Blender Git Loki

Merge branch 'master' into depsgraph_refactor

Depsgraph: Implement updateDepsgraph for the new normals modifier

Why it's called target when it's a source..

Merge branch 'master' into depsgraph_refactor

Depsgraph: Traversal for layer flush should be aware of cyclic dependencies

Quite straoghtforward, should have think of it earlier.

Also made transitive relations disabled by default now. They're killing some
relations which keeps thread safety.

So now koro is kinda finally working. There still some issues caused by
dependency cycles which are solved differently from master. Plus threaded
nature of armatures updates also affects in this case..

Merge branch 'master' into depsgraph_refactor

Depsgraph: Implement russian roulette cycles solver

The intention is to mark some relations from cycles as cyclic so
graph evaluation will ignore them when calculating number of pending
relations which are not satisfied yet.

This should unlock some rigs which are currently not possible to
evaluate at all because for cycle dependencies are never met.

In fact, with transitive relations reduction koro rig now is kinda
working. It's flickering but bones are not locked. What's weird is
that disabling transitive reduction makes some bones to not being
evaluated.

This is weird because it means either:

- Dependency cycle detector doesn't work 100% correct.
- Transitive reduction kills some extra relations which it shouldn't.
- We've got issue in graph traversal somewhere.

Depsgraph: Add simple relation cycle detector

For now it only reports if there are cycles in the graph and prints
nodes identifiers and relation names in the terminal.

TODO:
- Add cycles solver
- Improve reported operations names

Merge branch 'master' into depsgraph_refactor

Conflicts:
source/blender/blenkernel/intern/armature.c
source/blender/modifiers/intern/MOD_hook.c

Depsgraph: Fix previous workaround for driving bone by itself

- No relations needs to be added in this case at all
- COver RNA paths now as well

Depsgrpah: Adding/removing drivers should tag relations for update

TODO: There are more places to ensure proper relations, plus animation is
also to be checked.

Depsgraph: Use DEBUG_PRINTF for IK chain debug prints

Depsgraph: Avoid dependency cycle when driver is applied n the bone and uses same bone as target

This is more like a workaround and need much deeper re-consideration to support IK chains and so.

At least now we've got this noted in the code.

Depsgraph: Correction for depsgraph rebuild

Make sure tags are not getting lost after graph rebuild.

Depsgraph: Ported over ability to use vertex-group targets for IK constraints

Vertex group targets are supported by the constraint system, but were not
accounted for by the new depsgraph building code. This adds that support.

NOTE: Ideally, this wouldn't require a separate set of logic, but we need
these links to be attached to the solver and not the constraint, so
things have to be done this way.

Depsgraph: Fix for deadlock when IK targets depend on a bone in the chain

This hack tags the target as being part of the IK chain, so that when adding the
constraints, the "ready" state will be used instead of "done" to avoid the deadlock.
There don't seem to be any readily observed ill-effects of this patch, so I guess
it's fine.

Depsgraph: Temporary debugging code to check that the IK chains are getting built correctly

Depsgraph: Fix for superfluous errors arising from trying to add animation link when no animation exists

This is for the animdata link used for rigs. The one for plain object data worked
already.

Depsgraph: Improved add_relation() errors to report what the non-failing link was

Depsgraph: Debug value 799 can now be used to disable the transitive reduction

Originally, I was going to have it off by default, but since it seems to keep
many rigs working, it looks like it's best to keep it this way for now.

Depsgraph: Tried out a few other OGDF methods - These still have various issues