import glob
import os
import datasets
from huggingface_hub import snapshot_download

_CITATION = """TO BE ADDED."""

_DESCRIPTION = """
HeuriGen is a collection of combinatorial-optimization problems
for benchmarking heuristic-program generation by LLMs.
"""

REPO_ID = "heurigen/heurigen-data"


class HeuriGenConfig(datasets.BuilderConfig):
    def __init__(self, **kwargs):
        super().__init__(version=datasets.Version("0.1.0"), **kwargs)


_BUILDER_CONFIGS = [
    HeuriGenConfig(
        name="operator_scheduling", description="DFG operator-scheduling instances"
    ),
    HeuriGenConfig(
        name="technology_mapping",
        description="Logic-network technology-mapping instances",
    ),
    HeuriGenConfig(
        name="global_routing", description="Netlist global-routing instances"
    ),
    HeuriGenConfig(
        name="egraph_extraction", description="E-graph extraction instances"
    ),
    HeuriGenConfig(name="intra_op_parallel", description="Intra-op parallel instances"),
    HeuriGenConfig(
        name="protein_sequence_design", description="Protein sequence design instances"
    ),
    HeuriGenConfig(name="pedigree", description="Pedigree problem instances"),
    HeuriGenConfig(
        name="pickup_delivery_time_windows", description="Pickup-delivery-time-windows (PDPTW) instances"
    ),
    HeuriGenConfig(name="crew_pairing", description="Crew pairing instances"),
    HeuriGenConfig(name="frequency_assignment", description="Frequency assignment instances"),
]


class HeuriGen(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIGS = _BUILDER_CONFIGS
    DEFAULT_CONFIG_NAME = "operator_scheduling"

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            citation=_CITATION,
            features=datasets.Features(
                {
                    "file_path": datasets.Value("string"),
                }
            ),
            supervised_keys=None,
            homepage=f"https://huggingface.co/datasets/{REPO_ID}",
        )

    def _split_generators(self, dl_manager):
        local_root = snapshot_download(
            repo_id=REPO_ID,
            repo_type="dataset",
            revision="main",
            allow_patterns=[f"{self.config.name}/**"],
            local_dir=dl_manager.manual_dir,
        )
        base = os.path.join(local_root, self.config.name)

        # helper – returns all files in that split, including files in subfolders but not the folders themselves
        def files(pattern):
            all_files = []
            for path in glob.glob(os.path.join(base, pattern), recursive=True):
                if os.path.isfile(path):
                    all_files.append(os.path.abspath(path))
            return sorted(all_files)

        demo_paths = files("demo/**/*")
        eval_paths = files("eval/**/*")

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"files": demo_paths},
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={"files": eval_paths},
            ),
        ]

    def _generate_examples(self, files):
        for idx, path in enumerate(files):
            yield idx, {"file_path": path}