[ENH] Use tree-sitter for robust C API parsing and auto-generation

[tazarov]

Description

Currently, the OrtApi struct generator (tools/gen_ortapi.go) uses regex-based parsing to extract function pointers from the ONNX Runtime C header file. While this works for the current API, it has limitations:

• Fragile: Regex patterns may break with header formatting changes
• Limited: Only extracts function names, not full signatures
• Manual: Requires manual pattern updates for new C API patterns

Proposal

Migrate to using tree-sitter-c for parsing the C header files. This would enable:

1. Robust parsing: Proper C AST parsing handles all valid C syntax
2. Full signature extraction: Parse return types, parameter types, and names
3. Auto-generate purego bindings: Generate not just struct definitions but also Go wrapper functions
4. Better validation: Detect breaking changes in C API structure

Potential Implementation

// Pseudo-code example
func parseWithTreeSitter(headerPath string) []FunctionDecl {
    // Use tree-sitter-c to parse the header
    parser := treesitter.NewParser()
    parser.SetLanguage(c.GetLanguage())
    tree := parser.Parse(readFile(headerPath))
    
    // Walk the AST to find OrtApi struct
    // Extract function pointer declarations with full signatures
    // Generate both struct and wrapper functions
}

Benefits

• Type safety: Generate typed wrappers instead of manual purego.RegisterFunc calls
• Documentation: Extract and preserve C API comments
• Maintainability: Easier to update when ONNX Runtime evolves
• Correctness: Eliminate regex parsing bugs

Related

• Current generator: tools/gen_ortapi.go
• Generated file: ort/ortapi_generated.go
• Issue [TST] Fix OrtApi struct layout to enable proper ReleaseEnv cleanup #20: Fixed by complete struct definition

Implementation Notes

This is a non-trivial enhancement that should be done incrementally:

1. First, implement tree-sitter parsing alongside existing regex approach
2. Validate both produce identical results
3. Switch to tree-sitter-only generation
4. Add auto-generation of wrapper functions (optional)

References

• tree-sitter-c
• go-tree-sitter