Agent/examples/process_pipeline/script/fix_broken_outputs.py

"""
批量修复有问题的 output 文件
根据 schema_errors_report.txt 中的错误，删除相应文件并重跑 pipeline
"""
import re
import subprocess
from pathlib import Path

PIPELINE_DIR = Path(__file__).parent.parent
OUTPUT_DIR = PIPELINE_DIR / "output"
REPORT_PATH = PIPELINE_DIR / "script" / "schema_errors_report.txt"

def parse_error_report(report_path):
    """解析错误报告，返回需要修复的需求列表"""
    errors = {}

    with open(report_path, "r", encoding="utf-8") as f:
        for line in f:
            # 匹配格式: output\XXX\file.json: Schema mismatch: error_detail
            match = re.search(r'output[/\\](\d+)[/\\]([^:]+):\s*Schema mismatch:\s*(.+)', line)
            if match:
                req_id = match.group(1)
                file_name = match.group(2)
                error_detail = match.group(3).strip()

                if req_id not in errors:
                    errors[req_id] = []
                errors[req_id].append({
                    "file": file_name,
                    "error": error_detail
                })

    return errors

def determine_cleanup_strategy(req_id, error_items):
    """
    根据错误文件类型决定清理策略
    返回: (策略类型, 需要删除的文件列表, 问题平台列表)

    策略类型:
    - single_platform: 单个平台的 raw_cases 有问题，只重跑该平台 + 级联删除后续文件
    - full: 多个平台或整体 raw_cases 有问题，完整重跑
    - from_blueprint: blueprint 有问题，删除 blueprint + capabilities + strategy
    - from_capabilities: capabilities 有问题，删除 capabilities + strategy
    - from_strategy: 只有 strategy 有问题，只删除 strategy
    """
    # 提取有问题的平台和错误类型
    problem_platforms = []
    has_empty_array = False

    for item in error_items:
        file_name = item["file"]
        error_detail = item["error"]

        # 检查是否是空数组问题
        if "array is empty" in error_detail:
            has_empty_array = True

        # 跳过 strategies[1].reasoning is empty（非选中策略的 reasoning 为空是正常的）
        if "strategies[1].reasoning is empty" in error_detail:
            continue

        if "raw_cases" in file_name:
            # 提取平台名，如 "raw_cases/case_bili.json" -> "bili"
            match = re.search(r'case_([a-z]+)\.json', file_name)
            if match:
                problem_platforms.append(match.group(1))

    has_raw_cases_error = len(problem_platforms) > 0
    has_blueprint_error = any(item["file"] == "blueprint.json" for item in error_items)
    has_capabilities_error = any(item["file"] == "capabilities_extracted.json" for item in error_items)
    has_strategy_error = any(item["file"] == "strategy.json" for item in error_items)

    if has_raw_cases_error:
        if len(problem_platforms) == 1:
            # 单个平台有问题，只重跑该平台，但要级联删除后续所有文件
            platform = problem_platforms[0]
            return "single_platform", [f"raw_cases/case_{platform}.json", "blueprint.json", "capabilities_extracted.json", "strategy.json"], [platform]
        else:
            # 多个平台有问题，或者 raw_cases 整体有问题，完整重跑
            return "full", ["raw_cases", "blueprint.json", "capabilities_extracted.json", "strategy.json"], problem_platforms

    elif has_blueprint_error:
        # blueprint 有问题，删除 blueprint 及后续文件
        return "from_blueprint", ["blueprint.json", "capabilities_extracted.json", "strategy.json"], []

    elif has_capabilities_error:
        # capabilities 有问题，删除 capabilities 和 strategy
        return "from_capabilities", ["capabilities_extracted.json", "strategy.json"], []

    elif has_strategy_error:
        # 只有 strategy 有问题，只删除 strategy
        return "from_strategy", ["strategy.json"], []

    return "unknown", [], []

def delete_files(req_id, files_to_delete):
    """删除指定的文件"""
    output_dir = OUTPUT_DIR / req_id
    deleted = []

    for file_name in files_to_delete:
        if file_name == "raw_cases":
            # 删除整个 raw_cases 目录
            raw_cases_dir = output_dir / "raw_cases"
            if raw_cases_dir.exists():
                import shutil
                shutil.rmtree(raw_cases_dir)
                deleted.append(f"{req_id}/raw_cases/")
        elif file_name.startswith("raw_cases/"):
            # 删除单个 case 文件
            file_path = output_dir / file_name
            if file_path.exists():
                file_path.unlink()
                deleted.append(f"{req_id}/{file_name}")
        else:
            file_path = output_dir / file_name
            if file_path.exists():
                file_path.unlink()
                deleted.append(f"{req_id}/{file_name}")

