Blender Git Loki

Depsgraph: WIP attempt to resolve some problems with IK chains

This commit tried to fix the problems which occur when a bone that's
part of an IK chain has a constraint to another bone in the same chain.
Since all constraints are (or should still) handled before the IK solver runs,
that ended up causing a deadlock, since the offending bone would rely on the
"final" (done_node) transform of another bone in the same IK chain, which could
only run after the IK solver runs - but that cannot run as it's waiting on this
constrained bone.

* Added a dedicated "BONE_READY" noop, which sits after the constraints op,
before the done node, and before any link to an IK solver. This replaces
all the calls to bone_final_transform() by having a dedicated node which
can be queried for this purpose

* Simplified the way that bone relationships get set up: The idea is that
all bones in the same armature will use the "READY" nodes instead of "DONE"
to avoid any lockups when dealing with IK. This way, we can do without all the
common-root checks before doing parent links and so forth, as well as just
setting out all the major relationships at the same time.

(There is of course the risk that this breaks some old rigs (* see note below))


The result of all these efforts is that all the simple test cases and also the
test file with the offending setup which inspired this fix all work now. However,
it also seems to now be causing a range of weird regressions with production rigs
(notably, strange new unstable popping in Koro and Victor Layout, as well as lag
issues on some simpler IK leg rigs).

Depsgraph: Use straight-lines for graphviz debug

Trying straight/orthogonal lines for relationships in graphviz debug graphs
instead of the curved defaults. The curvy lines were clumping together in confusing
ways, making it hard to tell what's going on. Still, this isn't much better, but
can at least see what's going on in the file I'm debugging.

Depsgraph: Report opcodes using their string identifiers instead of numbers

Depsgraph: Code cleanup - Remove unneeded define

Depsgraph: Noted 2 places where driver evaluation should not be taking place

We can't remove these outright yet, since these callbacks are still shared between
old depsgraph and new ubereval nodes.

Depsgraph: Knock out BKE_animsys_evaluate_all_animation() for new depsgraph

To debug what animdata is not yet included, we need to disable this. Eventually,
it should go, but we may need to put it back temporarily if we don't get
everything hooked up before setting more users loose on the codebase.

Depsgraph: GPencil animdata is now included in the depsgraph

Depsgraph: Clean up errors when building rigidbody graphs

* Constraints may not always be present
* There were missing links to "done" node

Depsgraph: Intial attempt at finishing off hooking up rigidbody sim code to the new depsgraph

Depsgraph: Added in missing calls to handle relations for subdata hanging off scene

Depsgraph: Make ID nodes fully fisible by default

Depsgraph: Fix missing editors ID update callback invoke

This way tweaking node trees reflects on shading update in GLSL viewport.

Depsgraph: Add shading component and empty operation in Material node

This way we can easily define relations between material and other entities, plus
this allows to have generic recalc flush mechanism working just fine.

Depsgraph: Create ID nodes for material, so this way tag update doesn't go berserk

Without this DEG_graph_id_tag_update would consider delayed ID update is needed
after the graph is rebuilt, which is just wrong.

There are no relations fr now, need of them is still being investigated.

Depsgraph: Fix missing rendered viewport update

Was only happening when blender is built with legacy depsgraph support.

Merge branch 'master' into depsgraph_refactor

Depsgraph: Fill in callbacks for mesh/curve geometry update

For now they'll only calculate bounding box there. in the future it might be
extended further.

Depsgraph: Headers cleanup

Depsgtraph: Move some more update callbacks to their final(ish) destination

Still having animation callbacks as a bad level implementation, but those are a
bit tricky since they somehow need to be able to access time source node.

Depsgraph: Header cleanup