CAD Viewer Architecture Review

A CAD viewer often starts as a small rendering tool and gradually becomes an application platform. Once model size, selection, annotation, measurement, and interaction requirements grow, the architecture needs clear boundaries.

Background

The viewer needs to handle imported geometry, scene organization, user interaction, and rendering performance. These concerns should not collapse into one large view class.

Problems

Common pressure points include:

• Import logic leaking into rendering code.
• Selection state being duplicated in multiple modules.
• Scene updates becoming hard to reason about.
• UI operations depending on low-level rendering details.

Design Goals

The first goal is to separate model data, scene representation, interaction state, and rendering resources. The second goal is to keep the update path explicit enough that performance work can be targeted.

Layered Structure

A practical structure can start with these layers:

1. Import and data normalization.
2. Domain model and metadata.
3. Scene graph or render scene construction.
4. Interaction tools, selection, and highlighting.
5. Rendering backend and GPU resource management.
6. UI shell and command routing.

add_library(cad_viewer_core)
target_sources(cad_viewer_core PRIVATE
  model/Document.cpp
  scene/SceneBuilder.cpp
  interaction/SelectionService.cpp
)

Follow-up Plan

Future notes can expand on large model loading, incremental scene updates, picking accuracy, render cache invalidation, and how to keep the viewer testable without requiring a full UI runtime.