import os
import posixpath

import datasets
from datasets import DatasetInfo, DownloadManager
from fsspec.core import url_to_fs


_EXTENSION = [".png", ".jpg", ".jpeg", ".webp", ".bmp"]


class DanbooruDataset(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIGS = [
        # add number before name for sorting
        datasets.BuilderConfig(name="full"),
    ]

    def _info(self) -> DatasetInfo:
        features = {
            "image": datasets.Image(),
            "post_id": datasets.Value("int64")
        }
        info = datasets.DatasetInfo(
            features=datasets.Features(features),
            supervised_keys=None,
            citation="",
        )
        return info

    def _split_generators(self, dl_manager: DownloadManager):
        fs, path = url_to_fs(dl_manager._base_path)
        urls = fs.glob(posixpath.join(path, "**/*.tar"), detail=False)
        archives = dl_manager.download(["hf://" + url for url in urls])
        archives = [dl_manager.iter_archive(archives) for archives in archives]
        return [datasets.SplitGenerator(name="train", gen_kwargs={"archives": archives})]

    def _generate_examples(self, archives):
        for archive in archives:
            for path, f in archive:
                path_root, path_ext = os.path.splitext(path)
                if path_ext.lower() not in _EXTENSION:
                    continue
                post_id = int(os.path.basename(path_root))
                yield path, {"image": {"bytes": f.read()}, "post_id": post_id}