Update app.py

app.py

@@ -51,7 +51,10 @@ llm = ChatGroq(
     api_key=GROQ_API_KEY,
 )
 
-PROGRAMMING_ASSISTANT_PROMPT = """
+# --- Define the standard error message for failed parsing
+LLM_PARSE_ERROR_MESSAGE = "I'm sorry, I couldn't process the last response correctly due to a formatting issue. Could you please rephrase or try a simpler query?"
+
+PROGRAMMING_ASSISTANT_PROMPT = f"""
 You are an expert programming assistant. Your role is to provide code suggestions, fix bugs, explain programming concepts, and offer contextual help based on the user's query and preferred programming language.
 
 **CONTEXT HANDLING RULES (Follow these strictly):**
@@ -61,7 +64,7 @@ You are an expert programming assistant. Your role is to provide code suggestion
 STRICT OUTPUT FORMAT (JSON ONLY):
 Return a single JSON object with the following keys. **The JSON object MUST be enclosed in a single ```json block.**
 - assistant_reply: string  // A natural language reply to the user (short and helpful). Do NOT include code blocks here.
-- code_snippet: string  // If suggesting code, provide it here in a markdown code block (e.g., ```python\\nprint('Hello')\\n```). If no code is required, use an empty string: "".
+- code_snippet: string  // If suggesting code, provide it here in a markdown code block. **CRITICALLY, you must escape all internal newlines as '\\n' and backslashes as '\\\\'** to keep the string value valid JSON. If no code is required, use an empty string: "".
 - state_updates: object  // updates to the internal state, must include: language, conversationSummary
 - suggested_tags: array of strings // a list of 1-3 relevant tags for the assistant_reply
 
@@ -72,8 +75,10 @@ Rules:
 """
 
 def extract_json_from_llm_response(raw_response: str) -> dict:
+    # The default object is only used if parsing fails, providing a clean error message.
+    # The actual state preservation logic is in the /chat route.
     default = {
-        "assistant_reply": "I'm sorry, I couldn't process the response correctly. Could you please rephrase?",
+        "assistant_reply": LLM_PARSE_ERROR_MESSAGE,
         "code_snippet": "",
         "state_updates": {"conversationSummary": "", "language": "Python"},
         "suggested_tags": [],
@@ -82,25 +87,25 @@ def extract_json_from_llm_response(raw_response: str) -> dict:
     if not raw_response or not isinstance(raw_response, str):
         return default
     
-    # Use a non-greedy regex to find the JSON content inside the first code block
+    # 1. Use regex to find the JSON content inside the first code block (```json)
     m = re.search(r"```json\s*([\s\S]*?)\s*```", raw_response)
     json_string = m.group(1).strip() if m else raw_response
     
-    # Further refine candidate to just the JSON object content
+    # 2. Refine candidate to just the JSON object content
     first = json_string.find('{')
     last = json_string.rfind('}')
     candidate = json_string[first:last+1] if first != -1 and last != -1 and first < last else json_string
     
-    # Remove trailing commas which can break JSON parsing
+    # 3. Remove trailing commas which can break JSON parsing
     candidate = re.sub(r',\s*(?=[}\]])', '', candidate)
     
     try:
         parsed = json.loads(candidate)
     except Exception as e:
-        logger.warning("Failed to parse JSON from LLM output: %s. Candidate: %s", e, candidate)
+        logger.warning("Failed to parse JSON from LLM output: %s. Candidate: %s", e, candidate[:200]) # Truncate candidate for cleaner logs
         return default
     
-    # Validate and clean up the parsed dictionary
+    # 4. Validate and clean up the parsed dictionary
     if isinstance(parsed, dict) and "assistant_reply" in parsed:
         parsed.setdefault("code_snippet", "")
         parsed.setdefault("state_updates", {})
@@ -114,7 +119,7 @@ def extract_json_from_llm_response(raw_response: str) -> dict:
         
         return parsed
     else:
-        logger.warning("Parsed JSON missing 'assistant_reply' or invalid format. Returning default.")
+        logger.warning("Parsed JSON missing required keys or invalid format. Returning default.")
         return default
 
 def detect_language_from_text(text: str) -> Optional[str]:
@@ -146,18 +151,17 @@ def chat():
     chat_history: List[Dict[str, str]] = data.get("chat_history") or []
     assistant_state: AssistantState = data.get("assistant_state") or {}
 
