chat_handler.py

"""
Flare – Chat Handler  (v1.7 · parameter parsing düzeltmesi)
==========================================
"""

import os
import re, json, sys, httpx
from datetime import datetime
from typing import Dict, List, Optional
from fastapi import APIRouter, HTTPException, Header
from pydantic import BaseModel
import requests

from prompt_builder import build_intent_prompt, build_parameter_prompt, build_smart_parameter_question_prompt, extract_params_from_question
from utils import log
from api_executor import call_api as execute_api  
from validation_engine import validate
from session import session_store, Session
from llm_interface import LLMInterface, SparkLLM, GPT4oLLM

# ─────────────────────────  CONFIG ───────────────────────── #
# Lazy loading for config
_cfg = None

def get_config():
    """Get or reload config"""
    global _cfg
    if _cfg is None:
        from config_provider import ConfigProvider
        _cfg = ConfigProvider.get()
    return _cfg

def reload_config():
    """Force reload config"""
    global _cfg
    from config_provider import ConfigProvider
    ConfigProvider._instance = None
    _cfg = ConfigProvider.get()
    return _cfg

# Global LLM instance
llm_provider: Optional[LLMInterface] = None

# ─────────────────────────  HELPERS ───────────────────────── #
def _trim_response(raw: str) -> str:
    """
    Remove everything after the first logical assistant block or intent tag.
    Also strips trailing 'assistant' artifacts and prompt injections.
    """
    # Stop at our own rules if model leaked them
    for stop in ["#DETECTED_INTENT", "⚠️", "\nassistant", "assistant\n", "assistant"]:
        idx = raw.find(stop)
        if idx != -1:
            raw = raw[:idx]
    # Normalise selamlama
    raw = re.sub(r"Hoş[\s-]?geldin(iz)?", "Hoş geldiniz", raw, flags=re.IGNORECASE)
    return raw.strip()

def _safe_intent_parse(raw: str) -> tuple[str, str]:
    """Extract intent name and extra tail."""
    m = re.search(r"#DETECTED_INTENT:\s*([A-Za-z0-9_-]+)", raw)
    if not m:
        return "", raw
    name = m.group(1)
    # Remove 'assistant' suffix if exists
    if name.endswith("assistant"):
        name = name[:-9]  # Remove last 9 chars ("assistant")
        log(f"🔧 Removed 'assistant' suffix from intent name")
    tail = raw[m.end():]
    log(f"🎯 Parsed intent: {name}")
    return name, tail
    
# ─────────────────────────  SPARK ───────────────────────── #
def initialize_llm(force_reload=False):
    """Initialize LLM provider based on work_mode"""
    global llm_provider
    
    # Get fresh config if forced or first time
    if force_reload:
        cfg = reload_config()
    else:
        cfg = get_config()
    
    work_mode = cfg.global_config.work_mode
    
    if cfg.global_config.is_gpt_mode():
        # GPT mode
        api_key = cfg.global_config.get_plain_token()
        if not api_key:
            raise ValueError("OpenAI API key not configured")
        
        model = cfg.global_config.get_gpt_model()
        llm_provider = GPT4oLLM(api_key, model)
        log(f"✅ Initialized {model} provider")
    else:
        # Spark mode
        spark_token = _get_spark_token()
        if not spark_token:
            raise ValueError("Spark token not configured")
            
        spark_endpoint = str(cfg.global_config.spark_endpoint)
        llm_provider = SparkLLM(spark_endpoint, spark_token)
        log("✅ Initialized Spark provider")

# ─────────────────────────  SPARK ───────────────────────── #
def _get_spark_token() -> Optional[str]:
    """Get Spark token based on work_mode"""
    cfg = get_config()
    work_mode = cfg.global_config.work_mode
    
    if work_mode in ("hfcloud", "cloud"):
        # Cloud mode - use HuggingFace Secrets
        token = os.getenv("SPARK_TOKEN")
        if not token:
            log("❌ SPARK_TOKEN not found in HuggingFace Secrets!")
        return token
    else:
        # On-premise mode - use .env file
        from dotenv import load_dotenv
        load_dotenv()
        return os.getenv("SPARK_TOKEN")

async def spark_generate(s: Session, prompt: str, user_msg: str) -> str:
    """Call LLM provider with proper error handling"""
    try:
        # Always reinitialize to get fresh config
        initialize_llm(force_reload=True)
        
        if not llm_provider:
            raise ValueError("Failed to initialize LLM provider")
        
