Behavior Preservation Checker

Behavior Preservation Checker

Overview

Validate that a migrated or refactored codebase preserves the original behavior by automatically comparing runtime behavior, test results, execution traces, and observable outputs between two repository versions.

Core Workflow

1. Setup Repositories

Prepare both repositories for comparison:

# Clone or locate repositories
ORIGINAL_REPO=/path/to/original
MIGRATED_REPO=/path/to/migrated

# Ensure both are at comparable states
cd $ORIGINAL_REPO && git checkout main
cd $MIGRATED_REPO && git checkout main

2. Run Behavior Comparison

Use the comparison script to analyze behavioral differences:

python scripts/behavior_checker.py \
    --original $ORIGINAL_REPO \
    --migrated $MIGRATED_REPO \
    --output behavior_report.json

3. Review Results

Examine the generated report for:

• Test result differences
• Execution trace divergences
• Output mismatches
• Performance regressions
• API contract violations

4. Fix Deviations

Follow actionable guidance to resolve behavioral differences.

Comparison Methods

Method 1: Test-Based Comparison

Run the same test suite on both repositories and compare results:

Workflow:

1. Identify common test suite (or create equivalent tests)
2. Run tests on original repository
3. Run tests on migrated repository
4. Compare pass/fail status, assertions, and outputs

Example:

# Run on original
cd $ORIGINAL_REPO
pytest tests/ --json-report --json-report-file=original_results.json

# Run on migrated
cd $MIGRATED_REPO
pytest tests/ --json-report --json-report-file=migrated_results.json

# Compare
python scripts/compare_test_results.py \
    original_results.json \
    migrated_results.json

Method 2: Execution Trace Comparison

Capture and compare execution traces:

Workflow:

1. Instrument code to capture function calls, arguments, and return values
2. Run identical inputs through both versions
3. Compare execution traces for divergences

Example:

# Trace original
python scripts/trace_execution.py \
    --repo $ORIGINAL_REPO \
    --input test_inputs.json \
    --output original_trace.json

# Trace migrated
python scripts/trace_execution.py \
    --repo $MIGRATED_REPO \
    --input test_inputs.json \
    --output migrated_trace.json

# Compare traces
python scripts/compare_traces.py \
    original_trace.json \
    migrated_trace.json

Method 3: Observable Output Comparison

Compare program outputs for identical inputs:

Workflow:

1. Define test inputs (API requests, CLI commands, function calls)
2. Capture outputs from both versions (stdout, files, API responses)
3. Compare outputs for differences

Example:

# Test API endpoints
python scripts/compare_api_outputs.py \
    --original-url http://localhost:8000 \
    --migrated-url http://localhost:8001 \
    --test-cases api_test_cases.json

Method 4: Property-Based Testing

Use property-based testing to find behavioral differences:

Workflow:

1. Define behavioral properties (invariants, contracts)
2. Generate random inputs
3. Verify properties hold for both versions
4. Report any property violations

Example:

# Property: sorting should produce same result
from hypothesis import given, strategies as st

@given(st.lists(st.integers()))
def test_sort_equivalence(input_list):
    original_result = original_sort(input_list)
    migrated_result = migrated_sort(input_list)
    assert original_result == migrated_result

Difference Detection

Test Result Differences

What to check:

• Tests that pass in original but fail in migrated
• Tests that fail in original but pass in migrated
• New test failures
• Changed assertion messages

Severity levels:

• Critical: Core functionality tests fail
• High: Integration tests fail
• Medium: Edge case tests fail
• Low: Flaky tests or timing-dependent failures

Execution Trace Differences

What to check:

• Different function call sequences
• Different argument values
• Different return values
• Missing or extra function calls

Example divergence:

Original trace:
  calculate(x=10) -> 20
  validate(20) -> True
  save(20) -> Success

Migrated trace:
  calculate(x=10) -> 21  # ← Difference!
  validate(21) -> True
  save(21) -> Success