-    # Initialize/Clean up state
+    # Initialize state from input. This is the "safe" state.
     state: AssistantState = {
         "conversationSummary": assistant_state.get("conversationSummary", ""),
         "language": assistant_state.get("language", "Python"),
         "taggedReplies": assistant_state.get("taggedReplies", []),
     }
     
-    # 1. Prepare LLM Messages from Full History
+    # 1. Prepare LLM Messages from Full History (same as before)
     llm_messages = [{"role": "system", "content": PROGRAMMING_ASSISTANT_PROMPT}]
     
     last_user_message = ""
-    
     for msg in chat_history:
         role = msg.get("role")
         content = msg.get("content")
@@ -166,15 +170,14 @@ def chat():
             if role == "user":
                 last_user_message = content
 
-    # 2. Language Detection & State Update
+    # 2. Language Detection & State Update (same as before)
     detected_lang = detect_language_from_text(last_user_message)
     if detected_lang and detected_lang.lower() != state["language"].lower():
         logger.info("Detected new language: %s", detected_lang)
         state["language"] = detected_lang
         
-    # 3. Inject Contextual Hint and State into the LAST user message
+    # 3. Inject Contextual Hint and State into the LAST user message (same as before)
     context_hint = f"Current Language: {state['language']}. Conversation Summary so far: {state['conversationSummary']}"
-    
     if llm_messages and llm_messages[-1]["role"] == "user":
         llm_messages[-1]["content"] = f"USER MESSAGE: {last_user_message}\n\n[CONTEXT HINT: {context_hint}]"
     elif last_user_message:
@@ -196,22 +199,36 @@ def chat():
              error_detail = "LLM Model Error: The model is likely decommissioned. Please check the 'LLM_MODEL' environment variable or the default model in app.py."
         return jsonify({"error": "LLM invocation failed", "detail": error_detail}), 500
 
-    # 4. State Update from LLM
+    # 4. State Update from LLM (NEW ROBUST LOGIC)
+    
+    # Check if parsing failed (by comparing the reply to the known error message)
+    if parsed_result.get("assistant_reply") == LLM_PARSE_ERROR_MESSAGE:
+        final_reply_content = LLM_PARSE_ERROR_MESSAGE
+        
+        # State and tags remain as initialized (from the input assistant_state), fulfilling the user request.
+        response_payload = {
+            "assistant_reply": final_reply_content,
+            "updated_state": state, # Keep the original input state
+            "suggested_tags": [],
+        }
+        return jsonify(response_payload)
+
+    # Parsing was successful. Safely update the state.
     updated_state_from_llm = parsed_result.get("state_updates", {})
     
     if 'conversationSummary' in updated_state_from_llm:
         state["conversationSummary"] = updated_state_from_llm["conversationSummary"]
-    if 'language' in updated_state_from_llm:
+    # We allow the language to be updated only if it's explicitly set by the LLM AND it's a valid change.
+    if 'language' in updated_state_from_llm and updated_state_from_llm['language'].strip():
         state["language"] = updated_state_from_llm["language"]
 
+
     assistant_reply = parsed_result.get("assistant_reply")
     code_snippet = parsed_result.get("code_snippet")
     
     # 5. Final Response Payload: Combine the reply and the code snippet
-    # The frontend is expecting the code to be *in* the assistant_reply, so we stitch it back together.
     final_reply_content = assistant_reply
     if code_snippet and code_snippet.strip():
-        # Add a newline for clean separation if the reply isn't just whitespace
         if final_reply_content.strip():
             final_reply_content += "\n\n"
         final_reply_content += code_snippet
@@ -220,7 +237,7 @@ def chat():
         final_reply_content = "I'm here to help with your code! What programming language are you using?"
 
     response_payload = {
-        "assistant_reply": final_reply_content, # Send combined reply + code
+        "assistant_reply": final_reply_content, 
         "updated_state": state,
         "suggested_tags": parsed_result.get("suggested_tags", []),
     }