        # Use the abstract interface
        raw = await llm_provider.generate(prompt, user_msg, s.chat_history)
        log(f"🪄 LLM raw response: {raw[:120]!r}")
        return raw
        
    except Exception as e:
        log(f"❌ LLM error: {e}")
        raise

# ─────────────────────────  FASTAPI ───────────────────────── #
router = APIRouter()

@router.get("/health")
def health_check():
    """Health check endpoint for monitoring"""
    return {
        "status": "ok",
        "sessions": len(session_store._sessions),
        "timestamp": datetime.now().isoformat()
    }

class StartRequest(BaseModel): 
    project_name: str

class ChatRequest(BaseModel):  
    user_input: str

class ChatResponse(BaseModel):  
    session_id: str
    answer: str

@router.post("/start_session", response_model=ChatResponse)
async def start_session(req: StartRequest):
    """Create new session"""
    try:
        cfg = get_config()

        # Validate project exists
        project = next((p for p in cfg.projects if p.name == req.project_name and p.enabled), None)
        if not project:
            raise HTTPException(404, f"Project '{req.project_name}' not found or disabled")
        
        # Create session
        session = session_store.create_session(req.project_name)
        greeting = "Hoş geldiniz! Size nasıl yardımcı olabilirim?"
        session.add_turn("assistant", greeting)
        
        return ChatResponse(session_id=session.session_id, answer=greeting)
        
    except Exception as e:
        log(f"❌ Error creating session: {e}")
        raise HTTPException(500, str(e))

@router.post("/chat", response_model=ChatResponse)
async def chat(body: ChatRequest, x_session_id: str = Header(...)):
    """Process chat message"""
    try:
        # Get session
        session = session_store.get_session(x_session_id)
        if not session:
            raise HTTPException(404, "Session not found")
        
        user_input = body.user_input.strip()
        if not user_input:
            raise HTTPException(400, "Empty message")
            
        log(f"💬 User input: {user_input}")
        log(f"📊 Session state: {session.state}, last_intent: {session.last_intent}")
        log(f"📊 Session version: {session.version_number}")
        
        session.add_turn("user", user_input)
        
        # Get version config from session
        version = session.get_version_config()
        if not version:
            raise HTTPException(500, "Version configuration lost")

        # Handle based on state
        if session.state == "await_param":
            log(f"🔄 Handling parameter followup for missing: {session.awaiting_parameters}")
            answer = await _handle_parameter_followup(session, user_input)  # version parametresi kaldırıldı
        else:
            log("🆕 Handling new message")
            answer = await _handle_new_message(session, user_input)  # version parametresi kaldırıldı
        
        session.add_turn("assistant", answer)
        return ChatResponse(session_id=session.session_id, answer=answer)
        
    except HTTPException:
        raise
    except Exception as e:
        log(f"❌ Chat error: {e}")
        session.reset_flow()
        error_msg = "Bir hata oluştu. Lütfen tekrar deneyin."
        session.add_turn("assistant", error_msg)
        return ChatResponse(session_id=x_session_id, answer=error_msg)

# ─────────────────────────  MESSAGE HANDLERS ───────────────────────── #
async def _handle_new_message(session: Session, user_input: str) -> str:
    """Handle new message (not parameter followup)"""
    
    # Get version config from session
    version = session.get_version_config()
    if not version:
        log("❌ Version config not found")
        return "Bir hata oluştu. Lütfen tekrar deneyin."
    
    # Build intent detection prompt
    prompt = build_intent_prompt(
        version.general_prompt,
        session.chat_history,
        user_input,
        version.intents,
        session.project_name
    )
    
    # Get LLM response
    raw = await spark_generate(session, prompt, user_input)
    
    # Empty response fallback
    if not raw:
        log("⚠️ Empty response from LLM")
        return "Üzgünüm, mesajınızı anlayamadım. Lütfen tekrar dener misiniz?"
    
    # Check for intent
    if not raw.startswith("#DETECTED_INTENT"):
        # Small talk response
        log("💬 No intent detected, returning small talk")
        return _trim_response(raw)
    
    # Parse intent
    intent_name, tail = _safe_intent_parse(raw)
    
    # Validate intent against version's intents
    valid_intents = {intent.name for intent in version.intents}
    if intent_name not in valid_intents:
        log(f"⚠️ Invalid intent: {intent_name} (valid: {valid_intents})")
        return _trim_response(tail) if tail else "Size nasıl yardımcı olabilirim?"
    
