"""Use the imported model snapshot; local names alone do not establish identity."""
import hashlib
import json
from pathlib import Path

import numpy as np
import pandas as pd

from . import get_model, list_available_models
from .model_data import ModelData
from .anos import detectar_ano_base, periodo_variavel_ano


def context_key(row):
    terrain = row.get('evolutivo', {}).get('terreno', {})
    context = terrain.get('contextoModelo') or {}
    return row.get('modelo', ''), hashlib.sha256(json.dumps(context, sort_keys=True, ensure_ascii=False).encode()).hexdigest()[:12]


def matching_local(row):
    name = row.get('modelo')
    context = row.get('evolutivo', {}).get('terreno', {}).get('contextoModelo') or {}
    if name not in list_available_models():
        return None, 'Modelo indisponível no catálogo.'
    local = get_model(name)
    if not context:
        return None, 'O arquivo não preserva o contexto necessário para confirmar a versão do modelo do catálogo.'
    prediction = context.get('prediction') or {}
    names, params = prediction.get('names') or [], prediction.get('params') or []
    sm = local.modelos_sm if local else None
    try:
        local_names = list(map(str, sm.model.exog_names))
        same = names == local_names and len(params) == len(sm.params) and np.allclose(params, sm.params, rtol=1e-12, atol=1e-12)
        # Equal parameters are necessary, but not sufficient: verify fitted data too.
        x = context.get('transformacoes', {}).get('X')
        y = context.get('transformacoes', {}).get('y')
        portable_x = pd.DataFrame(x).reindex(columns=names, fill_value=1).to_numpy(dtype=float)
        portable_y = np.asarray(y, dtype=float).reshape(-1)
        same = same and portable_x.shape == sm.model.exog.shape and portable_y.shape == sm.model.endog.shape
        same = same and np.allclose(portable_x, sm.model.exog, rtol=1e-12, atol=1e-12) and np.allclose(portable_y, sm.model.endog, rtol=1e-12, atol=1e-12)
        covariance = np.asarray(prediction.get('covariance'),dtype=float)
        same = same and covariance.shape == np.asarray(sm.cov_params()).shape and np.allclose(covariance,sm.cov_params(),rtol=1e-12,atol=1e-12)
        same = same and context.get('transformacoes',{}).get('info') == local.formatted_top_transformation_info
        if same:
            return local, ''
    except (AttributeError, TypeError, ValueError, KeyError):
        pass
    return None, 'A correspondência da versão do catálogo não foi confirmada com os coeficientes e a amostra salvos; os anexos desse modelo não serão usados.'


def snapshot_model(row):
    context = row.get('evolutivo', {}).get('terreno', {}).get('contextoModelo') or {}
    prediction = context.get('prediction') or {}
    names, params = prediction.get('names') or [], prediction.get('params') or []
    if not names or len(names) != len(params):
        return None
    raw = context.get('dados', {}).get('df') or {}
    try:
        df = pd.DataFrame(**raw) if isinstance(raw, dict) and 'columns' in raw else pd.DataFrame(raw)
    except (ValueError, TypeError):
        df = pd.DataFrame()
    transforms = context.get('transformacoes') or {}
    info = transforms.get('info') or []
    mapping = dict(item.split(': ', 1) for item in info if isinstance(item, str) and ': ' in item)
    # Transcribe the fitted expression; do not fit or evaluate a new model.
    terms = []
    for name, param in zip(names, params):
        coefficient = format(float(param), '.12g')
        if name == 'const':
            terms.append(coefficient)
        else:
            transformed = mapping.get(name, '(x)').replace('x', f'[{name}]')
            terms.append(f'({coefficient} × {transformed})')
    dependent = next((item.split(': ', 1) for item in info if isinstance(item, str) and ': ' in item and item.split(': ', 1)[0] not in names), ['Y', 'y'])
    lhs = dependent[1].replace('y', dependent[0])
    evidence = context.get('normativeEvidence') or {}
    annual_names = [str(name) for name in names if periodo_variavel_ano(str(name))]
    reference = context.get('referencia_temporal')
    base_year = detectar_ano_base(df, annual_names, reference)
    if base_year is None and reference is None:
        # Legacy EVO contexts omitted the declaration. A name alone is not
        # enough to borrow it: matching_local verifies the fitted model/data.
        local, _ = matching_local(row)
        if local is not None:
            base_year = local.ano_base
    return ModelData(
        nome=row.get('modelo') or context.get('name') or 'Modelo importado', path=Path('contexto_importado'),
        xy_preview=df, tabelas_coef=pd.DataFrame({'coef': params}, index=names),
        modelos_resumos={'n': evidence.get('n', len(df)), 'k': evidence.get('k', len(names)-1),
                         'r2': evidence.get('r2'), 'equacao': f'{lhs} = ' + ' + '.join(terms)},
        formatted_top_transformation_info=info,
        variaveis_dicotomicas=transforms.get('dicotomicas') or [],
        periodo_dados_mercado=context.get('periodo_dados_mercado') or {},
        ano_base=base_year,
    )