Perfomance and Accuracy

[aram-devdocs]

Performance and Accuracy Improvements

Current Problems

• Generated files are becoming extremely large (tens of thousands of lines)
• Complex inline type generation causing IDE and TypeScript server crashes
• Circular dependencies leading to performance issues
• Overuse of z.any and any types in generated schemas
• Input/Output type accuracy needs improvement

Goals

Performance

• Implement circular dependency management solution
• Add caching system for previously encountered types
• Add configurable "level of detail" for type generation
• Make imports from other files optional vs inline generation
• Implement depth limit in config to cap infinite nesting

Accuracy

• Reduce/eliminate use of z.any in schema generation
• Improve type accuracy in input/output files
• Add relationship flattening configuration options
• Implement type caching to better handle repeated types

Code Organization

• Separate schema generation into dedicated module
• Add pre-generation cleanup of output directory
• Implement generation change logging
• Consolidate file writing/generation logic
• Add granular type definition controls

Key Areas to Focus

Code References

1. Type Generation

// generator.ts:69-90
// Review model file generation loop
for (const m of models) {
  const content = generateModelFileInline(m, [...models, ...types], enumMap, modelMap, config);
  const filePath = resolvePath(outputDir, `model_${m.name}.ts`);
  await writeTsFile({ filePath, content, config });
  modelFiles.push(filePath);
}

2. Configuration
```ts
// generator.ts:24-45
// Expand config options for granular control
const config: Config = {
  // Add new options:
  // - relationshipFlatteningLevel
  // - typeGenerationDetail  
  // - importStrategy
  // - cacheStrategy
}

3. File Management

// libs/utils/writeTsFile.ts
// Add pre-write cleanup and logging
export async function writeTsFile({
  filePath,
  content,
  config,
}: {
  filePath: string;
  content: string;
  config: Config;
})