Add Phase 6 verification scripts and manual testing tools

verify_enhanced_features.py

@@ -0,0 +1,378 @@
+"""
+Manual verification script for Phase 6 testing.
+Tests session management, file operations, and multi-language execution.
+"""
+
+import requests
+import time
+import sys
+from pathlib import Path
+
+BASE_URL = "http://localhost:7860"
+
+def print_test(name):
+    print(f"\n{'='*60}")
+    print(f"TEST: {name}")
+    print(f"{'='*60}")
+
+def print_success(msg):
+    print(f"✅ {msg}")
+
+def print_error(msg):
+    print(f"❌ {msg}")
+    
+def print_info(msg):
+    print(f"ℹ️  {msg}")
+
+
+def test_1_session_lifecycle():
+    """Test creating, getting, and destroying a session"""
+    print_test("Session Lifecycle")
+    
+    # Create session
+    print_info("Creating session...")
+    resp = requests.post(f"{BASE_URL}/sessions", json={
+        "metadata": {"test": "lifecycle"},
+        "timeout_minutes": 30
+    })
+    
+    if resp.status_code != 201:
+        print_error(f"Failed to create session: {resp.status_code}")
+        return None
+    
+    session = resp.json()
+    session_id = session["session_id"]
+    print_success(f"Session created: {session_id}")
+    
+    # Wait for container to be ready
+    time.sleep(3)
+    
+    # Get session details
+    print_info("Getting session details...")
+    resp = requests.get(f"{BASE_URL}/sessions/{session_id}")
+    if resp.status_code == 200:
+        print_success("Session retrieved successfully")
+    else:
+        print_error(f"Failed to get session: {resp.status_code}")
+    
+    # List sessions
+    print_info("Listing all sessions...")
+    resp = requests.get(f"{BASE_URL}/sessions")
+    if resp.status_code == 200:
+        sessions = resp.json()
+        print_success(f"Found {len(sessions)} active session(s)")
+    else:
+        print_error(f"Failed to list sessions: {resp.status_code}")
+    
+    return session_id
+
+
+def test_2_file_operations(session_id):
+    """Test file upload, list, download"""
+    print_test("File Operations")
+    
+    if not session_id:
+        print_error("No session available")
+        return False
+    
+    # Upload file
+    print_info("Uploading test file...")
+    file_content = b"print('Hello from uploaded file!')\nprint(f'2 + 2 = {2+2}')"
+    resp = requests.post(
+        f"{BASE_URL}/sessions/{session_id}/files",
+        files={"file": ("test_script.py", file_content, "text/x-python")}
+    )
+    
+    if resp.status_code != 200:
+        print_error(f"Failed to upload file: {resp.status_code} - {resp.text}")
+        return False
+    
+    upload_result = resp.json()
+    print_success(f"File uploaded: {upload_result['filename']} ({upload_result['size']} bytes)")
+    
+    # List files
+    print_info("Listing files...")
+    resp = requests.get(f"{BASE_URL}/sessions/{session_id}/files")
+    if resp.status_code == 200:
+        files = resp.json()
+        print_success(f"Found {len(files)} file(s) in workspace")
+        for f in files:
+            print(f"  - {f['filename']} ({f['size']} bytes)")
+    else:
+        print_error(f"Failed to list files: {resp.status_code}")
+        return False
+    
+    # Download file
+    print_info("Downloading file...")
+    resp = requests.get(f"{BASE_URL}/sessions/{session_id}/files/test_script.py")
+    if resp.status_code == 200:
+        if resp.content == file_content:
+            print_success("File downloaded successfully and content matches")
+        else:
+            print_error("Downloaded file content doesn't match")
+            return False
+    else:
+        print_error(f"Failed to download file: {resp.status_code}")
+        return False
+    
+    return True
+
+
+def test_3_session_execution(session_id):
+    """Test code execution in session"""
+    print_test("Session-Based Code Execution")
+    
+    if not session_id:
+        print_error("No session available")
+        return False
+    
+    # Execute Python code
+    print_info("Executing Python code...")
