Files
zui/frontend/docs/IMPROVEMENTS.md
2026-03-12 17:07:42 +01:00

48 lines
2.7 KiB
Markdown

# Codebase simplification and design patterns
This doc summarizes recent improvements and suggested next steps for readability, extension, and consistency.
## Done
### 1. Single place for template/reachability (DRY)
- **Added** `lib/graph/templateRefs.ts`: `isReachable`, `resolveExtendsRef`, `getTemplateRefs`.
- **Refactored** `config/renderingLogic.ts` and `useRenderingNodeState.ts` to use these helpers instead of duplicating the same logic.
- **Pattern:** Extract shared pure helpers into a small lib module; keep call sites thin and consistent.
### 2. Naming and comments
- **Renamed** `outputMenuRegistry.tsx``outputMenuHandlers.tsx` (no registry, only helpers).
- **Updated** `rendering.ts` and `sourceRenderingLogic.ts`: output menu is described as coming from the node descriptor (`getOutputMenuContent`), not a separate registry.
- **Documented** `NodeMenubar`: extra content can come from props or from the descriptor (`getNodeMenuExtraContent`).
- **Documented** `nodeTypes.ts`: clarifies React Flow types vs node type id (nodeRegistry).
### 3. Central pipeline entry
- **rendering.ts** documents the 3-step pipeline, how to add a source/output type, and points to `templateRefs.ts` for shared helpers.
## Design patterns in use
| Pattern | Where |
|----------------|--------------------------------------------|
| **Registry** | nodeRegistry, sourceRenderingLogic |
| **Builder** | nodeTypeBuilder (descriptor per node type) |
| **Pipeline** | Resolve → Render → Display (rendering.ts) |
| **Strategy** | Source logic per node type; output menu per descriptor |
| **Shared helpers** | templateRefs, outputMenuHandlers, renderingUtils |
## Suggested next steps
1. **CanvasPage** (~950 lines): Split into smaller units, e.g.:
- `useCanvasGraph()` or similar for graph state and connection rules.
- A dedicated component for the context menu (add node, paste, etc.).
- Keeps CanvasPage as composition + layout.
2. **useRenderingNodeState**: Consider extracting:
- Signature building (config/edges/variables/functions/data) into a pure function or small module, e.g. `buildSourceSignatures(nodes, edges, id, incomingIds, getConfigContent)`.
- Makes the hook easier to read and the logic testable in isolation.
3. **Config types**: If you add more output types (e.g. Mermaid), consider a small registry API (`registerConfigType`, `getConfigType`) instead of a single large `CONFIG_TYPES` array, so extensions can register without editing the core list.
4. **Consistent node shape in lib**: `templateRefs` uses `EdgeLike` / `NodeLike`; other graph code uses inline `{ source, target }` or `nodes as ...`. You could standardize on the same minimal types where appropriate to reduce casts.