    # Short message guard (less than 3 words usually means incomplete request)
    if len(user_input.split()) < 3 and intent_name != "flight-info":
        log(f"⚠️ Message too short ({len(user_input.split())} words) for intent {intent_name}")
        return _trim_response(tail) if tail else "Lütfen talebinizi biraz daha detaylandırır mısınız?"
    
    # Find intent config
    intent_config = next((i for i in version.intents if i.name == intent_name), None)
    if not intent_config:
        log(f"❌ Intent config not found for: {intent_name}")
        return "Üzgünüm, bu işlemi gerçekleştiremiyorum."
    
    # Set intent in session
    session.last_intent = intent_name
    log(f"✅ Intent set: {intent_name}")
    
    # Log intent parameters
    log(f"📋 Intent parameters: {[p.name for p in intent_config.parameters]}")
    
    # Extract parameters
    return await _extract_parameters(session, intent_config, user_input)

async def _handle_parameter_followup(session: Session, user_input: str, version) -> str:
    """Handle parameter collection followup with smart question generation"""
    if not session.last_intent:
        log("⚠️ No last intent in session")
        session.reset_flow()
        return "Üzgünüm, hangi işlem için bilgi istediğimi unuttum. Baştan başlayalım."
    
    # Get intent config
    intent_config = next((i for i in version.intents if i.name == session.last_intent), None)
    if not intent_config:
        log(f"❌ Intent config not found for: {session.last_intent}")
        session.reset_flow()
        return "Bir hata oluştu. Lütfen tekrar deneyin."
    
    # Try to extract missing parameters
    missing = session.awaiting_parameters
    log(f"🔍 Trying to extract missing params: {missing}")
    
    prompt = build_parameter_prompt(intent_config, missing, user_input, session.chat_history)
    raw = await spark_generate(session, prompt, user_input)
    
    # Try parsing with or without #PARAMETERS: prefix
    success = _process_parameters(session, intent_config, raw)
    
    # Hangi parametreler hala eksik?
    still_missing = _get_missing_parameters(session, intent_config)
    
    # Sorulan ama cevaplanmayan parametreleri belirle
    asked_but_not_answered = []
    for param in session.awaiting_parameters:
        if param in still_missing:
            asked_but_not_answered.append(param)
    
    # Cevaplanmayanları session'a kaydet
    if asked_but_not_answered:
        session.mark_parameters_unanswered(asked_but_not_answered)
        log(f"❓ Parameters not answered: {asked_but_not_answered}")
    
    # Cevaplananları işaretle
    for param in session.awaiting_parameters:
        if param not in still_missing:
            session.mark_parameter_answered(param)
            log(f"✅ Parameter answered: {param}")
    
    if still_missing:
        # Maksimum deneme kontrolü
        if session.missing_ask_count >= 3:
            session.reset_flow()
            return "Üzgünüm, istediğiniz bilgileri anlayamadım. Başka bir konuda yardımcı olabilir miyim?"
        
        session.missing_ask_count += 1
        
        # Akıllı soru üret
        question = await _generate_smart_parameter_question(
            session, intent_config, still_missing, version
        )
        
        # Sorulan parametreleri tahmin et ve kaydet
        params_in_question = extract_params_from_question(question, still_missing, intent_config)
        session.record_parameter_question(params_in_question)
        session.awaiting_parameters = params_in_question
        
        log(f"🤖 Smart question generated for params: {params_in_question}")
        return question
    
    # All parameters collected, call API
    log("✅ All parameters collected, calling API")
    session.state = "call_api"
    return await _execute_api_call(session, intent_config)

# ─────────────────────────  PARAMETER HANDLING ───────────────────────── #
async def _extract_parameters(session: Session, intent_config, user_input: str) -> str:
    """Extract parameters from user input with smart question generation"""
    
    # Yeni intent için parametre takibini sıfırla
    if session.parameter_ask_rounds == 0:
        session.reset_parameter_tracking()
    
    missing = _get_missing_parameters(session, intent_config)
    log(f"🔍 Missing parameters: {missing}")
    
    if not missing:
        # All parameters already available
        log("✅ All parameters already available")
        return await _execute_api_call(session, intent_config)
    
    # Build parameter extraction prompt
    prompt = build_parameter_prompt(intent_config, missing, user_input, session.chat_history)
    raw = await spark_generate(session, prompt, user_input)
    
    # Try processing with flexible parsing
    success = _process_parameters(session, intent_config, raw)
    
    if success:
        missing = _get_missing_parameters(session, intent_config)
        log(f"📊 After extraction, missing: {missing}")
    else:
        log("⚠️ Failed to extract parameters from response")
    
    if missing:
        # Still missing parameters - generate smart question
        session.state = "await_param"
        session.missing_ask_count = 0
        
