z3/.github/workflows/README-deeptest.md

DeepTest - Automated Test Case Generator

Overview

DeepTest is an AI-powered GitHub Agentic Workflow that automatically generates comprehensive test cases for Z3 source files. It analyzes a given source file and creates high-quality, production-ready tests covering basic functionality, edge cases, error handling, and integration scenarios.

Features

• Comprehensive Test Coverage: Generates tests for basic functionality, edge cases, error handling, and integration scenarios
• Multiple Languages: Supports C++, Python, Java, C#, and other Z3 API languages
• Smart Analysis: Uses Serena language server for Python code analysis and grep/glob for C++ analysis
• Automated PR Creation: Creates a pull request with generated tests automatically
• Follows Z3 Conventions: Generates tests that match existing Z3 testing patterns and style
• Persistent Memory: Uses cache memory to track progress across runs

How to Use

Method 1: Workflow Dispatch (Manual Trigger)

1. Go to Actions → Deeptest in the GitHub repository
2. Click Run workflow
3. Enter the file path (e.g., src/util/vector.h)
4. Optionally link to an issue number
5. Click Run workflow

Method 2: Slash Command (In Issues/PRs)

Simply comment on any issue or pull request:

/deeptest src/util/vector.h

The workflow will:

1. Extract the file path from your comment
2. Analyze the source file
3. Generate comprehensive tests
4. Create a pull request with the test files
5. Add a comment with statistics and instructions

What Gets Generated

DeepTest creates test files that include:

For C++ Files

• Unit tests using Z3's testing framework
• Located in src/test/test_<module_name>.cpp
• Follows existing test patterns in the repository
• Includes necessary headers and setup/teardown code

For Python Files

• Unit tests using unittest or pytest
• Located in src/api/python/test_<module_name>.py
• Follows patterns from existing Python tests
• Includes proper imports and test fixtures

Test Categories

Generated tests cover:

1. Basic Functionality: Happy path scenarios with typical inputs
2. Edge Cases: Boundary values, empty inputs, zero/negative values, very large inputs
3. Error Handling: Invalid parameters, null pointers, exceptions, assertion violations
4. Integration Tests: Realistic SMT-LIB2 formulas, solver workflows, theory combinations

Output

After running, DeepTest will:

1. Create a Pull Request with:

   • Title: [DeepTest] Add comprehensive tests for <filename>
   • Generated test file(s)
   • Detailed description of test coverage
   • Instructions for running the tests
   • Labels: automated-tests, deeptest

2. Post a Comment with:

   • Test statistics (number of test cases by category)
   • Coverage percentage
   • Link to the created PR
   • Instructions for running the tests

Example Usage

Example 1: Test a C++ utility file

/deeptest src/util/vector.h

Example 2: Test a Python API file

/deeptest src/api/python/z3/z3.py

Example 3: Via workflow dispatch

• File path: src/ast/ast.cpp
• Issue number: 1234 (optional)

Running Generated Tests

After the PR is merged, run the tests:

# Build Z3
python scripts/mk_make.py
cd build && make -j$(nproc)

# Run the new tests
./test-z3 [test-name-pattern]

Configuration

The workflow is configured with:

• Timeout: 30 minutes
• Permissions: Read-only (safe-outputs handle writes)
• Network: Default curated allow-list
• Tools: Serena (Python), bash, edit, grep, glob, GitHub API
• Cache: Persistent memory enabled

Customization

To modify the agent behavior without recompiling:

1. Edit .github/agentics/deeptest.md
2. Changes take effect immediately (no compilation needed)
3. For configuration changes in .github/workflows/deeptest.md, run:

   gh aw compile deeptest
   

Limitations

• Does not modify existing source files (only creates new test files)
• Focuses on public APIs and functions
• May not cover all internal implementation details
• Generated tests should be reviewed before merging

Security

• Read-only permissions for the main job
• Safe outputs handle all write operations (PR creation, comments)
• Network access restricted to curated allow-list
• No secrets exposed to the AI agent

Support

If DeepTest encounters issues or needs additional tools:

• It will automatically create an issue labeled missing-tool
• The issue will expire after 1 week if not addressed
• Check the workflow run logs for detailed error information

See Also

• GitHub Agentic Workflows Documentation
• Z3 Testing Guide
• Z3 Build Instructions