#
import os
import torch
from datasets import DatasetInfo, GeneratorBasedBuilder, BuilderConfig, Split, SplitGenerator

class LPBFConfig(BuilderConfig):
    def __init__(self, **kwargs):
        super().__init__(**kwargs)

class LPBFDataset(GeneratorBasedBuilder):
    VERSION = "1.0.0"
    BUILDER_CONFIGS = [
        LPBFConfig(name="default", version=VERSION, description="Laser powder bed fusion additive manufacturing dataset")
    ]

    # No fixed schema because we return a PyG graph object
    def _info(self):
        return DatasetInfo(
            description="Laser powder bed fusion additive manufacturing dataset",
            features=None,
        )

    def _split_generators(self, dl_manager):
        """
        After extraction, the dataset structure should look like:
          LPBF/train/*.pt
          LPBF/test/*.pt
        """
        data_dir = dl_manager.download_and_extract(
            "https://huggingface.co/datasets/vedantpuri/LPBF_FLARE/resolve/main/LPBF.tar.gz"
        )

        # if the tar contains a subdir called LPBF/
        # then dl_manager extracts to something like: [...]/LPBF/
        split_train = os.path.join(data_dir, "LPBF", "train")
        split_test = os.path.join(data_dir, "LPBF", "test")

        return [
            SplitGenerator(name=Split.TRAIN, gen_kwargs={"data_dir": split_train}),
            SplitGenerator(name=Split.TEST, gen_kwargs={"data_dir": split_test}),
        ]


    def _generate_examples(self, data_dir):
        files = sorted([f for f in os.listdir(data_dir) if f.endswith(".pt")])
        for idx, fname in enumerate(files):
            path = os.path.join(data_dir, fname)
            obj = torch.load(path, map_location="cpu")
            # Return a dict so HF can wrap it
            yield idx, {"graph": obj}