during a WM_draw. tiny is max 8* size of pointer and it is maintained at
that size. So no worries there.
* cleanup some code to be certain that deinitialization happens
correctly.
Two bugs: first, the point-in-face function had a bug that made it fail for the xz or yz planes.
Second, in ortho mode, simultaneous linehits need careful sorting.
Also, in orth mode it is somewhat random whether the edges exactly behind the front ones are reported 'hit' by bvh tree, so put in code to prevent cuts along the parallel-to-view faces when not in cut-through mode.
Issue was caused by getting image from compositor node conversion code,
now it'll check whether rendering happens and if so, frame wouldn't be
stored in the cache.
This possible fixes#32465: Memory leak when rendering
Checked by Brecht when were in Blender Institute.
Discovered when was looking into #32296: Node Texture - Node Material - GLSL Viewport rendering issue
* Location: Basically the same as the location from Object Info node for object instances on particles, but in principle there could be additional offsets for dupli objects, so included for completeness.
* Size: Single float scale of the particle. Also directly translates to object scale for current dupli objects, but handy to have as a single float to start with instead of a scale vector (currently not even exposed in Object Info).
* Rotation: This is a quaternion, which are not yet supported by Cycles nodes. The float4 is copied to internal Cycles data and stored in the particles texture data, but the node doesn't have a socket for it yet and the data is not yet written to the stack. Code is just commented out so could be enabled quickly if/when rotation support is added to cycles.
* Velocity: Linear velocity vector of particles.
* Angular Velocity: Angular velocity around principle axes.
The texture data is currently packed tightly into the particles texture, which saves a few bytes, but requires an additional texture lookup for some vector attributes which spread over two float4s. Could also add another float4 to particle size to avoid this.
bone renaming
* Renaming F-Curves now checks if the corresponding F-Curve's group can also be
renamed accordingly.
* Changed the RNA updates for bone renaming so that they properly update the
channel lists
WM_keymap_add_item(keymap, "TEXT_OT_insert", KM_TEXTINPUT, KM_ANY, KM_ANY, 0);
<--- this catch-all key map item at the end of the Text Editor keymap was
gobbling all the key events for user-defined hotkeys added after it in the
keymap. This includes all hotkeys for new operators defined by addons (via
keymap.keymap_items.new()).
As a slightly hacky workaround for this, I've added an extra parameter to
keymap_items.new() which will force the newly added item to get added to the
start of the key map items list (i.e. with top priority). To enable, simply add,
head=True to keymap_items.new() calls. For example:
keymap.keymap_items.new("MY_OP_my_operator_id", type='E', value='PRESS',
alt=True, head=True)
This should be useful for cases where there are similar catch-alls, where it is
necessary to insert our item before the offending catch-all (without knowing
which one it is). However, in general, it's recommended not to use this, unless
all other methods (inlcuding choosing another key combination if your first
choice doesn't work) fails.
Around Frame is enabled
It turns out that the final "rebuild pose" at the end of each ghost drawing step
was at fault here, as it resulted in bones referring to a different set of
constraints after each ghost drawing step. However, most of the time, these new
references pointed to non-existent locations, causing the segfaults.
To fix this, I've removed this last rebuild step, which seems to work fine (no
major problems seem to arise from this). Also, to ensure that the other object
parameters have not changed (as a result of the reevaluation that needs to go
on), we now evaluate the object once more after all the ghost drawing (but
before final restore) so that object parameters (and potentially sub-data) are
all restored to their original values.
An upside of removing the final rebuild step is that "In Range" and "On
Keyframe" ghosting options now work with proxies again. Previously, trying to
move bones (but without autokey enabled) would mean that bones would snap back
to their unkeyed positions.
