# core/fable_engine.py
"""
FableEngine – Optimisation itérative et création automatique de skills.
Utilise BlackBoxAdapter pour interagir avec le LLM.
"""
import asyncio
import logging
import time
from typing import Optional, Dict, Any

from core.skill_schema import Skill, SkillMetadata
from core.vector_memory import VectorMemory

logger = logging.getLogger("vortex.fable_engine")


class FableEngine:
    def __init__(self, llm_engine, memory, error_tree, vector_memory: VectorMemory, blackbox=None):
        """
        llm_engine : conservé pour compatibilité, mais non utilisé directement pour les appels LLM.
        blackbox : instance de BlackBoxAdapter qui expose une méthode async `query(prompt)`.
        """
        self.llm = llm_engine
        self.memory = memory
        self.error_tree = error_tree
        self.vm = vector_memory
        self.blackbox = blackbox

    async def run(self, task: str, max_iter: int = 3, min_score: float = 0.85) -> Dict[str, Any]:
        logger.info(f"[Fable] Début optimisation – tâche : {task[:80]} (max {max_iter} itérations)")
        best_code = None
        best_score = 0.0
        best_skill = None

        if not self.blackbox:
            raise RuntimeError("FableEngine a besoin d'une instance BlackBoxAdapter (blackbox).")

        for i in range(max_iter):
            prompt = self._build_prompt(task, best_code, best_score, i)
            # Appel via BlackBoxAdapter (même interface que le chat)
            response = await self.blackbox.query(prompt)
            # Extraire le texte : réponse peut être un objet avec .content ou une chaîne
            if hasattr(response, 'content'):
                text = response.content
            else:
                text = str(response)
            code = self._extract_code(text)
            if not code:
                continue

            score = await self._evaluate(code, task)
            logger.info(f"[Fable] Itération {i+1}/{max_iter} : score {score:.3f}")
            if score > best_score:
                best_score = score
                best_code = code
                if score >= min_score:
                    skill = self._create_skill(task, best_code, best_score)
                    best_skill = skill
                    logger.info(f"[Fable] Skill sauvegardé : {skill.metadata.name} (score {best_score:.3f})")
            if score >= 0.99:
                break

        return {
            "skill": best_skill.to_dict() if best_skill else {},
            "score": best_score,
            "iterations": i + 1,
            "code": best_code
        }

    def _build_prompt(self, task: str, current_code: Optional[str], current_score: float, iteration: int) -> str:
        if current_code is None:
            return f"Écris du code Python pour résoudre cette tâche :\n{task}\nFournis uniquement le code."
        return (
            f"Tâche : {task}\n"
            f"Code actuel (score {current_score:.3f}) :\n```python\n{current_code}\n```\n"
            f"Améliore ce code (itération {iteration+1}). Fournis uniquement le code amélioré."
        )

    def _extract_code(self, response: str) -> Optional[str]:
        if "```" in response:
            parts = response.split("```")
            for i, part in enumerate(parts):
                if i % 2 == 1:  # contenu entre les backticks
                    if part.startswith("python"):
                        part = part[6:]
                    return part.strip()
        return response.strip()

    async def _evaluate(self, code: str, task: str) -> float:
        """Évaluation heuristique simple (à remplacer par un vrai sandbox si besoin)."""
        score = 0.5
        if len(code) > 50: score += 0.1
        if len(code) > 100: score += 0.1
        if "def " in code: score += 0.1
        if "return " in code: score += 0.1
        if "import " in code: score += 0.05
        # Bonus mots-clés de la tâche
        task_words = set(task.lower().split())
        code_words = set(code.lower().split())
        common = task_words.intersection(code_words)
        score += min(0.1, len(common) * 0.02)
        return min(score, 1.0)

    def _create_skill(self, task: str, code: str, score: float) -> Skill:
        name = f"skill_{int(time.time())}"
        metadata = SkillMetadata(
            name=name,
            description=task[:100],
            category="auto_fable",
            version="1.0",
            tags=["fable", "optimized"]
        )
        skill = Skill(metadata=metadata, code=code, score=score)
        self.vm.add_skill(skill)
        return skill