    return deleted

def generate_rerun_commands(errors_dict):
    """生成重跑命令"""
    commands = []

    for req_id, error_files in sorted(errors_dict.items()):
        strategy, files_to_delete, problem_platforms = determine_cleanup_strategy(req_id, error_files)
        req_index = int(req_id) - 1  # run_pipeline.py 使用 0-based index

        if strategy == "single_platform":
            # 只重跑单个平台，blueprint/capabilities/strategy 已被级联删除
            platforms_str = ",".join(problem_platforms)
            commands.append({
                "req_id": req_id,
                "index": req_index,
                "strategy": strategy,
                "files_to_delete": files_to_delete,
                "command": f"python run_pipeline.py --index {req_index} --platforms {platforms_str}"
            })
        elif strategy == "full":
            # 完整重跑（包括所有 research）
            commands.append({
                "req_id": req_id,
                "index": req_index,
                "strategy": strategy,
                "files_to_delete": files_to_delete,
                "command": f"python run_pipeline.py --index {req_index}"
            })
        elif strategy == "from_blueprint":
            # 跳过 research，从 blueprint 开始
            commands.append({
                "req_id": req_id,
                "index": req_index,
                "strategy": strategy,
                "files_to_delete": files_to_delete,
                "command": f"python run_pipeline.py --index {req_index} --skip-research"
            })
        elif strategy == "from_capabilities":
            # 跳过 research；blueprint 仍在会被跳过，capabilities + strategy 会重跑
            commands.append({
                "req_id": req_id,
                "index": req_index,
                "strategy": strategy,
                "files_to_delete": files_to_delete,
                "command": f"python run_pipeline.py --index {req_index} --skip-research"
            })
        elif strategy == "from_strategy":
            # 跳过 research；blueprint/capabilities 仍在会被跳过，只重跑 strategy
            commands.append({
                "req_id": req_id,
                "index": req_index,
                "strategy": strategy,
                "files_to_delete": files_to_delete,
                "command": f"python run_pipeline.py --index {req_index} --skip-research"
            })

    return commands

def main():
    import argparse
    parser = argparse.ArgumentParser(description="Fix broken output files based on schema validation")
    parser.add_argument("--dry-run", action="store_true", help="Show what would be done without actually doing it")
    parser.add_argument("--delete-only", action="store_true", help="Only delete broken files, don't rerun")
    parser.add_argument("--rerun-only", action="store_true", help="Only rerun (assume files already deleted)")
    parser.add_argument("--workers", type=int, default=4, help="Number of parallel CMD windows to spawn")
    parser.add_argument("--no-spawn", action="store_true", help="Only generate worker .bat files, don't auto-spawn windows")
    args = parser.parse_args()

    report_path = REPORT_PATH
    if not report_path.exists():
        print(f"[ERROR] {report_path} not found. Run validate_schema.py first.")
        return

    print("[*] Parsing error report...")
    errors_dict = parse_error_report(report_path)

    if not errors_dict:
        print("[OK] No errors found in report!")
        return

    print(f"Found {len(errors_dict)} requirements with errors:")
    for req_id, items in sorted(errors_dict.items()):
        files_str = ", ".join(item["file"] for item in items)
        print(f"  - REQ_{req_id}: {files_str}")

    print("\n" + "="*60)
    print("[*] Generating cleanup and rerun plan...")
    print("="*60)

    commands = generate_rerun_commands(errors_dict)

    for cmd_info in commands:
        print(f"\n[FIX] REQ_{cmd_info['req_id']} ({cmd_info['strategy']}):")
        print(f"   Delete: {', '.join(cmd_info['files_to_delete'])}")
        print(f"   Rerun:  {cmd_info['command']}")

    if args.dry_run:
        print("\n[DRY-RUN] No files were deleted or commands executed.")
        return

    # 执行删除
    if not args.rerun_only:
        print("\n" + "="*60)
        print("[*] Deleting broken files...")
        print("="*60)

        for cmd_info in commands:
            deleted = delete_files(cmd_info['req_id'], cmd_info['files_to_delete'])
            for f in deleted:
                print(f"   ✓ Deleted: {f}")

    if args.delete_only:
        print("\n[DELETE-ONLY] Files deleted. Skipping rerun.")
        return

