cweill / gotests / 18882875710

94%

push

github

feat: AI-powered test case generation (#194)

* feat: add AI test generation foundation (WIP)

Implementing AI-powered test case generation as proposed in issue #41.
This is the foundation layer with provider interface and Ollama support.

New package: internal/ai/
- provider.go: Provider interface and config structs
- ollama.go: Ollama provider implementation with retry logic
- prompt.go: Prompt templates for LLM test case generation
- validator.go: In-memory compilation validation using go/parser

CLI additions:
- `-ai`: Enable AI test case generation
- `-ai-model`: Specify model (default: qwen2.5-coder:0.5b)
- `-ai-endpoint`: Ollama endpoint (default: localhost:11434)
- `-ai-cases`: Number of cases to generate (default: 3)

Options propagation:
- Added UseAI, AIModel, AIEndpoint, AICases to Options structs
- Flows from CLI flags → process.Options → gotests.Options

Still TODO:
- Integrate AI into output processing
- Modify templates for AI case injection
- Testing and validation

Related to #41

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* feat: complete AI test case generation integration

Implements full integration of AI-powered test case generation:

1. Added function body extraction:
   - Modified goparser to extract function body source code
   - Added Body field to models.Function for AI context
   - Implemented extractFunctionBody helper using AST positions

2. Enhanced AI prompt with one-shot examples:
   - Added example for simple functions (Max)
   - Added example for error-returning functions (Divide)
   - Includes function body in prompt for better context
   - Aligned prompt with wantName() helper conventions

3. Template integration:
   - Updated function.tmpl to render AI-generated test cases
   - Falls back to TODO comment when AI is not enabled/fails
   - Properly handles Args and Want maps from TestCase struct

4. Configuration improvements:
   - Set temperature... (continued)

690 of 753 new or added lines in 10 files covered. (91.63%)

1552 of 1647 relevant lines covered (94.23%)

1320.52 hits per line