from pathlib import Path

from tht.cli.schema_cmd import physical_path


def _extract_lsh_values(dwh, physical, annotations, limit, eligibility_cfg=None):
    from tht.db.sampling import SkippedColumn, TruncatedColumn, is_text_type
    from tht.mschema.eligibility import effective_eligibility

    values, skipped, truncated = {}, [], []
    for table_name, table in physical.tables.items():
        table_ann = annotations.tables.get(table_name)
        for column_name, column in table.columns.items():
            ann_col = table_ann.columns.get(column_name) if table_ann else None
            from_catalog = column.eligibility_reason in {"catalog", "sensitive"}
            if not is_text_type(column.type):
                continue
            if from_catalog and column.eligibility_reason == "sensitive":
                continue
            if from_catalog and eligibility_cfg is not None and (
                column_name.lower() in {name.lower() for name in eligibility_cfg.ignore_columns}
            ):
                continue
            if not from_catalog and not effective_eligibility(column, ann_col)[0]:
                continue
            try:
                distinct = dwh.distinct_values(table_name, column_name, limit=limit)
            except Exception as exc:  # noqa: BLE001 - an unreadable DWH column is non-fatal
                skipped.append(SkippedColumn(table_name, column_name, f"errore: {exc}"))
                continue
            vals = [str(value) for value in distinct.values if value not in (None, "")]
            if from_catalog and eligibility_cfg is not None:
                from tht.mschema.eligibility import classify_column

                lengths = [len(value) for value in vals]
                eligible, reason = classify_column(
                    column.type,
                    column.is_enum,
                    (sum(lengths) / len(lengths)) if lengths else None,
                    max(lengths) if lengths else None,
                    eligibility_cfg,
                )
                column.eligible = eligible
                column.eligibility_reason = reason
                if not eligible:
                    continue
            if vals:
                values.setdefault(table_name, {})[column_name] = vals
            if distinct.truncated:
                truncated.append(TruncatedColumn(table_name, column_name, len(vals)))
    return values, skipped, truncated


def build_lsh_artifacts(
    cfg, *, dwh=None, verbose: bool = False, physical_file: Path | None = None,
    output_dir: Path | None = None,
):
    """Run the existing LSH extraction/build algorithm and persist its outputs."""
    from tht.adapters.factory import build_dwh
    from tht.lshindex import build_index, save_index

    if physical_file is None and cfg.paths.catalog_metadata_snapshot is not None:
        from tht.mschema.context import load_schema_context

        context = load_schema_context(cfg)
        physical, annotations = context.physical, context.annotations
    else:
        from tht.cli.schema_cmd import annotations_path
        from tht.mschema.models import Annotations, PhysicalSchema

        phys_file = physical_file or physical_path(cfg)
        if not phys_file.exists():
            raise FileNotFoundError("physical catalog is missing; run schema introspect first")
        physical = PhysicalSchema.from_yaml(phys_file)
        annotations = Annotations.from_yaml(annotations_path(cfg))
    target = dwh if dwh is not None else build_dwh(cfg)
    values, skipped, truncated = _extract_lsh_values(
        target, physical, annotations, cfg.lsh.max_values_per_column, cfg.eligibility
    )
    lsh, minhashes = build_index(values, cfg.lsh, verbose=verbose)
    save_index(
        lsh, minhashes, cfg.lsh, output_dir or (cfg.paths.indexes / "lsh"),
        name=cfg.database.db_schema,
    )
    return minhashes, skipped, truncated, values