Blender Git Loki

Depsgraph: Move task callback to the depsgraph_eval file

It's just more a hassle of having separate file and some public-ish
API to pass all the data around. Easier just have single file with
all evaluation.

Depsgraph: Avoid having one global lock for all the graphs

That's a bit crazy to have single global lock in the .cpp file which
could be easily coupled into Depsgraph structure so it would never
interfere if several threads traverses the graphs.

Depsgraph: A bit more cleanup

Hopefully it's only me who touches this areas atm.

Depsgraph: Code cleanup, wrapping and whitespace

Depsgraph: Fix one more use-after-free issue

Depsgraph: Move using std::vector to depsgraph_types.h

Such a generic types declarations shouldn't happen inbetween of
declaration of own stuff.

Depsgraph: Code cleanup

Depsgraph: Fix order of depsgraph destroying

First need to remove all the nodes and links, then free the root, otherwise it's
possible to access freed data on relations destroy.

This is actually much bigger can of worms i'm afraid, need to make sure no access
to previously freed nodes happens when doing full depsgraph destroy.

Depsgraph: Correction to previous commit, it's unsafe to cast nodes

We shouldn't cast time source node to operation node and access the flag.

Depsgraph: Adding/removing modifiers should tag relations for update

Depsgraph: Fix assert failure in the parents counting

Currently time source is allowed parent for modifiers which depends on time,
so we can not expect parents to be operation nodes.

Depsgraph: Compilation error fixes for the strict compiler rules

Depsgraph Debugging: Rudimentary stats for depsgraph

Added a method/operator which prints out how many operations, relations, and outer nodes
we have in the graph at a time. This is mostly useful to check on the complexity of
rig files (e.g. koro from caminandes).

Code Cleanup: Style + Removing dead code

Fix: Relationships to time source nodes were getting removed by reachability pruning

Moved DepsRelations (i.e. inlinks/outlinks lists) back to DepsNode as originally intended

Otherwise, it really doesn't make much sense that we're deriving all these different
node types from DepsNode, when these can't represent relationships between different
nodes. Also, this will be useful for some upcoming new node types for keeping track
of other kinds of dependencies other than what's executable.

Made debugging identifiers for nodes less "scary" looking

Fixed logic for verify_entry_exit_ops()

This was assuming that there were relationships between nodes when checking for
inlinks and outlinks. However, since relationships are computed in a later pass,
there are always none, and it ends up nearly always adding unncessary entry/exit
noops (unless there's just a single op in the component).

Fix: Some of the object transform relations weren't quite right

* For now, we need to connect the parent to the component init, or else
the parent matrix doesn't accumulate correctly for object parenting

* Final transform op wasn't getting hooked in at all

Object transform evaluation is now done in new-depsgraph-driven callbacks instead of ubereval

* The ubereval is still lurking around for handling proxy cruft

* I've implemented a stripped-down version of the parenting code for
use here (notably, without the slow parenting, and originmat support).
The old/full version did a bit too much for us to be able to keep the
old code running while we worked on the new version too, without making
the code too messy.

Later on, when we work out the geometry handling a bit more, we may need
to replace the parent-matrix finding code it uses too (i.e. to make it
grab data from context instead - especially in cases where it hunts for
this info itself currently).