    # 生成多窗口并行批处理脚本
    print("\n" + "="*60)
    print("[*] Generating multi-window parallel batch scripts...")
    print("="*60)

    total_tasks = len(commands)
    num_workers = min(args.workers, total_tasks)

    # 使用 round-robin 分配任务到各个 worker
    worker_tasks = {i: [] for i in range(num_workers)}
    for i, cmd_info in enumerate(commands):
        worker_id = i % num_workers
        worker_tasks[worker_id].append(cmd_info)

    print(f"Distributing {total_tasks} tasks across {num_workers} workers...")

    # bat 文件生成在 pipeline 根目录（run_pipeline.py 所在目录）
    bat_dir = PIPELINE_DIR

    # 为每个 worker 生成独立的批处理文件
    for worker_id, tasks in worker_tasks.items():
        if not tasks:
            continue

        bat_path = bat_dir / f"fix_worker_{worker_id}.bat"
        with open(bat_path, "w", encoding="utf-8") as f:
            f.write("@echo off\n")
            f.write("cd /d %~dp0\n")
            f.write("if exist ..\\..\\venv\\Scripts\\activate.bat call ..\\..\\venv\\Scripts\\activate.bat\n")
            f.write("if exist ..\\..\\.venv\\Scripts\\activate.bat call ..\\..\\.venv\\Scripts\\activate.bat\n")
            f.write(f"echo Worker {worker_id} starting with {len(tasks)} tasks...\n")
            f.write("echo ========================================\n\n")

            for cmd_info in tasks:
                f.write(f"echo.\n")
                f.write(f"echo ========================================\n")
                f.write(f"echo [Worker {worker_id}] Fixing REQ_{cmd_info['req_id']} ({cmd_info['strategy']})\n")
                f.write(f"echo ========================================\n")
                f.write(f"{cmd_info['command']}\n")
                f.write(f"timeout /t 2 > NUL\n\n")

            f.write(f"echo.\n")
            f.write(f"echo ========================================\n")
            f.write(f"echo Worker {worker_id} completed all tasks!\n")
            f.write(f"echo ========================================\n")
            f.write("pause\n")

        print(f"  Worker {worker_id}: {len(tasks)} tasks -> {bat_path}")

    # 生成主启动脚本
    launcher_path = bat_dir / "fix_broken_outputs_parallel.bat"
    with open(launcher_path, "w", encoding="utf-8") as f:
        f.write("@echo off\n")
        f.write("cd /d %~dp0\n")
        f.write(f"echo Spawning {num_workers} parallel workers to fix {total_tasks} broken outputs...\n")
        f.write("echo ========================================\n\n")

        for worker_id in range(num_workers):
            if worker_id in worker_tasks and worker_tasks[worker_id]:
                f.write(f'start "Fix_Worker_{worker_id}" cmd /c fix_worker_{worker_id}.bat\n')
                f.write(f"timeout /t 1 > NUL\n")

        f.write("\necho.\n")
        f.write(f"echo All {num_workers} workers have been launched in separate windows.\n")
        f.write("echo ========================================\n")
        f.write("pause\n")

    print(f"\n[OK] Parallel batch scripts generated!")
    print(f"\nTo execute all fixes in parallel, run:")
    print(f"   {launcher_path}")
    print(f"\nOr manually launch individual workers:")
    for worker_id in range(num_workers):
        if worker_id in worker_tasks and worker_tasks[worker_id]:
            print(f"   fix_worker_{worker_id}.bat")

    # 如果不是 no-spawn 模式，自动启动窗口
    if not args.no_spawn:
        import subprocess
        print(f"\n[*] Auto-spawning {num_workers} worker windows...")
        for worker_id in range(num_workers):
            if worker_id in worker_tasks and worker_tasks[worker_id]:
                bat_file = bat_dir / f"fix_worker_{worker_id}.bat"
                subprocess.Popen(
                    ["cmd.exe", "/c", "start", f"Fix_Worker_{worker_id}", "cmd.exe", "/c", str(bat_file.absolute())],
                    cwd=str(bat_dir)
                )
                print(f"  Launched Worker {worker_id}")
        print(f"\n[OK] All workers launched! Check the new CMD windows.")

if __name__ == "__main__":
    main()