Method flush_eval_queue

src/benchmark/leetcode.py:1007–1209

刷新评测队列：批量 push 代码并依次评测。 Returns: 评测完成的任务列表

(self)

            return False
    
    async def flush_eval_queue(self) -> List[Task]:
        """
        刷新评测队列：批量 push 代码并依次评测。
        
        Returns:
            评测完成的任务列表
        """
        if not self._pending_queue:
            return []
        
        async with submit_lock:
            logger.info(f"| 🔒 Acquired lock for batch evaluation ({len(self._pending_queue)} tasks)")
            
            # 1. 确保 submitter 已初始化
            if not self._submitter_started:
                logger.info("| 🔍 Starting CodeSubmitter browser...")
                try:
                    await self._submitter.initialize()
                    self._submitter_started = True
                except Exception as e:
                    logger.error(f"| ❌ Failed to start submitter: {e}")
                    # 标记所有任务为失败，并保存结果（output_file 在 __init__ 中已设置）
                    for task in self._pending_queue:
                        task.score = 0.0
                        task.extra["submit_time"] = 0.0
                        task.extra["spend_time"] = task.extra.get("inference_time", 0.0)
                        self._tasks.append(task)
                        
                        # 保存错误结果
                        error_result = Result(
                            task_id=task.task_id,
                            prompt=task.input,
                            prediction="browser_init_error",
                            answer="None",
                            score=0.0,
                            metrics={"error": str(e), "inference_time": task.extra.get("inference_time", 0.0), "submit_time": 0.0},
                            extra=None,
                            start_time=task.extra.get("inference_start_time", time.time()),
                            end_time=time.time(),
                            spend_time=task.extra.get("inference_time", 0.0)
                        )
                        await self._submitter.save_result(error_result)
                    
                    result = self._pending_queue.copy()
                    self._pending_queue.clear()
                    return result
            
            # 2. 准备批量写入的文件
            batch_tasks = self._pending_queue.copy()
            self._pending_queue.clear()
            
            file_contents = []
            valid_tasks = []  # 有代码的任务
            
            for task in batch_tasks:
                code_content = task.result
                if not code_content:
                    # 无代码，直接标记失败