Blender Git Loki

Extended TypeDesc to support compound type definitions (structures).

New util file for structure type implementations in C/C++.

Rename: POINTER base type to "RNAPOINTER".

This is so the POINTER type can be used for general struct pointers and arrays.

RNA pointers should eventually be replaced with more specific types for
data that can be accessed in the bvm system.

Fix for incorrect pointer type in smoke effector updates.

Moved util_typedesc file into the compile module.

Split off from the bvm::Function class the basic instruction list functionality.

The point of this is to keep type definitions out of the core bvm eval code.
TypeDesc is only used for the outermost layer of bvm::Function in order to
copy arguments to and results from the eval stack. OpCodes themselves are
agnostic to higher type definitions and just work with raw data.

TypeDesc registration in NodeGraph for reusable type definitions.

Eventually this could allow actual user typedefs (through the API),
but primarily this is to make complex type definitions possible without
having to do a full definition every time, and simply use the type name.

Merge branch 'master' into object_nodes

Fix a memleak of the return value user counter from mesh_ptr.

The returned mesh_ptr result needs to be cleared to ensure the user
counter is also freed. To avoid deleting the mesh data as well a new
function 'reset' was added, which frees the user counter but only resets
the data pointer without freeing it.

Another geometry node for loading another objects 'final derived' mesh.

Modifiers should be able to work on meshes directly as much as possible,
without direct reference to an object. The space transform should
ultimately be passed along with the DM (TODO, currently requires explicit
inputs).

Prepend a value node on all (non-constant) node inputs without a link.

In principle the input value could be used directly in codegen, but this
approach means one less condition to check and has the same effect.

Fix for incorrect output type displace element.index.

Added a simple operator and panel for one-click debug display of bvm node visualizations.

This was formerly part of an addon, but is useful enough to include directly in the branch
for debugging purposes. It uses the existing graphviz dump functions to write into a
temp dot file, then runs graphviz on it and opens the resulting image in the standard
system image viewer.

This is tailored for linux systems, but that shouldn't be a big issue since most developers
(who are the target audience) use this platform anyway.

Use the new node block structure in code generation, to determine dependencies of symbols.

More elegant construction of implicit node blocks, by copying argument node first.

Represent nested node blocks in the graphviz debug dump.

Cleanup: class for debug dumping functions for node graphs.

Moved the block definition from codegen to the node graph optimization.

This now involves making actual copies of nodes, so that each block has
a unique instance of each node and ambiguity is avoided.

Generalized algorithm for constructing node blocks based on local arguments.

Nodes can have expression inputs and override argument nodes with local
variables. This means that a local block must be created for each expression
input. The boundaries of such blocks are defined implicitly by upstream nodes
which define their own local arguments (or ultimately by the graph arguments).

The current version is still not quite generic enough to allow arbitrary
nesting of blocks, but it should ensure that expressions don't include all
the parent nodes which don't depend on local arguments and therefore should
be calculated in the parent block (a form of constant folding).

Use a global node index assignment to enable sorted sets for nodes.

This simplifies code generation because we can use a set with efficient
lookup which also keeps a valid ordering of nodes as dictated by links.
Since any generated code uses only subsets of nodes, the ordering still
is valid for any BasicBlock.