This blog post is related to my Google Summer of Code 2024 project: Procedural Fragment Shader Generation Using Classic Machine Learning.

After encountering numerous challenges with the nodeeditor package, I made the strategic decision to develop a custom implementation. This decision was driven by the need for precision and maintainability, especially considering that this project’s ultimate goal extends beyond a Visual Shader Editor to a machine learning approach for shader generation.

Analysis of Third-Party Solution

Limitations of nodeeditor Package

1. Scope Mismatch: The library encompasses backend functionality that I had already implemented prior to the midterm evaluation. It exceeds the requirements of a GUI library, introducing unnecessary complexity.
2. Bug Density: The package contains numerous bugs that would require significant time investment to resolve.
3. Integration Challenges: Attempts to integrate the library with my midterm evaluation code proved problematic, potentially rendering integration unfeasible.
4. Architectural Constraints: As documented in Issue #149, the global theming system prevents the creation of multiple editor instances. While a fix exists in PR #172, it remains unmerged.

Custom Implementation Benefits

Simplified Port Management

The nodeeditor library appears to use a complex algorithm for port detection and event handling. My approach simplifies this by treating ports as discrete objects:

1. Port Objects as Children: By making ports children of nodes, we achieve clearer object hierarchy and simpler event propagation.

2. Event Handler Reduction: This approach requires only three key functions:

   on_port_pressed

   Initializes or resets the “temporary” connection object. A temporary connection exists in a dragging state, remaining uncommitted until finalized.

   on_port_dragged

   Manages the continuous update of temporary connections during mouse movement. The implementation is straightforward, focusing solely on creation and updating of the temporary connection object.

   on_port_dropped

   Handles the connection finalization when the mouse is released. Port detection is simplified as each port is a discrete object.

While implementing this custom solution was time-intensive and occasionally daunting due to its scope, the resulting system is more maintainable and aligned with our specific needs.

Event System Architecture

The implementation uses a simple but effective upward event propagation model:

• Events emit from widgets and traverse up the widget tree until reaching the target
• This ensures encapsulation, as children cannot directly access parents
• Event handling occurs at the appropriate level in the widget hierarchy

Testing Framework Development

Visual Shader Editor Testing Infrastructure

I’ve established a new Tests subdirectory within RGM, focusing on:

• Foundation for testing MainWindow and VisualShaderEditor classes
• Exploration of various testing methodologies, guided by resources like Difference between Mocks, Fakes, Stubs and Dummies

Testing Challenges and Solutions

1. Protobuf and Model Mocking: Initial attempts to mock MessageModel and ProtoModel classes proved problematic
2. Isolation vs. Integration: Balancing the need to test the VisualShaderEditor in isolation while also testing system integration
3. Workaround Implementation: Created a parameter-less constructor for VisualShaderEditor to facilitate testing

Current Testing Status

While UI testing presents significant challenges, we’ve established a solid foundation:

• Created the Tests directory structure
• Implemented tests for the VisualShader class
• Acknowledged the complexity of comprehensive UI testing and prioritized accordingly

Project Extension Considerations

Despite aspirations to extend the project to include machine learning components, time constraints have necessitated focusing on core functionality. The integration complexity between Qt5, Protobuf, gRPC, VisualShader, VisualShaderEditor, JDI, and ENIGMA’s Graphics System requires careful consideration and extensive testing.

Notes

As confirmed by Robert on August 30, 2024:

R0bert — 30/08/2024 17:58

be careful im not sure if we can extend yours again this year, so dont count on it i saw mentor thread talking about that

This reinforces our focus on delivering a robust core implementation rather than expanding scope.