# Author: Simon-Pierre Boucher — simon-pierre.boucher@uqo.ca
"""Shared fixtures. Tests run on the SYNTHETIC table (built in memory from the generator, no private data).
``pytest --real`` additionally runs the tests marked ``real`` on data/processed/*."""
from __future__ import annotations

import polars as pl
import pytest

from agentic_valuation.config import PROCESSED, SYNTHETIC, load_config
from agentic_valuation.data.build import build_from_synthetic
from agentic_valuation.data.splits import assign_splits, stratified_subsample
from agentic_valuation.data.synthetic import generate
from agentic_valuation.features.matrix import build_feature_matrix

SYN_N_BY_SPLIT = {"validation": 40, "test": 60, "holdout": 30}


def pytest_addoption(parser: pytest.Parser) -> None:
    parser.addoption("--real", action="store_true", default=False, help="also run checks on the real processed data")


def pytest_configure(config: pytest.Config) -> None:
    config.addinivalue_line("markers", "real: needs data/processed (private); enabled with --real")


def pytest_collection_modifyitems(config: pytest.Config, items: list[pytest.Item]) -> None:
    if config.getoption("--real"):
        return
    skip = pytest.mark.skip(reason="needs --real and the private processed data")
    for item in items:
        if "real" in item.keywords:
            item.add_marker(skip)


@pytest.fixture(scope="session")
def cfg() -> dict:
    return load_config("data")


@pytest.fixture(scope="session")
def syn_analytic() -> pl.DataFrame:
    src = SYNTHETIC / "synthetic_transactions.parquet"
    raw = pl.read_parquet(src) if src.exists() else generate(20_000)
    df, _ = build_from_synthetic(raw)
    return df


@pytest.fixture(scope="session")
def syn_splits(syn_analytic: pl.DataFrame, cfg: dict) -> pl.DataFrame:
    return assign_splits(syn_analytic, cfg)


@pytest.fixture(scope="session")
def syn_subsamples(syn_analytic: pl.DataFrame, syn_splits: pl.DataFrame, cfg: dict) -> pl.DataFrame:
    return stratified_subsample(syn_analytic, syn_splits, cfg, n_by_split=SYN_N_BY_SPLIT)


@pytest.fixture(scope="session")
def syn_matrix(syn_analytic: pl.DataFrame, syn_splits: pl.DataFrame) -> pl.DataFrame:
    return build_feature_matrix(syn_analytic, syn_splits, geo=None, comps_summary=True)


@pytest.fixture(scope="session")
def real_bundle() -> dict[str, pl.DataFrame]:
    paths = {k: PROCESSED / f"{k}.parquet" for k in ("analytic", "splits", "llm_subsamples", "feature_matrix", "geo_features")}
    if not paths["analytic"].exists():
        pytest.skip("data/processed/analytic.parquet missing")
    return {k: pl.read_parquet(p) for k, p in paths.items() if p.exists()}