Output Differences

What to check:

• Different stdout/stderr
• Different file contents
• Different API response bodies
• Different status codes
• Different error messages

Tolerance levels:

# Exact match required
assert original_output == migrated_output

# Numerical tolerance
assert abs(original_value - migrated_value) < 0.001

# Structural equivalence (ignore formatting)
assert json.loads(original) == json.loads(migrated)

Actionable Guidance

Pattern 1: Logic Error

Symptom: Different outputs for same inputs

Diagnosis:

python scripts/isolate_difference.py \
    --original $ORIGINAL_REPO \
    --migrated $MIGRATED_REPO \
    --failing-test test_calculation

Guidance:

1. Identify the diverging function
2. Compare implementations side-by-side
3. Check for off-by-one errors, operator changes, or logic inversions
4. Add unit test for the specific case

Pattern 2: Missing Functionality

Symptom: Tests pass in original but fail in migrated with "not implemented" or "attribute error"

Diagnosis:

python scripts/find_missing_functions.py \
    --original $ORIGINAL_REPO \
    --migrated $MIGRATED_REPO

Guidance:

1. List all missing functions/methods
2. Implement missing functionality
3. Verify with targeted tests

Pattern 3: API Contract Violation

Symptom: Different response structure or status codes

Diagnosis:

python scripts/compare_api_contracts.py \
    --original-spec openapi_original.yaml \
    --migrated-spec openapi_migrated.yaml

Guidance:

1. Document API contract differences
2. Update migrated API to match original contract
3. Add contract tests to prevent future violations

Pattern 4: Performance Regression

Symptom: Migrated version is significantly slower

Diagnosis:

python scripts/benchmark_comparison.py \
    --original $ORIGINAL_REPO \
    --migrated $MIGRATED_REPO \
    --iterations 100

Guidance:

1. Profile both versions to identify bottlenecks
2. Check for algorithmic changes (O(n) → O(n²))
3. Look for missing optimizations or caching
4. Verify database query efficiency

Pattern 5: State Management Issues

Symptom: Tests fail intermittently or depend on execution order

Diagnosis:

python scripts/detect_state_issues.py \
    --repo $MIGRATED_REPO \
    --test-suite tests/

Guidance:

1. Identify shared state between tests
2. Add proper setup/teardown
3. Ensure test isolation
4. Check for global variable usage

Report Format

The behavior checker generates a comprehensive JSON report:

{
  "summary": {
    "total_tests": 150,
    "passed_both": 140,
    "failed_both": 2,
    "passed_original_failed_migrated": 5,
    "failed_original_passed_migrated": 3,
    "behavioral_equivalence": "92.7%"
  },
  "differences": [
    {
      "type": "test_failure",
      "test_name": "test_user_authentication",
      "severity": "critical",
      "original_result": "passed",
      "migrated_result": "failed",
      "error_message": "AssertionError: Expected 200, got 401",
      "guidance": "Check authentication logic in migrated version",
      "affected_files": ["auth/login.py"]
    }
  ],
  "recommendations": [
    "Fix 5 critical test failures before deployment",
    "Review 3 output differences for correctness"
  ]
}

Best Practices

1. Start with tests: Ensure comprehensive test coverage before migration
2. Incremental validation: Check behavior after each migration step
3. Document intentional changes: Mark expected behavioral differences
4. Use multiple comparison methods: Combine tests, traces, and outputs
5. Automate the process: Integrate into CI/CD pipeline
6. Set tolerance thresholds: Define acceptable differences (e.g., timing, formatting)

Resources

• references/comparison_techniques.md: Detailed comparison methodologies
• references/difference_patterns.md: Common behavioral difference patterns
• scripts/behavior_checker.py: Main comparison orchestrator
• scripts/compare_test_results.py: Test result comparison
• scripts/trace_execution.py: Execution trace capture
• scripts/compare_traces.py: Trace comparison and analysis