        # Akıllı soru üret
        question = await _generate_smart_parameter_question(
            session, intent_config, missing, version
        )
        
        # Sorulan parametreleri tahmin et ve kaydet
        params_in_question = extract_params_from_question(question, missing, intent_config)
        session.record_parameter_question(params_in_question)
        session.awaiting_parameters = params_in_question
        
        log(f"🤖 Smart question generated for initial params: {params_in_question}")
        return question
    
    # All parameters collected
    log("✅ All parameters collected after extraction")
    return await _execute_api_call(session, intent_config)

async def _generate_smart_parameter_question(
    session: Session, 
    intent_config,
    missing_params: List[str],
    version
) -> str:
    """LLM kullanarak doğal parametre sorusu üret"""
    
    # Config'i al
    collection_config = cfg.global_config.parameter_collection_config
    
    # Öncelik sıralaması: önce cevaplanmayanlar
    prioritized_params = []
    
    # 1. Daha önce sorulup cevaplanmayanlar
    for param in session.unanswered_parameters:
        if param in missing_params:
            prioritized_params.append(param)
            log(f"🔝 Priority param (unanswered): {param}")
    
    # 2. Hiç sorulmamışlar
    for param in missing_params:
        if param not in prioritized_params:
            prioritized_params.append(param)
            log(f"➕ Additional param (not asked): {param}")
    
    # Maksimum parametre sayısını belirle
    max_params = min(
        len(prioritized_params),
        collection_config.max_params_per_question
    )
    params_to_ask = prioritized_params[:max_params]
    
    log(f"📋 Params to ask in this round: {params_to_ask}")
    
    # Proje dilini belirle
    project_language = "Turkish"  # Default
    if hasattr(version, 'project') and hasattr(version.project, 'default_language'):
        lang_map = {
            "tr": "Turkish", "en": "English", "de": "German",
            "fr": "French", "es": "Spanish"
        }
        project_language = lang_map.get(version.project.default_language, "Turkish")
    
    # Prompt oluştur
    prompt = build_smart_parameter_question_prompt(
        collection_config,
        intent_config,
        params_to_ask,  # Sadece bu turda sorulacak parametreler
        session,
        project_language
    )
    
    # LLM'den soru üret
    response = await spark_generate(session, prompt, "")
    
    # Güvenlik: Eğer LLM boş veya hatalı response verirse fallback
    if not response or len(response.strip()) < 10:
        log("⚠️ Empty or invalid response from LLM, using fallback")
        # En yüksek öncelikli parametre için fallback soru
        param = params_to_ask[0]
        param_config = next(p for p in intent_config.parameters if p.name == param)
        
        # Parametrenin kaç kez sorulduğuna göre farklı fallback
        ask_count = session.get_parameter_ask_count(param)
        if ask_count == 0:
            return f"{param_config.caption} bilgisini alabilir miyim?"
        elif ask_count == 1:
            return f"Lütfen {param_config.caption.lower()} bilgisini paylaşır mısınız?"
        else:
            return f"Devam edebilmem için {param_config.caption.lower()} bilgisine ihtiyacım var."
    
    # Response'u temizle
    clean_response = response.strip()
    
    # Eğer response yanlışlıkla başka şeyler içeriyorsa temizle
    if "#" in clean_response or "assistant:" in clean_response.lower():
        # İlk satırı al
        clean_response = clean_response.split('\n')[0].strip()
    
    log(f"💬 Generated smart question: {clean_response[:100]}...")
    return clean_response

def _get_missing_parameters(session: Session, intent_config) -> List[str]:
    """Get list of missing required parameters"""
    missing = [
        p.name for p in intent_config.parameters 
        if p.required and p.variable_name not in session.variables
    ]
    log(f"📊 Session variables: {list(session.variables.keys())}")
    return missing

def _process_parameters(session: Session, intent_config, raw: str) -> bool:
    """Process parameter extraction response with flexible parsing"""
    try:
        # Try to parse JSON, handling both with and without #PARAMETERS: prefix
        json_str = raw
        if raw.startswith("#PARAMETERS:"):
            json_str = raw[len("#PARAMETERS:"):]
            log(f"🔍 Found #PARAMETERS: prefix, removing it")
        
