app.py

from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from functools import lru_cache
from transformers import TapexTokenizer, BartForConditionalGeneration
from deep_translator import GoogleTranslator
from pathlib import Path
import os, json, pandas as pd, torch

# ------------------------
# Config
# ------------------------
HF_MODEL_ID  = os.getenv("HF_MODEL_ID",  "stvnnnnnn/tapex-wikisql-best")
SPLIT        = os.getenv("TABLE_SPLIT",  "validation")   # "validation" ~ "dev"
INDEX        = int(os.getenv("TABLE_INDEX", "10"))
MAX_ROWS     = int(os.getenv("MAX_ROWS",    "12"))

# ------------------------
# App
# ------------------------
app = FastAPI(title="NL→SQL – TAPEX + WikiSQL (API)")
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"], allow_methods=["*"], allow_headers=["*"], allow_credentials=True,
)

class NLQuery(BaseModel):
    nl_query: str

# ------------------------
# Modelo
# ------------------------
tok = TapexTokenizer.from_pretrained(HF_MODEL_ID)
model = BartForConditionalGeneration.from_pretrained(HF_MODEL_ID)
if torch.cuda.is_available():
    model = model.to("cuda")

# ------------------------
# Utilidades de carga robustas
# ------------------------
def _read_json_or_jsonl(p: Path) -> dict:
    """
    Lee un JSON normal (.json) o un JSONL (.jsonl) y devuelve el primer objeto.
    """
    txt = p.read_text(encoding="utf-8").strip()
    if p.suffix.lower() == ".jsonl":
        for line in txt.splitlines():
            s = line.strip()
            if s:
                return json.loads(s)
        raise ValueError(f"{p} está vacío.")
    return json.loads(txt)

@lru_cache(maxsize=32)
def get_table(split: str, index: int, max_rows: int) -> pd.DataFrame:
    """
    1) Intenta cargar ./data/<split>.json o ./data/<split>.jsonl (mapeando 'validation'->'dev').
    2) Si no existe, hace fallback a un ejemplo de WikiSQL (conversión Parquet oficial).
    """
    base_dir = Path(__file__).parent
    data_dir = base_dir / "data"

    # Normalizamos nombre local (para demo usamos 'dev')
    local_name = "dev" if split.lower() in ("validation", "dev") else split.lower()

    # 1) Buscar archivo local
    for candidate in (data_dir / f"{local_name}.json", data_dir / f"{local_name}.jsonl"):
        if candidate.exists():
            js = _read_json_or_jsonl(candidate)
            header = [str(h) for h in js["header"]]
            rows   = js["rows"][:max_rows]
            df = pd.DataFrame(rows, columns=header)
            df.columns = [str(c) for c in df.columns]
            return df

    # 2) Fallback: cargar un ejemplo del dataset WikiSQL (Parquet convertido)
    try:
        from datasets import load_dataset  # import diferido para arrancar más rápido
        ds = load_dataset("Salesforce/wikisql", split="validation", revision="refs/convert/parquet")
        if not (0 <= index < len(ds)):
            index = 0  # seguridad
        ex = ds[index]
        header = [str(h) for h in ex["table"]["header"]]
        rows   = ex["table"]["rows"][:max_rows]
        df = pd.DataFrame(rows, columns=header)
        df.columns = [str(c) for c in df.columns]
        return df
    except Exception as e:
        raise RuntimeError(f"No se pudo obtener una tabla: {e}")

# ------------------------
# Endpoints
# ------------------------
@app.get("/api/health")
def health():
    return {"ok": True, "model": HF_MODEL_ID, "split": SPLIT, "index": INDEX}

@app.get("/api/preview")
def preview():
    try:
        df = get_table(SPLIT, INDEX, MAX_ROWS)
        return {"columns": df.columns.tolist(), "rows": df.head(8).to_dict(orient="records")}
    except Exception as e:
        return {"error": str(e)}

@app.post("/api/nl2sql")
def nl2sql(q: NLQuery):
    try:
        text = (q.nl_query or "").strip()
        if not text:
            raise ValueError("Consulta vacía.")

        # Detectar si parece SQL
        lower = text.lower().strip()
        looks_like_sql = lower.startswith(("select", "with", "insert", "update", "delete", "create", "drop", "alter"))

        # Traducir a inglés si no es SQL
        query_en = text
        if not looks_like_sql:
            try:
                translated = GoogleTranslator(source="auto", target="en").translate(text)
                if translated:
                    query_en = translated
            except Exception:
                query_en = text  # fallback seguro

        # Procesar con TAPEX
        df = get_table(SPLIT, INDEX, MAX_ROWS)
        enc = tok(table=df, query=query_en, return_tensors="pt", truncation=True)
        if torch.cuda.is_available():
            enc = {k: v.to("cuda") for k, v in enc.items()}
        out = model.generate(**enc, max_length=160, num_beams=1)
        sql = tok.batch_decode(out, skip_special_tokens=True)[0]

        return {
            "consulta_original": text,
            "consulta_traducida": query_en,
            "sql_generado": sql
        }

    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))