+    resp = requests.post(
+        f"{BASE_URL}/sessions/{session_id}/execute",
+        json={
+            "code": "import sys\nprint(f'Python version: {sys.version}')\nprint('Execution successful!')",
+            "language": "python",
+            "working_dir": "/workspace"
+        }
+    )
+    
+    if resp.status_code != 200:
+        print_error(f"Failed to execute code: {resp.status_code} - {resp.text}")
+        return False
+    
+    result = resp.json()
+    print_success("Code executed successfully")
+    print(f"  Output: {result['stdout'].strip()}")
+    print(f"  Exit code: {result['exit_code']}")
+    print(f"  Execution time: {result['execution_time']}s")
+    
+    return True
+
+
+def test_4_file_execution(session_id):
+    """Test executing uploaded file"""
+    print_test("Execute Uploaded File")
+    
+    if not session_id:
+        print_error("No session available")
+        return False
+    
+    # Execute the previously uploaded file
+    print_info("Executing uploaded Python file...")
+    resp = requests.post(
+        f"{BASE_URL}/sessions/{session_id}/execute-file",
+        json={
+            "filepath": "/workspace/test_script.py",
+            "language": "python",
+            "args": []
+        }
+    )
+    
+    if resp.status_code != 200:
+        print_error(f"Failed to execute file: {resp.status_code} - {resp.text}")
+        return False
+    
+    result = resp.json()
+    print_success("File executed successfully")
+    print(f"  Output:\n{result['stdout']}")
+    print(f"  Exit code: {result['exit_code']}")
+    
+    return True
+
+
+def test_5_persistent_state(session_id):
+    """Test that files persist across executions"""
+    print_test("Persistent State Verification")
+    
+    if not session_id:
+        print_error("No session available")
+        return False
+    
+    # First execution: create a file
+    print_info("Creating data file in first execution...")
+    resp = requests.post(
+        f"{BASE_URL}/sessions/{session_id}/execute",
+        json={
+            "code": "with open('/workspace/persistent.txt', 'w') as f: f.write('Data persists!')",
+            "language": "python"
+        }
+    )
+    
+    if resp.status_code != 200:
+        print_error(f"Failed first execution: {resp.status_code}")
+        return False
+    
+    print_success("File created in first execution")
+    
+    # Second execution: read the file
+    print_info("Reading file in second execution...")
+    resp = requests.post(
+        f"{BASE_URL}/sessions/{session_id}/execute",
+        json={
+            "code": "with open('/workspace/persistent.txt', 'r') as f: print(f'Read: {f.read()}')",
+            "language": "python"
+        }
+    )
+    
+    if resp.status_code != 200:
+        print_error(f"Failed second execution: {resp.status_code}")
+        return False
+    
+    result = resp.json()
+    if "Data persists!" in result['stdout']:
+        print_success("File persisted across executions!")
+        print(f"  Output: {result['stdout'].strip()}")
+    else:
+        print_error("File did not persist")
+        return False
+    
+    return True
+
+
+def test_6_concurrent_sessions():
+    """Test creating multiple concurrent sessions"""
+    print_test("Concurrent Sessions")
+    
+    session_ids = []
+    
+    # Create 3 sessions
+    for i in range(3):
+        print_info(f"Creating session {i+1}/3...")
+        resp = requests.post(f"{BASE_URL}/sessions", json={
+            "metadata": {"test": "concurrent", "index": i}
+        })
+        
+        if resp.status_code == 201:
+            session_id = resp.json()["session_id"]
+            session_ids.append(session_id)
+            print_success(f"Session {i+1} created: {session_id[:8]}...")
+        else:
+            print_error(f"Failed to create session {i+1}")
+    
+    # List all sessions
+    resp = requests.get(f"{BASE_URL}/sessions")
+    if resp.status_code == 200:
+        all_sessions = resp.json()
+        print_success(f"Total active sessions: {len(all_sessions)}")
+    
+    # Cleanup
+    print_info("Cleaning up concurrent sessions...")
+    for sid in session_ids:
+        requests.delete(f"{BASE_URL}/sessions/{sid}")
+    
+    print_success("Concurrent session test completed")
+    return True
+
+
+def test_7_cleanup(session_id):
+    """Test session cleanup"""
+    print_test("Session Cleanup")
+    
+    if not session_id:
+        print_info("No session to cleanup")
+        return True
+    
+    print_info(f"Destroying session {session_id[:8]}...")
+    resp = requests.delete(f"{BASE_URL}/sessions/{session_id}")
+    
+    if resp.status_code == 200:
+        print_success("Session destroyed successfully")
+        
+        # Verify it's gone
+        resp = requests.get(f"{BASE_URL}/sessions/{session_id}")
+        if resp.status_code == 404:
+            print_success("Session verified as destroyed")
+            return True
+        else:
+            print_error("Session still exists after destruction")
+            return False
+    else:
+        print_error(f"Failed to destroy session: {resp.status_code}")
+        return False
+
+
+def test_8_stateless_execution():
+    """Test backward compatibility - stateless execution"""
+    print_test("Stateless Execution (Backward Compatibility)")
+    
+    print_info("Executing code without session...")
