclem
Differential Revision: https://developer.blender.org/D5147
There are several aspects to the fix:
- Always calculate bounding box for meshes and curves from dependency
graph evaluation function.
There is a reason why mesh was tagged for geometry update, so can not
be spare here in attempts to avoid calculation.
- Remove texture space evaluation from RNA accessor.
Such data is to be evaluated by a dependency graph.
Don't see a reason to be different here: we never force evaluation of
any kind from RNA.
- Copy bounding box and texture space to original object for active
dependency graph.
This matches object-level bounding box and allows to remove bounding
box evaluation from RNA.
