Blender Git Commits

svn merge ^/trunk/blender -r48153:48158

option to disable feather, since its so slow - for interactively editing masks its useful to be able to disable.
also rename RNA to 'use_antialiasing'

Fixes for 8vytes scalar types used in DNA (int64 and double)

- makesdna wasn't checking whether such scalars are aligned to 8 bytes.
Now it should be handled correct.
- Some scalars in Object structure weren't 8 bytes aligned, which lead
to some incorrectly loaded files.

Fixed by adding void *pad. It's a bit tricky part of patch, but can't
see clearer way to make alignment correct

Usually ints/chars were used for padding, but in this case there are
some leading pointer properties before int64 properties and using
pointer as a padding resolves alignment on both 32 and 64 bit
platforms.

Using pointers as padding weren't needed before, because all types
were correctly aligned independent of whether pointers are 4 or 8
bytes.

This fixes #31774: Empty offset Y parameter is resetting

update libopenjpeg from 1.3 to 1.5, since 1.3 has a bug saving alpha channels.

the patches/ dir is NOT applied but perhaps we can get OSX and BSD working using the opj_config.h

Fixes [#31433] BMesh: Knife tool Angle Constraint function

The problem was that calculating the angle was reliant on previous mouse coordinates, which did not update to match snapped-to vertices and edges.

This solves the issue by updating mouse coordinates in knife_find_closest_edge() and knife_find_closest_vertex(). Additionally, when angle constraints are enabled, edge/vertex snapping is now prevented.

svn merge ^/trunk/blender -r48147:48153

* Fix for the Cycles XML commit yesterday, blur_glossy has a different internal variable name.

falloff options for dilate/erode feather compo node.

Fix [#31714] Pinning object context causes texture list to disappear

Problem was py code of main texture panel was not doing any check on the pinned id, assuming it managed the textures itself - but this is not the case of the Object datablock...

All work actually done by Sergey, was just missing the Lamp specific case. Checked both in code and with tests, quite sure all cases are now correctly handled!

Fix for color ramp RNA paths in node trees. The path generation for color ramps in nodes was incomplete (not prepending the ID-to-node path), which prevented keyframing color ramp elements. Path lookup for color ramps is still brute-force and slow, but this is a general design problem with nested RNA structs.

Fix for properly converting 2.4x IPO Actuators to 2.6x Action Actuators. Previously the converted Action Actuators would not have an action assigned. This fix is based on code provided by Maxim Aleynikov in his report: [#30410] not full conversion IPO Actuator in Action Actuator.

svn merge ^/trunk/blender -r48140:48147

Fix for [#31396] "bge.logic.LibLoad fails to import text blocks" reported by Leonard Ritter.

Blender's import function check's the Text datablocks in main for additional modules for importing. However, libloaded scenes were 1) not loading Text datablocks and 2) not letting bpy know about them. Text datablocks are now loaded if a Scene is loaded and bpy can now looking through extra Mains to find additional modules.

Fix for [#31792] "Character Physics type not detected by near sensor" reported by Mikko-Pentti Eronen.

Near sensors only pick up "actors," but objects with character physics did not have the actor option displayed. By setting the character physics object to actor, it can be picked up by the near sensor. However, it collides with the near sensor, which sounds like bug [#31701]

workarond for outliner group object selection trick - which is so slow it had to be disabled. Now use the first object of the group, which isn't great but is at least usable.

correction to own commit - thanks Anthony Edlin for pointing out the error.

Cycles XML:
* Add recent integrator options.

svn merge ^/trunk/blender -r48133:48140

fix for segfault loading a file with a proxy that is a missing link.

Refactoring of tiles opencl implementation:
- Moved methods from NodeOperation to OpenCLDevice
- Added check on Nvidia for local size