"""Prepare explicit, hashed data for a new Sudoku training workflow."""

from pathlib import Path

from omegaconf import DictConfig, OmegaConf
from softasp_core.config.schemas import DatasetConfig
from softasp_domain.sudoku.downloader import SudokuDownloader
from softasp_domain.sudoku.preprocessor import SudokuPreprocessor

from softasp.sudoku_data import select_hdf5
from softasp.sudoku_reporting import file_sha256


def prepare_scratch_data(cfg: DictConfig) -> Path:
    """Generate in a fresh directory, or validate an explicitly pinned existing file."""
    if cfg.pipeline.data_mode == "existing":
        return select_hdf5(cfg)
    # Generation requires the actual uniqueness checker, never its legacy fallback.
    import clingo  # noqa: F401
    from torchvision.datasets import MNIST

    raw_dir = (
        (
            Path(cfg.paths.raw_data_root)
            if cfg.paths.get("raw_data_root")
            else Path(cfg.paths.data_root) / "raw" / "sudoku"
        )
        .expanduser()
        .resolve()
    )
    if cfg.pipeline.download:
        SudokuDownloader(raw_dir).run()
    # Check complete data rather than trusting the downloader's coarse cache check.
    for train in (True, False):
        MNIST(root=str(raw_dir / "mnist"), train=train, download=False)
    destination = (
        Path(cfg.paths.data_root).expanduser().resolve()
        / "processed"
        / "sudoku"
        / "from_scratch"
        / cfg.pipeline.name
    )
    destination.mkdir(parents=True, exist_ok=False)
    dataset = DatasetConfig(**cfg.dataset)
    path = SudokuPreprocessor(raw_dir, destination, cfg.preprocessing, dataset).run()
    OmegaConf.update(
        cfg,
        "data_artifact",
        {
            "path": str(path.resolve()),
            "sha256": file_sha256(path),
            # A newly generated file is not the archived legacy artifact.
            "protocol_id": None,
        },
    )
    return select_hdf5(cfg)