+    resp = requests.post(f"{BASE_URL}/execute", json={
+        "code": "print('Stateless execution works!')\nprint(f'Result: {10 * 10}')",
+        "language": "python"
+    })
+    
+    if resp.status_code == 200:
+        result = resp.json()
+        print_success("Stateless execution successful")
+        print(f"  Output: {result['stdout'].strip()}")
+        print(f"  Exit code: {result['exit_code']}")
+        return True
+    else:
+        print_error(f"Stateless execution failed: {resp.status_code}")
+        return False
+
+
+def main():
+    """Run all verification tests"""
+    print("\n" + "="*60)
+    print("ENHANCED SANDBOX - MANUAL VERIFICATION")
+    print("="*60)
+    
+    # Check if API is running
+    try:
+        resp = requests.get(f"{BASE_URL}/health", timeout=2)
+        if resp.status_code != 200:
+            print_error("API server is not healthy")
+            sys.exit(1)
+        print_success("API server is running")
+    except requests.exceptions.RequestException as e:
+        print_error(f"Cannot connect to API server at {BASE_URL}")
+        print_info("Please run 'python app.py' first")
+        sys.exit(1)
+    
+    results = {}
+    session_id = None
+    
+    # Run all tests
+    try:
+        session_id = test_1_session_lifecycle()
+        results['session_lifecycle'] = session_id is not None
+        
+        results['file_operations'] = test_2_file_operations(session_id)
+        results['session_execution'] = test_3_session_execution(session_id)
+        results['file_execution'] = test_4_file_execution(session_id)
+        results['persistent_state'] = test_5_persistent_state(session_id)
+        results['concurrent_sessions'] = test_6_concurrent_sessions()
+        results['stateless_execution'] = test_8_stateless_execution()
+        results['cleanup'] = test_7_cleanup(session_id)
+        
+    except Exception as e:
+        print_error(f"Test failed with exception: {e}")
+        import traceback
+        traceback.print_exc()
+    
+    # Summary
+    print("\n" + "="*60)
+    print("VERIFICATION SUMMARY")
+    print("="*60)
+    
+    passed = sum(1 for v in results.values() if v)
+    total = len(results)
+    
+    for test, result in results.items():
+        status = "✅ PASS" if result else "❌ FAIL"
+        print(f"{status} - {test.replace('_', ' ').title()}")
+    
+    print(f"\nResults: {passed}/{total} tests passed")
+    
+    if passed == total:
+        print_success("All verification tests passed!")
+        sys.exit(0)
+    else:
+        print_error(f"{total - passed} test(s) failed")
+        sys.exit(1)
+
+
+if __name__ == "__main__":
+    main()

verify_phase6.ps1

@@ -0,0 +1,142 @@
+# Phase 6: Integration & Testing Verification Script
+# Run this script after Docker is started to verify all enhanced sandbox features
+
+Write-Host "========================================" -ForegroundColor Cyan
+Write-Host "Enhanced Sandbox - Phase 6 Verification" -ForegroundColor Cyan
+Write-Host "========================================" -ForegroundColor Cyan
+Write-Host ""
+
+# Check Docker
+Write-Host "1. Checking Docker..." -ForegroundColor Yellow
+try {
+    docker ps *> $null
+    if ($LASTEXITCODE -eq 0) {
+        Write-Host "   ✅ Docker is running" -ForegroundColor Green
+    }
+    else {
+        Write-Host "   ❌ Docker is not running. Please start Docker Desktop." -ForegroundColor Red
+        exit 1
+    }
+}
+catch {
+    Write-Host "   ❌ Docker is not available." -ForegroundColor Red
+    exit 1
+}
+Write-Host ""
+
+# Run unit tests
+Write-Host "2. Running Unit Tests..." -ForegroundColor Yellow
+pytest tests/test_executor.py -v
+if ($LASTEXITCODE -ne 0) {
+    Write-Host "   ⚠️  Some executor tests failed" -ForegroundColor Yellow
+}
+Write-Host ""
+
+# Build devenv image
+Write-Host "3. Building Development Environment Image..." -ForegroundColor Yellow
