api.py

import os
import re
import tempfile
from fastapi import FastAPI, UploadFile, File, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from google import genai

from pliego import pliego_pdf_to_markdown
from rup import rup_pdf_to_markdown
from main import generate_analysis

app = FastAPI()

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],  # Reemplaza "*" con tu URL de Vercel en producción
    allow_methods=["POST"],
    allow_headers=["*"],
)

client = genai.Client(api_key=os.environ["GOOGLE_API_KEY"])


def parse_markdown_analysis(text: str) -> dict:
    """
    Parsea la respuesta Markdown de Gemini en datos estructurados
    compatibles con el modelo de datos del frontend.
    """
    gap_rows = []
    lines = text.split("\n")
    in_table = False

    for line in lines:
        stripped = line.strip()
        if "|" not in stripped:
            continue
        # Omitir líneas separadoras como |---|---|
        if re.match(r"^\|[\s\-:|]+\|$", stripped):
            continue

        cells = [c.strip() for c in stripped.split("|")]
        cells = [c for c in cells if c]  # quitar vacíos de | inicial/final

        if len(cells) < 4:
            continue

        # Detectar cabecera de la tabla
        if "requisito" in cells[0].lower():
            in_table = True
            continue

        if not in_table:
            continue

        cumple_cell = cells[3] if len(cells) > 3 else ""
        note = cells[4] if len(cells) > 4 else ""

        if "✅" in cumple_cell:
            status = "pass"
        elif "❌" in cumple_cell:
            status = "fail"
        else:
            status = "warning"

        req_lower = cells[0].lower()
        if any(w in req_lower for w in ["liquidez", "endeudamiento", "capital", "patrimonio", "financiero", "razón", "cobertura"]):
            category = "Financiero"
        elif any(w in req_lower for w in ["unspsc", "código", "clasificador"]):
            category = "UNSPSC"
        elif any(w in req_lower for w in ["experiencia", "contrato", "smmlv", "k de"]):
            category = "Experiencia"
        else:
            category = "Habilitante"

        gap_rows.append({
            "requirement": cells[0],
            "category": category,
            "required": cells[1],
            "yours": cells[2],
            "status": status,
            "note": note,
        })

    # Detectar si se recomienda Consorcio o Unión Temporal
    lower_text = text.lower()
    consortium_needed = "consorcio" in lower_text or "unión temporal" in lower_text
    consortium_reason = ""
    if consortium_needed:
        for line in lines:
            if "consorcio" in line.lower() or "unión temporal" in line.lower():
                clean = re.sub(r"[#*`>]", "", line).strip()
                if len(clean) > 20:
                    consortium_reason = clean
                    break

    # Estado global de la licitación
    fail_count = sum(1 for r in gap_rows if r["status"] == "fail")
    if fail_count > 0:
        overall_status = "no-viable"
    elif all(r["status"] == "pass" for r in gap_rows) and gap_rows:
        overall_status = "viable"
    else:
        overall_status = "revision"

    return {
        "status": overall_status,
        "gap_analysis": gap_rows,
        "risks": [],
        "consortium_data": {
            "needed": consortium_needed,
            "reason": consortium_reason,
        },
    }


@app.post("/analizar")
async def analizar(
    pliego: UploadFile = File(...),
    rup: UploadFile = File(...)
):
    # 1. Guardar PDFs en archivos temporales
    with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp_pliego:
        tmp_pliego.write(await pliego.read())
        pliego_path = tmp_pliego.name

    with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp_rup:
        tmp_rup.write(await rup.read())
        rup_path = tmp_rup.name

    try:
        # 2. Convertir PDFs a texto
        pliego_text = pliego_pdf_to_markdown(pliego_path, os.devnull)
        rup_text = rup_pdf_to_markdown(rup_path, os.devnull)

        # 3. Correr el análisis con Gemini
        response = generate_analysis(client, pliego_text, rup_text)

        if not response.text:
            raise HTTPException(status_code=500, detail="El modelo no devolvió respuesta.")

        # 4. Parsear respuesta Markdown a datos estructurados
        structured = parse_markdown_analysis(response.text)

        return {
            "resultado": response.text,
            **structured,
        }

    finally:
        # 5. Limpiar archivos temporales siempre, incluso si hay error
        os.unlink(pliego_path)
        os.unlink(rup_path)