The Big Idea

The architecture of Dynamo was previously a tangled mess of dependencies. This PR refactors DynamoCore into a modular, layered architecture, where the boundaries of the various components of the application are clearly defined.

Or, in a nutshell, we are no longer accessing DynamoModel from remote parts of the application. This means that:

1. In order to test parts of the application, we don't need to create an entire instance of the application.
2. We can safely assume that different parts of the application aren't affecting other remote parts of the application unless there is a direct dependency between the two components.

What's been changed?

• XML parsing
  • The logic for loading whole Workspaces, Nodes, Connectors, and Notes are now all located in a single place: NodeGraph
  • Loading .dyn and .dyf files now leverage this new API, previously we had duplicate (and unsynchronized) code for loading both.
• Node instantiation
  • NodeFactory has been overhauled; previously we were doing weird string parsing to determine how to create new nodes. Nothing was consistent, the API was cumbersome to use, and there was a ton of duplicate code.
  • All node instantiation is now processed through NodeFactory. Different types of nodes register INodeLoader instances with the NodeFactory. For special kinds of nodes (Custom Nodes and Zero Touch Nodes), there are special INodeLoaders that handle the necessary extra logic for creating instances. This means that NodeFactory no longer needs to know about the Function and DSFunction classes.
  • All external implementations of Nodes which extend NodeModel are now instantiated using a pre-compiled Expression, as opposed to relying on an Activator. This means that instantiating large numbers of new nodes should be significantly faster (copy/paste, undo/redo, loading files).
• Custom Node Infrastructure
  • CustomNodeManager has become the primary interface for dealing with Custom Nodes.
    • It keeps track of CustomNodeWorkspaces, instead of just CustomNodeDefinitions.
  • CustomNodeWorkspace now emits events when the CustomNodeDefinition and CustomNodeInfo has changed.
    • This makes sense because the state of a CustomNodeWorkspace defines what the CustomNodeDefinition and CustomNodeInfo are.
    • These events are watched by CustomNodeManager, which then forwards them to DynamoModel, which then propagates them to the EngineController, Search Library, and Custom Node Instances.
  • Custom Node (Function) Instances are now created from CustomNodeManager, in order to ensure that they are kept in sync with their CustomNodeDefinition.
• Evaluation
  • Since HomeWorkspaceModels are the only kind of Workspaces that can be evaluated, EngineController has been moved to HomeWorkspaceModel
    • Consequentially, all logic originally on DynamoModel associated with EngineController has now been moved to HomeWorkspaceModel Unfortunately, CodeBlockNodeModels have a dependency on EngineController for Auto-complete. Further work can be done to decouple Auto-complete from EngineController, which would allow for total removal of EngineController from DynamoModel.
    • This sets the stage for eventually having multiple HomeWorkspaceModels active at once.
• Search
  • Search has been re-implemented to be as general as possible.
  • Search has been refactored to support the new method of instantiating Nodes.
  • The actual search method has been separated from the categorization infrastructure.
  • Categorization of Search Elements has been greatly generalized and simplified.

Reviewers:

• @lukechurch
• @ikeough
• @Benglin
• @pboyer