코드 실행 검증 루프

Python 3.11+ / Docker

#!/usr/bin/env python3
"""
코드 실행 검증기
생성된 코드를 Docker 샌드박스에서 실제 실행해 검증
"""

import subprocess
import tempfile
import os
import re
import json

def strip_html_tags(html: str) -> str:
    """HTML span 태그 제거해 순수 코드 추출"""
    return re.sub(r'<[^>]+>', '', html)

def validate_python(code: str) -> tuple[bool, str]:
    """Python 코드 실행 검증"""
    clean_code = strip_html_tags(code)
    with tempfile.NamedTemporaryFile(suffix='.py', mode='w', delete=False) as f:
        f.write(clean_code)
        tmp_path = f.name

    try:
        result = subprocess.run(
            ['python3', '-c', f'import ast; ast.parse(open("{tmp_path}").read())'],
            capture_output=True, text=True, timeout=5
        )
        if result.returncode != 0:
            return False, result.stderr

        # 실제 실행 (Docker 샌드박스 권장)
        result = subprocess.run(
            ['python3', tmp_path],
            capture_output=True, text=True, timeout=10
        )
        return result.returncode == 0, result.stderr or result.stdout
    except subprocess.TimeoutExpired:
        return False, "실행 시간 초과"
    finally:
        os.unlink(tmp_path)

def validate_go(code: str) -> tuple[bool, str]:
    """Go 코드 문법 검증"""
    clean_code = strip_html_tags(code)
    with tempfile.TemporaryDirectory() as tmpdir:
        # go.mod 생성
        with open(os.path.join(tmpdir, 'go.mod'), 'w') as f:
            f.write('module temp\ngo 1.22\n')
        with open(os.path.join(tmpdir, 'main.go'), 'w') as f:
            f.write(clean_code)

        result = subprocess.run(
            ['go', 'vet', './...'],
            capture_output=True, text=True,
            timeout=15, cwd=tmpdir
        )
        return result.returncode == 0, result.stderr

def validate_item(item: dict) -> dict:
    """항목의 코드를 검증하고 결과 반환"""
    lang = item.get('lang', '')
    code = item.get('code', '')

    validators = {
        'python': validate_python,
        'go':     validate_go,
    }

    if lang not in validators:
        return {'valid': True, 'note': f'{lang} 검증 미지원'}

    valid, output = validators[lang](code)
    return {
        'valid':   valid,
        'output':  output[:200] if output else '',
        'id':      item.get('id'),
    }

if __name__ == '__main__':
    import sys
    lang = sys.argv[1] if len(sys.argv) > 1 else 'python'
    ctx = {}
    exec(open(f'data/{lang}.js').read()
         .replace('window.DATA', 'DATA')
         .replace('(function () {', '')
         .replace('})();', ''), ctx)

    items  = ctx.get('DATA', {}).get(lang, {}).get('items', [])
    passed = 0
    failed = []

    for item in items[:5]:  # 처음 5개만 테스트
        result = validate_item(item)
        status = '✅' if result['valid'] else '❌'
        print(f"{status} {item['id']}: {item['title']}")
        if result['valid']:
            passed += 1
        else:
            failed.append(item['id'])
            print(f"   오류: {result['output'][:100]}")

    print(f"\n통과: {passed}/{len(items[:5])}")