directed-test-input-generator

# Directed Test Input Generator

Generate test inputs that target specific code paths and hard-to-reach behaviors using program analysis, coverage feedback, and LLM-driven semantic understanding.

## Overview

Directed test input generation combines multiple techniques to create test inputs that explore specific execution paths:

1. **Path Analysis**: Extract control flow paths and their constraints
2. **Constraint Solving**: Generate inputs satisfying path conditions
3. **Coverage Guidance**: Use coverage feedback to iteratively reach new paths
4. **LLM Semantic Understanding**: Leverage code understanding for meaningful inputs

## Quick Start

### Basic Workflow

```python
# 1. Analyze code to extract paths
from scripts.path_analyzer import analyze_code_paths

paths = analyze_code_paths(source_code)

# 2. Generate inputs for each path
from scripts.input_generator import generate_test_suite

test_suite = generate_test_suite(paths)

# 3. Execute tests and measure coverage
for path_id, test_data in test_suite.items():
    result = execute_test(function, test_data['inputs'])
    verify_coverage(result, test_data['target_line'])
```

## Core Techniques

### 1. Path Analysis and Extraction

Extract execution paths and their constraints from code:

```python
from scripts.path_analyzer import analyze_code_paths, print_paths

source = """
def validate_age(age, country):
    if age < 0:
        raise ValueError("Invalid age")
    if age < 18:
        return "minor"
    if age >= 65 and country == "US":
        return "senior_us"
    return "adult"
"""

paths = analyze_code_paths(source)
print_paths(paths)

# Output:
# Path #0: exception handler (ValueError) (line 3)
#   Conditions:
#     - age < 0
#
# Path #1: if branch (line 5)
#   Conditions:
#     - age >= 0
#     - age < 18
#
# Path #2: if branch (line 7)
#   Conditions:
#     - age >= 0
#     - age >= 18
#     - age >= 65
#     - country == US
```

### 2. Constraint-Based Input Generation

Generate inputs that satisfy specific path constraints:

```python
from scripts.input_generator import TestInputGenerator

generator = TestInputGenerator()

# Generate input for path: age >= 65 AND country == "US"
constraints = {
    "age": ["age >= 65"],
    "country": ["country == US"]
}

inputs = generator.generate_for_path(constraints)
# Result: {"age": 65, "country": "US"}
```

### 3. Edge Case Generation

Generate boundary values systematically:

```python
from scripts.input_generator import EdgeCaseGenerator

# Generate edge cases for integer type
edge_cases = EdgeCaseGenerator.generate_edge_cases(int)
# [0, -1, 1, -2147483648, 2147483647, ...]

# Generate boundary values around a threshold
boundaries = EdgeCaseGenerator.generate_boundary_values(">", 65)
# [64, 65, 66, 67] - values around the boundary
```

### 4. Coverage-Guided Generation

Iteratively generate inputs to maximize coverage:

```python
def coverage_guided_testing(function, max_iterations=100):
    """
    Generate test inputs guided by coverage feedback.
    """
    covered_lines = set()
    test_corpus = []

    # Start with initial inputs
    current_inputs = generate_seed_inputs(function)

    for i in range(max_iterations):
        # Execute and measure coverage
        new_coverage = execute_with_coverage(function, current_inputs)

        if new_coverage - covered_lines:
            # New coverage reached - save inputs
            test_corpus.append(current_inputs)
            covered_lines.update(new_coverage)

        # Mutate inputs to explore new paths
        current_inputs = mutate_toward_uncovered(current_inputs, covered_lines)

    return test_corpus
```

See **[coverage_strategies.md](references/coverage_strategies.md)** for detailed coverage-guided strategies.

### 5. LLM-Driven Semantic Generation

Use LLM understanding of code semantics to generate meaningful inputs:

```python
# Example: LLM generates semantically appropriate inputs
function_code = """
def book_flight(passenger_age, departure_date, destination_country):
    # ... booking logic
"""

# LLM understands parameter semantics and generates realistic inputs:
llm_generated = [
    {
        "passenger_age": 35,           # Adult
        "departure_date": "2026-06-15",  # Future date
        "destination_country": "US"      # Valid country code
    },
    {
        "passenger_age": 8,            # Child passenger
        "departure_date": "2026-07-20",
        "destination_country": "UK"
    },
    {
        "passenger_age": 72,           # Senior citizen
        "departure_date": "2026-08-10",
        "destination_country": "CA"
    }
]
```

See **[llm_patterns.md](references/llm_patterns.md)** for comprehensive LLM-driven generation patterns.

## Common Use Cases

### Use Case 1: Target Specific Branch

Generate input to reach an uncovered branch:

```python
# Target: age > 100 branch
def check_age(age):
    if age > 100:
        return "exceptionally_old"  # Want to test this
    return "normal"

# Analyze paths
paths = analyze_code_paths(check_age_source)
target_path = paths[0]  # age > 100 path

# Generate input
generator = TestInputGenerator()
constraints = target_path.get_constraints()
test_input = generator.generate_for_path(constraints)
# Result: {"age": 101}

# Test
assert check_age(**test_input) == "exceptionally_old"
```

### Use Case 2: Systematic Boundary Testing

Test all boundaries in a function:

```python
def calculate_discount(age, is_premium):
    if age < 18:
        return 0.0
    elif age < 65:
        return 0.1 if is_premium else 0.05
    else:
        return 0.2 if is_premium else 0.15

# Generate boundary test suite
boundaries = [
    {"age": 17, "is_premium": False},  # Just below 18
    {"age": 18, "is_premium": False},  # Exactly 18
    {"age": 19, "is_premium": True},   # Just above 18
    {"age": 64, "is_premium": False},  # Just below 65
    {"age": 65, "is_premium": True},   # Exactly 65
    {"age": 66, "is_premium": False},  # Just above 65
]

for inputs in boundaries: