Particle system code used global variable to sort hair by orig index,
which is not safe for threading at all.
Replaced this with usage of reentrant version of qsort, which is
now implemented in BLI. It was moved from recast navigation code
to BLI, so more areas could use it (if needed).
This replaces code (pseudo-code):
spin_lock();
update_child_dag_nodes();
schedule_new_nodes();
spin_unlock();
with:
update_child_dag_nodes_with_atomic_ops();
schedule_new_nodes();
The reason for this is that scheduling new nodes implies
mutex lock, and having spin around it is a bad idea.
Alternatives could have been to use spinlock around
child nodes update only, but that would either imply having
either per-node spin-lock or using array to put nodes
ready for update to an array.
Didn't like an alternatives, using atomic operations makes
code much easier to follow, keeps data-flow on cpu nice.
Same atomic ops might be used in other performance-critical
areas later.
Using atomic ops implementation from jemalloc project.
Some places like proxy rebuild didn't increent
custom shape user counter which lead to user
decrement errors later when freeing pose channels.
Try to keep custom object counter relevent, but
some corner cases might still be missing.
Issue was caused by reload caused by input colorspace change.
For generated images generated alpha flags weren't saved to
DNA, which lead to fallback from 32 bit depth to 24 when
doing any kind reload of generated image.
The same alpha loss happens when you save .blend file with
generated images.
Now added generated depth to DNA, so reload image and .blend
file wouldn't loss alpha.
They were caused by not having a free_windowmanager_cb set and by not having registered SpaceTypes, which meant data allocated for thosse SpaceTypes could not be freed. These were solved by defining a free_windowmanager_cb for the player that just frees wmWindows, and by making sure we only allocate memory for registered SpaceTypes.
after objects are deleted until another big object is added. There's no good reason
to do this, or to think that our pool is somehow much faster than using the OpenGL
API to allocate and free buffers.
entered explicitly, bypassing the ui limit.
NOTE: this only applies to newly added nodes. For legacy reasons the limit values are stored directly in each socket, so changing them for existing nodes would require a do_versions, seems unnecessary
for such a minor change.
when the group contained two objects duplicating the same group. Also added the
dependency check to the "add to active group" operator now for consistency.
Made first/last tessellated curve have proper
direction and tilt. Before direction/tilt
from second/previous to last tessellated curve
segments.
Thanks Brecht for review and tests!
decrement the object user count when removing the armature. This could cause
the object to stick when it shouldn't, in particular when that object is part
of a group.
of the scene user count. The node->id pointer for these nodes is now initialized in the respective init callbacks already. The explicit assignment and increment in the ED_node_composit_default is not
necessary and just adds an increment without checking previous values and decrementing them properly.
Note that the current system still leaves the scene with "fake" users (rlayer + composite nodes) which are actually part of the scene data itself. But that's design issue with the "local" node tree data
and doesn't do any real harm.
