Blender Git Loki

Keep a list of input arguments of a function in addition to return values.

Get rid of the extra list of "outputs" for subgraph codegen.

Instead we now pass a socket index map directly to subgraph codegen,
and then read out the final output indices externally.

Pass input arguments as const void* when calling bvm functions.

This is not used yet, but will replace the EvalData pointer.

New set of node types for 'arguments' which will act as stack entry placeholders.

These are like value nodes in that an output socket stack entry is reserved for them.
However, like 'pass' nodes they don't actually perform an operation. Their purpose is
to provide a place on the stack for input arguments that other nodes can then read from.

Expose the get_input function of the node graph just like get_output.

Bump the BVM stack size to 4095.

Currently there is no optimization for stack usage, which would ultimately
reduce the required stack size dramatically. 255 is still very low though.

Use individual node types for constructing transform matrices.

Transforms can be created as translations, rotations (euler and axis/angle)
and scale.

The UI nodes include and automatic multiplication node for convenience.

Use an explicit transform input in the array modifier node.

Transforms can only be created from translations atm.

Fix for mismatching socket type identifiers.

Have to be careful about distinguishing socket type identifiers
and bvm type identifiers.

Fix for ugly socket color because of swapped G and B ...

Merge branch 'master' into object_nodes

Some basic math operations for 4x4 matrices.

Moved math eval functions to a dedicated file.

Introduced a UI node socket type for affine transforms.

This matches the matrix44 internal bvm type. It is useful particulary
for describing rotations.

Fixed and improved the socket conversion function.

Cleanup: Renamed socket type utility functions for clarity.

Moved the socket converter function into the socket_type module.

Fix for re-entrant update problems, using a base class with a context manager.

Make node groups work by linking proxy nodes to the external sockets of the parent node graph.

Leave the creation of proxies for bNode inputs/outputs to the node compiler.