Better interruption handler.

Author: wehos

main_logic/omni_realtime_client.py

@@ -151,6 +151,9 @@ def __init__(
         # Fatal error detection - 检测到致命错误后立即中断
         self._fatal_error_occurred = False  # 致命错误标志
 
+        # Interruption state - suppress output after user interruption until next response
+        self._interrupted = False  # 打断状态标志，防止重复消息块
+        
         # Native image input rate limiting
         self._last_native_image_time = 0.0  # 上次原生图片输入时间戳
 
@@ -586,6 +589,9 @@ async def handle_interruption(self):
 
         logger.info("Handling interruption")
 
+        # Mark as interrupted to suppress any remaining output until next response
+        self._interrupted = True
+
         # 1. Cancel the current response
         if self._current_response_id:
             await self.cancel_response()
@@ -664,6 +670,7 @@ async def handle_messages(self) -> None:
                 elif event_type == "response.created":
                     self._current_response_id = event.get("response", {}).get("id")
                     self._is_responding = True
+                    self._interrupted = False  # Clear interruption flag on new response
                     self._is_first_text_chunk = self._is_first_transcript_chunk = True
                     # 清空转录 buffer，防止累积旧内容
                     self._output_transcript_buffer = ""
@@ -690,7 +697,7 @@ async def handle_messages(self) -> None:
                     self._print_input_transcript = False
                     self._output_transcript_buffer = ""
 
-                if not self._skip_until_next_response:
+                if not self._skip_until_next_response and not self._interrupted:
                     if event_type in ["response.text.delta", "response.output_text.delta"]:
                         if self.on_text_delta:
                             if "glm" not in self.model: