hanabhi/gridworld-env

1# Copyright (c) Meta Platforms, Inc. and affiliates.2# All rights reserved.3#4# This source code is licensed under the BSD-style license found in the5# LICENSE file in the root directory of this source tree.6 7#!/usr/bin/env python38"""9Simple test showing how users will use CodingEnv.from_docker_image().10 11This is the simplest possible usage12"""13 14import sys15from pathlib import Path16 17# Add src to path18sys.path.insert(0, str(Path(__file__).parent.parent / "src"))19 20from coding_env import CodeAction, CodingEnv21 22 23def main():24    """Test CodingEnv.from_docker_image()."""25    print("=" * 60)26    print("CodingEnv.from_docker_image() Test")27    print("=" * 60)28    print()29 30    try:31        # This is what users will do - just one line!32        print("Creating client from Docker image...")33        print("  CodingEnv.from_docker_image('coding-env:latest')")34        print()35 36        client = CodingEnv.from_docker_image("coding-env:latest")37 38        print("✓ Client created and container started!\n")39 40        # Now use it like any other client41        print("Testing the environment:")42        print("-" * 60)43 44        # Reset45        print("\n1. Reset:")46        result = client.reset()47        print(f"   stdout: {result.observation.stdout}")48        print(f"   stderr: {result.observation.stderr}")49        print(f"   exit_code: {result.observation.exit_code}")50 51        # Get initial state52        state = client.state()53        print(f"   State: episode_id={state.episode_id}, step_count={state.step_count}")54 55        # Execute some Python code56        print("\n2. Execute Python code:")57 58        code_samples = [59            "print('Hello, World!')",60            "x = 5 + 3\nprint(f'Result: {x}')",61            "import math\nprint(f'Pi is approximately {math.pi:.4f}')",62            "# Multi-line calculation\nfor i in range(1, 4):\n    print(f'{i} squared is {i**2}')",63        ]64 65        for i, code in enumerate(code_samples, 1):66            result = client.step(CodeAction(code=code))67            print(f"   {i}. Code: {code.replace(chr(10), '\\n')[:50]}...")68            print(f"      → stdout: {result.observation.stdout.strip()}")69            print(f"      → exit_code: {result.observation.exit_code}")70            if result.observation.stderr:71                print(f"      → stderr: {result.observation.stderr}")72 73        # Test error scenarios74        print("\n3. Test error scenarios:")75 76        error_samples = [77            ("Division by zero", "x = 1 / 0\nprint('Should not reach here')"),78            ("Undefined variable", "print(undefined_variable)"),79            ("Syntax error", "print('Hello'"),80        ]81 82        for i, (description, code) in enumerate(error_samples, 1):83            result = client.step(CodeAction(code=code))84            print(f"   {i}. {description}")85            print(f"      Code: {code.replace(chr(10), '\\n')[:40]}...")86            print(f"      → exit_code: {result.observation.exit_code}")87            if result.observation.stderr:88                # Truncate long error messages89                error_msg = result.observation.stderr[:100]90                if len(result.observation.stderr) > 100:91                    error_msg += "..."92                print(f"      → stderr: {error_msg}")93 94        # Check final state95        print("\n4. Check final state:")96        state = client.state()97        print(f"   episode_id: {state.episode_id}")98        print(f"   step_count: {state.step_count}")99        print(f"   last_exit_code: {state.last_exit_code}")100 101        print("\n" + "-" * 60)102        print("\n✓ All operations successful!")103        print()104 105        print("Cleaning up...")106        client.close()107        print("✓ Container stopped and removed")108        print()109 110        print("=" * 60)111        print("Test completed successfully! 🎉")112        print("=" * 60)113 114        return True115 116    except Exception as e:117        print(f"\n❌ Test failed: {e}")118        import traceback119        traceback.print_exc()120        return False121 122 123if __name__ == "__main__":124    success = main()125    exit(0 if success else 1)126