Key Concepts

Where the geometric model resource has a fixed set of component types, the graph-based model is intended to hold application-specific components. Furthermore, it uses modern C++ template metaprogramming techniques to offer type safety.


instances contain graph nodes and arcs (which are traditionally called graph edges — but we wish to avoid confusion with geometric edges). All of the nodes you create are owned by the resource, along with the arcs between them. By default, nodes are held in a NodeSet whose underlying storage is a std::set<std::shared_ptr<smtk::graph::Component>> with a comparator that orders components by UUID. However alternate types of storage are supported and the programming interface is geared to make indexing nodes by their UUID, type, and name possible.

The resource class itself requires a single template parameter that is a type-traits object which contains

  • a NodeTypes type-alias, expected to be a std::tuple of subclasses of the graph component class described below;

  • an ArcTypes type-alias, expected to be a std::tuple of arcs allowed to connect nodes to form a graph; and

  • an optional NodeContainer type-alias which the resource will inherit as the container for shared-pointers to nodes (instead of NodeSet) if it is present.

Thus the resource class is not a concrete implementation but rather a base class for other resources.


is a subclass of the geometry subsystem’s Component; instances of graph components serve as nodes in the resource’s graph. This subclass exists to extend the base API with methods for accessing graph nodes related by registered arc types. As above, you are expected to subclass this class with node types specific to your application.


is any struct or class holding type-traits and methods that specify how arcs of the given type behave (i.e., what types of nodes they may connect) and, optionally, are stored. An arc class may present either explicit or implicit arcs.

Explicit arcs are those which are explicitly stored by SMTK as pairs of connected node IDs.

Implicit arcs are those which are implied by the nature of data being represented (e.g., in a structured grid, points are implicitly connected to neighors; also, every node in one layer of a neural network is connected to every node in the next layer – given a pair of nodes in the neural network it is easy to decide whether they are connected by looking at their layer assignment) or by a third-party library (e.g., a CAD modeling kernel already stores relationships between edges and faces; rather than duplicate and maintain this relationship in SMTK, we forward requests about connectivity to the CAD library).

Resources that include an arc class in their list of arc types decorate the type-traits class you pass with templates that completely implement a convenient and consistent interface from the description and any partial implementation your arc class provides.

The decorated version of your arc class is then instantiated and stored as an entry in the resource’s ArcMap. The arc map, held by a resource instance, is where explicit arc storage lives (if any).

Developers who wish to construct their own implementation of a graph resource should start by going through smtk/graph/testing/cxx/TestArcs.cxx to understand the front-facing API before going through the implementation.


Create runtime-configurable classes (ProgrammableResource, ProgrammableNode, ProgrammableArc) that can be Python-wrapped and configured via a JSON serialization.