+Write-Host "   (This may take 10-15 minutes on first build)" -ForegroundColor Gray
+docker build -f sandbox/images/devenv.Dockerfile -t sandbox-devenv:latest sandbox/images/
+if ($LASTEXITCODE -eq 0) {
+    Write-Host "   ✅ Development environment image built successfully" -ForegroundColor Green
+}
+else {
+    Write-Host "   ❌ Failed to build image" -ForegroundColor Red
+    exit 1
+}
+Write-Host ""
+
+# Verify environment
+Write-Host "4. Verifying Development Environment..." -ForegroundColor Yellow
+docker run --rm sandbox-devenv:latest /opt/environment_check.sh
+if ($LASTEXITCODE -eq 0) {
+    Write-Host "   ✅ Environment verified" -ForegroundColor Green
+}
+else {
+    Write-Host "   ⚠️  Some tools missing in environment" -ForegroundColor Yellow
+}
+Write-Host ""
+
+# Start API server in background
+Write-Host "5. Starting API Server..." -ForegroundColor Yellow
+$apiProcess = Start-Process python -ArgumentList "app.py" -PassThru -NoNewWindow
+Start-Sleep -Seconds 5
+
+# Check if server started
+try {
+    $response = Invoke-WebRequest -Uri "http://localhost:7860/health" -UseBasicParsing
+    if ($response.StatusCode -eq 200) {
+        Write-Host "   ✅ API server started successfully" -ForegroundColor Green
+    }
+}
+catch {
+    Write-Host "   ❌ API server failed to start" -ForegroundColor Red
+    Stop-Process -Id $apiProcess.Id -Force
+    exit 1
+}
+Write-Host ""
+
+# Run session management tests
+Write-Host "6. Testing Session Management..." -ForegroundColor Yellow
+pytest tests/test_session_manager.py -v
+if ($LASTEXITCODE -eq 0) {
+    Write-Host "   ✅ Session management tests passed" -ForegroundColor Green
+}
+else {
+    Write-Host "   ⚠️  Some session tests failed" -ForegroundColor Yellow
+}
+Write-Host ""
+
+# Run file operation tests
+Write-Host "7. Testing File Operations..." -ForegroundColor Yellow
+pytest tests/test_file_operations.py -v
+if ($LASTEXITCODE -eq 0) {
+    Write-Host "   ✅ File operation tests passed" -ForegroundColor Green
+}
+else {
+    Write-Host "   ⚠️  Some file operation tests failed" -ForegroundColor Yellow
+}
+Write-Host ""
+
+# Run API integration tests
+Write-Host "8. Testing API Integration..." -ForegroundColor Yellow
+pytest tests/test_api.py -v
+if ($LASTEXITCODE -eq 0) {
+    Write-Host "   ✅ API integration tests passed" -ForegroundColor Green
+}
+else {
+    Write-Host "   ⚠️  Some API tests failed" -ForegroundColor Yellow
+}
+Write-Host ""
+
+# Manual verification tests
+Write-Host "9. Running Manual Verification Tests..." -ForegroundColor Yellow
+python verify_enhanced_features.py
+if ($LASTEXITCODE -eq 0) {
+    Write-Host "   ✅ Manual verification passed" -ForegroundColor Green
+}
+else {
+    Write-Host "   ⚠️  Some manual tests failed" -ForegroundColor Yellow
+}
+Write-Host ""
+
+# Cleanup
+Write-Host "10. Cleanup..." -ForegroundColor Yellow
+Stop-Process -Id $apiProcess.Id -Force
+Write-Host "    ✅ API server stopped" -ForegroundColor Green
+Write-Host ""
+
+Write-Host "========================================" -ForegroundColor Cyan
+Write-Host "Verification Complete!" -ForegroundColor Cyan
+Write-Host "========================================" -ForegroundColor Cyan
+Write-Host ""
+Write-Host "Summary:" -ForegroundColor White
+Write-Host "  - Docker: Running" -ForegroundColor Green
+Write-Host "  - DevEnv Image: Built" -ForegroundColor Green
+Write-Host "  - Unit Tests: Check output above" -ForegroundColor Yellow
+Write-Host "  - Integration Tests: Check output above" -ForegroundColor Yellow
+Write-Host "  - API Server: Tested" -ForegroundColor Green
+Write-Host ""
+Write-Host "Next steps:" -ForegroundColor White
+Write-Host "  1. Review test results above" -ForegroundColor Gray
+Write-Host "  2. Run 'python app.py' to start the server" -ForegroundColor Gray
+Write-Host "  3. Test with example requests from README.md" -ForegroundColor Gray