Blender Git Loki

Fix for unfreed memory of the default empty mesh.

When using a static variable for this it needs to be destroyed explicitly.

Reference counting for mesh_ptr variables on the stack.

Any extensive data stored as a pointer on the stack should be ref-counted,
so that it stays alive only as long as used by some node.

The way this works is by enclosing instructions that access the stack
entry with 'init'/'release' opcodes. These are added automatically based
on output socket types (nodes don't have to be created explicitly for them).

Merge branch 'master' into object_nodes

Extended graphviz debug printing to also show the graph input/output connections.

Node graph optimization: skip over 'pass' nodes.

These nodes don't have opcodes any more, they are merely placeholders
to simplify node graph construction. Ultimately they will be removed
in the 'finalize' function, along with any other unreachable nodes.

Removed unused function declaration get_node_type_from_opcode

Use a single geometry subtree also for the graphviz debug function.

Use the geometry subtree in object nodes rather than the toplevel node tree.

This is currently a hack, since only a single geotree is used. Eventually
all geometry subtree result should just be combined.

Removed unused code.

Added a 'PASS_MESH' node.

Make an extra copy of the resulting node DM after eval.

This is necessary unfortunately because of the unreliable nature of dm->needsFree.
Without it the BKE_object_free_derived_caches function can destroy our static
empty DM at any point.

Added an assertion to make sure we don't add invalid node instances with unknown types.

Added a graphviz debug dump function for the object modifier node tree.

Use the default output value for graph returns in case they remain unlinked.

In case of meshes this will always be an empty mesh, but it at least gives a valid return value.

Foundations for a mesh data type in bvm

Mesh type for storing DerivedMesh as a temporary storage on the node eval stack.

Use a full TypeDesc in node sockets.

For the time being this is synonymous with the base type, but in future
it could allow (fixed-size) arrays and vector semantics.

Set of basic node classes for representing modifiers in pynodes.

Simple integration of object nodes into the modifier stack evaluation.

Merge branch 'master' into object_nodes