        # Clean up any trailing content after JSON
        # Find the closing brace for the JSON object
        brace_count = 0
        json_end = -1
        in_string = False
        escape_next = False
        
        for i, char in enumerate(json_str):
            if escape_next:
                escape_next = False
                continue
                
            if char == '\\':
                escape_next = True
                continue
                
            if char == '"' and not escape_next:
                in_string = not in_string
                continue
                
            if not in_string:
                if char == '{':
                    brace_count += 1
                elif char == '}':
                    brace_count -= 1
                    if brace_count == 0:
                        json_end = i + 1
                        break
        
        if json_end > 0:
            json_str = json_str[:json_end]
            log(f"🔍 Cleaned JSON string: {json_str[:200]}")
        
        data = json.loads(json_str)
        
        extracted = data.get("extracted", [])
        log(f"📦 Extracted data: {extracted}")
        
        any_valid = False
        
        for param_data in extracted:
            param_name = param_data.get("name")
            param_value = param_data.get("value")
            
            if not param_name or not param_value:
                log(f"⚠️ Invalid param data: {param_data}")
                continue
            
            # Find parameter config
            param_config = next(
                (p for p in intent_config.parameters if p.name == param_name),
                None
            )
            if not param_config:
                log(f"⚠️ Parameter config not found for: {param_name}")
                continue

            # Date tipi için özel kontrol
            if param_config.type == "date":
                try:
                    # ISO format kontrolü
                    from datetime import datetime
                    datetime.strptime(str(param_value), "%Y-%m-%d")
                except ValueError:
                    log(f"❌ Invalid date format for {param_name}: {param_value}")
                    continue
            
            # Validate parameter
            if validate(str(param_value), param_config):
                session.variables[param_config.variable_name] = str(param_value)
                any_valid = True
                log(f"✅ Extracted {param_name}={param_value} → {param_config.variable_name}")
            else:
                log(f"❌ Invalid {param_name}={param_value}")
        
        return any_valid
        
    except json.JSONDecodeError as e:
        log(f"❌ JSON parsing error: {e}")
        log(f"❌ Failed to parse: {raw[:200]}")
        
        # Fallback: Try to extract simple values from user input
        # This is especially useful for single parameter responses
        if session.state == "await_param" and len(session.awaiting_parameters) > 0:
            # Get the first missing parameter
            first_missing = session.awaiting_parameters[0]
            param_config = next(
                (p for p in intent_config.parameters if p.name == first_missing),
                None
            )
            
            if param_config and session.chat_history:
                # Get the last user input
                last_user_input = session.chat_history[-1].get("content", "").strip()
                
                # For simple inputs like city names, try direct assignment
                if param_config.type in ["str", "string"] and len(last_user_input.split()) <= 3:
                    if validate(last_user_input, param_config):
                        session.variables[param_config.variable_name] = last_user_input
                        log(f"✅ Fallback extraction: {first_missing}={last_user_input}")
                        return True
        
        return False
    except Exception as e:
        log(f"❌ Parameter processing error: {e}")
        return False
        
# ─────────────────────────  API EXECUTION ───────────────────────── #
async def _execute_api_call(session: Session, intent_config) -> str:
    """Execute API call and return humanized response"""
    try:
        session.state = "call_api"
        api_name = intent_config.action

        cfg = get_config()

        api_config = cfg.get_api(api_name)
        
        if not api_config:
            log(f"❌ API config not found: {api_name}")
            session.reset_flow()
            return intent_config.fallback_error_prompt or "İşlem başarısız oldu."
        
        log(f"📡 Calling API: {api_name}")
        log(f"📦 API variables: {session.variables}")
        
        # Execute API call with session
        response = execute_api(api_config, session)
        api_json = response.json()
        log(f"✅ API response: {api_json}")
        
        # Humanize response
        session.state = "humanize"
        if api_config.response_prompt:
            prompt = api_config.response_prompt.replace(
                "{{api_response}}", 
                json.dumps(api_json, ensure_ascii=False)
            )
            human_response = await spark_generate(session, prompt, json.dumps(api_json))
            # Trim response to remove any trailing "assistant" artifacts
            trimmed_response = _trim_response(human_response)
            session.reset_flow()
            return trimmed_response if trimmed_response else f"İşlem sonucu: {api_json}"
        else:
            session.reset_flow()
            return f"İşlem tamamlandı: {api_json}"
            
    except requests.exceptions.Timeout:
        log(f"⏱️ API timeout: {api_name}")
        session.reset_flow()
        return intent_config.fallback_timeout_prompt or "İşlem zaman aşımına uğradı."
    except Exception as e:
        log(f"❌ API call error: {e}")
        session.reset_flow()
        return intent_config.fallback_error_prompt or "İşlem sırasında bir hata oluştu."