streaming fixed

app.py

@@ -668,35 +668,58 @@ with gr.Blocks(title="Audit Q&A", css= "style.css", theme=theme,elem_id = "main-
             None  # no need to store query
         )
     
+    # async def handle_chat_flow(warning_active, query, chatbot, sources, reports, subtype, year, client_ip, session_id):
+    #     """Handle chat flow with hard stop for warnings"""
+    #     # Check if any filters are selected
+    #     no_filters = (not reports) and (not sources) and (not subtype) and (not year)
+        
+    #     if warning_active:  # Only check warning_active, not no_filters
+    #         # If warning is active, return unchanged components
+    #         return (
+    #             chatbot,  # unchanged chatbot
+    #             "",  # empty sources
+    #             None,  # no feedback state
+    #             session_id  # keep session
+    #         )
+        
+    #     # Include start_chat functionality here
+    #     history = chatbot + [(query, None)]
+    #     history = [tuple(x) for x in history]
+        
+    #     # Proceed with chat
+    #     async for update in chat(query, history, sources, reports, subtype, year, client_ip, session_id):
+    #         # The last update will be returned
+    #         chatbot_update, sources_update, feedback_update, session_update = update
+        
+    #     return (
+    #         chatbot_update,
+    #         sources_update,
+    #         feedback_update,
+    #         session_update
+    #     )
+    
     async def handle_chat_flow(warning_active, query, chatbot, sources, reports, subtype, year, client_ip, session_id):
         """Handle chat flow with hard stop for warnings"""
         # Check if any filters are selected
         no_filters = (not reports) and (not sources) and (not subtype) and (not year)
         
         if warning_active:  # Only check warning_active, not no_filters
-            # If warning is active, return unchanged components
-            return (
+            # If warning is active, yield unchanged components instead of returning
+            yield (
                 chatbot,  # unchanged chatbot
                 "",  # empty sources
                 None,  # no feedback state
                 session_id  # keep session
             )
+            return  # Exit the generator
         
         # Include start_chat functionality here
         history = chatbot + [(query, None)]
         history = [tuple(x) for x in history]
         
-        # Proceed with chat
+        # Proceed with chat and yield each update
         async for update in chat(query, history, sources, reports, subtype, year, client_ip, session_id):
-            # The last update will be returned
-            chatbot_update, sources_update, feedback_update, session_update = update
-        
-        return (
-            chatbot_update,
-            sources_update,
-            feedback_update,
-            session_update
-        )
+            yield update
     
     #-------------------- Gradio voodoo -------------------------
     
@@ -710,7 +733,7 @@ with gr.Blocks(title="Audit Q&A", css= "style.css", theme=theme,elem_id = "main-
     #         queue=True, concurrency_limit=8, api_name="chat_textbox")
     #     .then(show_feedback, [feedback_state], [feedback_row, feedback_thanks, feedback_state], api_name="show_feedback_textbox")
     #     .then(finish_chat, None, [textbox], api_name="finish_chat_textbox"))
-    
+
     (textbox
         .submit(
             check_filters,
@@ -723,10 +746,11 @@ with gr.Blocks(title="Audit Q&A", css= "style.css", theme=theme,elem_id = "main-
             [client_ip],
             show_progress=False
         )
-        .then(  # Remove start_chat from here
+        .then(
             handle_chat_flow,
             [warning_state, textbox, chatbot, dropdown_sources, dropdown_reports, dropdown_category, dropdown_year, client_ip, session_id],
-            [chatbot, sources_textbox, feedback_state, session_id]
+            [chatbot, sources_textbox, feedback_state, session_id],
+            queue=True  # Changed from streaming=True
         )
         .then(
             handle_feedback,
@@ -758,10 +782,11 @@ with gr.Blocks(title="Audit Q&A", css= "style.css", theme=theme,elem_id = "main-
         get_client_ip_handler,
         [textbox],
         [client_ip]
-    ).then(  # Remove start_chat call and let handle_chat_flow handle the history update
+    ).then(
         handle_chat_flow,
         [warning_state, textbox, chatbot, dropdown_sources, dropdown_reports, dropdown_category, dropdown_year, client_ip, session_id],
-        [chatbot, sources_textbox, feedback_state, session_id]
+        [chatbot, sources_textbox, feedback_state, session_id],
+        queue=True  # Changed from streaming=True
     ).then(
         handle_feedback,
         [feedback_state],