Blender Git Loki

Merge branch 'master' into depsgraph_refactor

Merge branch 'master' into depsgraph_refactor

Merge branch 'master' into depsgraph_refactor

Conflicts:
CMakeLists.txt

Depsgraph: Temp solution for bone parent

Now the secret agent walk cycle works fine, and quite enough of the
knowledge is gathered now.

Time to start going granular now!

Subgraph: Workaround for dependency from self in the constraint stack

Once the granularity in the bones is reached we can get rid of this.

Depsgraph: Cleanup of animation callbacks

- Use single callback for all action fcurves
- Driver evaluation was actually evaluating the whole animation system,
not just a given fcurve.

Depsgraph: Workaround for bone constraint targets

Depsgraph: Corrections for Spline IK solver

The issue was caused by the missing relation between Spline IK solver node
and spline itself. This happened because builder used bone node type for
the IK solver, but relation builder was expecting pose eval node type.

Also added a dependency between IK solver node and flush nodes, so flush
happens after the solver. Well, it's a bit more tricky internally, because
we still using uber object update nodes, but adding that relation makes it
so DAG update flush all happens nicely.

Depsgraph: Updates for the cleaned UI API

Merge remote-tracking branch 'origin/master' into depsgraph_refactor

Depsgraph: Cleanup of time source relation a bit

This way it is now possible to add a relation between TimeSourceKey and
OperationKey.

Some areas were trying to add such a relation already, but it just didn't
work.

Not totally happy with the implementation, but think the patch is good
enough for the first cleanup iteration.

Merge branch 'master' into depsgraph_refactor

Depsgraph: Initial move towards animation support

It kinda works now, but gives issues with threaded update.
Plus it' really need to be cleaned up. Committing mainly to
have a starting point which does something.

Depsgraph: Bunch of node relation fixes, some other changes as well

This commit mainly ensures that relations between operations are more or less
correct (well, they're correct for the tests files i've got here), Namely:

- Added component relation between transform and geometry components.
This is a temporary workaround just to play around a bit, because it solves
some missing updates but also introduces unwanted updates.

For example, such relation solves missing updates of boolean modifier, but
introduces extra data updates when dragging default cube.

Think proper way to deal with this would be to make it so in the example above
boolean modifier adds a relation between modifier node and transform component.

- Added an uber object transform update, which is the last in the transform
component.

This way we can have proper object transform update without introducing storage
of the intermediate matricies for now.

This is a temporary solution for until we've got proper granularity.

- Made it so firts modifier depends on the object geometry.

- Made it so local constraints depends on local transform operation.

- Seems pose flush should depend on all the pose channels, otherwise what it is
gonna to flush?

- Fixed target key of constraint targeted to a bone.

This commit also gets rid of few temp hacks, and now the have working Victor rig.

Merge remote-tracking branch 'origin/master' into depsgraph_refactor

Depsgraph: Implement updateDepsgraph for armature modifier

So now dependencies between object and armature via the armature
modifier are all correct.

Depsgraph: Fix compilation with scons and compositor enabled

Depsgraph: Add depsgraph update function to the hook modifier

This is mainly to test how this is expected to work and it seems
to be rather straightforward. Now new depsgraph can use hook :)

Depsgraph: Forgot this in previous commit

Depsgraph: Hook up constraints evaluation callback

Quite straightforward. There's one extra flag to the old object where_is
which helps preventing duplicated code needed for old dependency graph
system In the future we'll get rid of this argument and decouple all the
functions in a nice way.

One retarded issue is setting negative scale flag on the object, which
currently happens in both local transform and constraints update.

Bigger issue here is the original object matrix which should be used
as an input for the constraints stack. Currently constraints will operate
on the previous final object matrix when changing the dependencies of the
constraint list.

This is possible to solve by storing intermediate data in the graph
(or somewhere else(, but for until that's nicely implemented we'll probably
couple constraints back. Will investigate this further.