from __future__ import annotations

import gzip
import hashlib
import json
from collections import Counter, defaultdict
from pathlib import Path


ROOT = Path(__file__).resolve().parent


def sha256_file(path: Path) -> str:
    digest = hashlib.sha256()
    with path.open("rb") as handle:
        for chunk in iter(lambda: handle.read(4 * 1024 * 1024), b""):
            digest.update(chunk)
    return digest.hexdigest()


def records(paths: list[Path]):
    for path in paths:
        with gzip.open(path, "rt", encoding="utf-8") as handle:
            for line in handle:
                yield json.loads(line)


def main() -> None:
    checked = 0
    for line in (ROOT / "SHA256SUMS").read_text(encoding="ascii").splitlines():
        expected, relative = line.split("  ", 1)
        path = ROOT / relative
        if sha256_file(path) != expected:
            raise RuntimeError(f"checksum mismatch: {relative}")
        checked += 1

    supervised = list(records([ROOT / "data" / "supervised_model_input.jsonl.gz"]))
    if len(supervised) != 168:
        raise RuntimeError("supervised row count mismatch")
    if Counter(row["label"] for row in supervised) != Counter({0: 112, 1: 56}):
        raise RuntimeError("supervised labels mismatch")
    if Counter(row["outer_fold"] for row in supervised) != Counter(
        {0: 39, 1: 30, 2: 45, 3: 27, 4: 27}
    ):
        raise RuntimeError("supervised fold counts mismatch")
    by_match = defaultdict(list)
    for row in supervised:
        by_match[row["matched_set_id"]].append(row)
        if len(row["sequence_ref"]) != 3000 or len(row["sequence_alt"]) != 3000:
            raise RuntimeError("supervised sequence length mismatch")
        differences = sum(a != b for a, b in zip(row["sequence_ref"], row["sequence_alt"]))
        if differences != 1 or row["sequence_ref"][1500] != row["ref"] or row["sequence_alt"][1500] != row["alt"]:
            raise RuntimeError("supervised centered mutation mismatch")
    if len(by_match) != 56 or any(len(rows) != 3 or sum(row["label"] for row in rows) != 1 for rows in by_match.values()):
        raise RuntimeError("supervised matched triplets mismatch")

    pseudo_paths = sorted((ROOT / "data").glob("pretraining-*.jsonl.gz"))
    pseudo_count = 0
    pseudo_ids = set()
    for row in records(pseudo_paths):
        pseudo_count += 1
        pseudo_ids.add(row["pseudo_index"])
        if len(row["sequence_ref"]) != 3000 or len(row["sequence_alt"]) != 3000:
            raise RuntimeError("pretraining sequence length mismatch")
        if row["sequence_ref"][1500] != row["ref"] or row["sequence_alt"][1500] != row["alt"]:
            raise RuntimeError("pretraining centered mutation mismatch")
        if sum(a != b for a, b in zip(row["sequence_ref"], row["sequence_alt"])) != 1:
            raise RuntimeError("pretraining mutation count mismatch")
    if pseudo_count != 20_000 or pseudo_ids != set(range(20_000)):
        raise RuntimeError("pretraining inventory mismatch")
    print(f"PASS: {checked} checksums; 168 supervised rows; 20,000 pretraining rows")


if __name__ == "__main__":
    main()