import os, subprocess, sys, zipfile

os.environ["HF_HUB_DISABLE_SYMLINKS_WARNING"] = "1"
os.environ["HF_HUB_DISABLE_HF_TRANSFER"] = "1"
os.environ["HF_HUB_ENABLE_XET"] = "0"

os.environ["NUMBA_CACHE_DIR"] = "/tmp/numba_cache"
os.makedirs("/tmp/numba_cache", exist_ok=True)
os.environ["NUMBA_DISABLE_JIT"] = "1"

from huggingface_hub import HfApi, HfFolder, upload_folder, snapshot_download

# 🔒 Eliminar hf_transfer si está presente
subprocess.run([sys.executable, "-m", "pip", "uninstall", "-y", "hf_transfer"])

# === Configuración ===
HF_MODEL_ID = "tu_usuario/xtts-v2-finetuned"   # <--- cambia con tu repo en HF
HF_TOKEN = os.environ.get("HF_TOKEN")          # Debe estar definido en tu Space/entorno
DATASET_PATH = "/tmp/dataset"        # Ruta a tu dataset
OUTPUT_PATH = "/tmp/output_model"
BASE_MODEL = "coqui/XTTS-v2"

os.makedirs("/tmp/xtts_cache", exist_ok=True)
os.chmod("/tmp/xtts_cache", 0o777)

os.makedirs("/tmp/xtts_model", exist_ok=True)
os.chmod("/tmp/xtts_model", 0o777)

os.makedirs("/tmp/xtts_model/.huggingface", exist_ok=True)
os.chmod("/tmp/xtts_model/.huggingface", 0o777)

# Continúa con tu lógica, usando las nuevas rutas de manera consistent

# 🔧 Forzar descarga sin symlinks ni hf_transfer
model_dir = snapshot_download(
    repo_id="coqui/XTTS-v2",
    local_dir="/tmp/xtts_model",   # descarga directa aquí
    cache_dir="/tmp/hf_cache",     # cache seguro en /tmp
    #local_dir_use_symlinks=False,  # 🔑 evita enlaces simbólicos
    resume_download=True,
    token=HF_TOKEN
)

print(f"✅ Modelo descargado en: {model_dir}")

CONFIG_PATH = "/tmp/xtts_model/config.json"
RESTORE_PATH = "/tmp/xtts_model/model.pth"

# === 1.B Extraer el dataset
def extract_zip(zip_file_path, destination_path):
    """
    Extracts the contents of a ZIP file to a specified directory.
    
    Args:
        zip_file_path (str): The full path to the ZIP file.
        destination_path (str): The directory where the contents will be extracted.
    """
    # Create the destination directory if it doesn't exist
    os.makedirs(destination_path, exist_ok=True)
    
    try:
        # Open the ZIP file in read mode
        with zipfile.ZipFile(zip_file_path, 'r') as zip_ref:
            # Extract all the contents to the specified directory
            zip_ref.extractall(destination_path)
        print(f"✅ Extracted '{zip_file_path}' to '{destination_path}' successfully.")
    except zipfile.BadZipFile:
        print(f"❌ Error: The file '{zip_file_path}' is not a valid ZIP file.")
    except FileNotFoundError:
        print(f"❌ Error: The file '{zip_file_path}' was not found.")
    except Exception as e:
        print(f"❌ An unexpected error occurred: {e}")

# Example usage:
zip_file = "/home/user/app/voxpopuli_es_500.zip"
dataset_folder = "/tmp/dataset"

# To protect against security vulnerabilities, it is important to sanitize the destination path.
# This prevents an attacker from using a malicious ZIP file to write outside the destination folder.
safe_destination = os.path.abspath(dataset_folder)

# Call the function
extract_zip(zip_file, safe_destination)

# === 2. Editar configuración para tu dataset VoxPopuli ===
print("=== Editando configuración para fine-tuning con VoxPopuli ===")
import json
with open(CONFIG_PATH, "r") as f:
    config = json.load(f)

config["output_path"] = OUTPUT_PATH
config["datasets"] = [
    {
        "formatter": "voxpopuli",
        "path": DATASET_PATH,
        "meta_file_train": "metadata.json"
    }
]
config["run_name"] = "xtts-finetune-voxpopuli"
config["lr"] = 1e-5  # más bajo para fine-tuning

with open(CONFIG_PATH, "w") as f:
    json.dump(config, f, indent=2)

# === 3. Lanzar entrenamiento ===
print("=== Iniciando fine-tuning de XTTS-v2 ===")

import librosa
from librosa.core import spectrum

subprocess.run([
   "python", "/home/user/TTS/TTS/bin/train_tts.py",
   "--config_path", CONFIG_PATH,
   "--restore_path", RESTORE_PATH
], check=True)

# subprocess.run([
#    "python", "-m", "TTS.bin.train",
#    "--config_path", CONFIG_PATH,
#    "--restore_path", RESTORE_PATH
# ], check=True)

# === 4. Subir modelo resultante a HF ===
print("=== Subiendo modelo fine-tuneado a Hugging Face Hub ===")
api = HfApi()
HfFolder.save_token(HF_TOKEN)

upload_folder(
    repo_id=HF_MODEL_ID,
    repo_type="model",
    folder_path=OUTPUT_PATH,
    token=HF_TOKEN
)

print("✅ Fine-tuning completado y modelo subido a Hugging Face.")