import os
from datasets import GeneratorBasedBuilder, DatasetInfo, SplitGenerator, Split, Features, Value, Sequence, Image


class TruckV2X(GeneratorBasedBuilder):
    VERSION = "1.0.0"

    def _info(self):
        return DatasetInfo(
            description="TruckV2X: A Truck-Centered Perception Dataset",
            features=Features({
                "town": Value("string"),
                "frame_id": Value("string"),
                "agent_data": {
                    "cav": {
                        "camera_images": Sequence(Image()),
                        "lidar_files": Sequence(Value("string")),
                        "yaml_file": Value("string"),
                    },
                    "roadside": {
                        "camera_images": Sequence(Image()),
                        "lidar_files": Sequence(Value("string")),
                        "yaml_file": Value("string"),
                    },
                    "tractor": {
                        "camera_images": Sequence(Image()),
                        "lidar_files": Sequence(Value("string")),
                        "yaml_file": Value("string"),
                    },
                    "trailer": {
                        "camera_images": Sequence(Image()),
                        "lidar_files": Sequence(Value("string")),
                        "yaml_file": Value("string"),
                    },
                }
            }),
            supervised_keys=None,
            homepage="TBD",
            license="MIT"
        )

    def _split_generators(self, dl_manager):
        # 直接用 dl_manager 下载并自动解压
        train_zips = [
            "train/Town1_1.zip", "train/Town1_4.zip", "train/Town1_5.zip",
            "train/Town2_0.zip", "train/Town2_1.zip", "train/Town2_2.zip",
            "train/Town2_3.zip", "train/Town3_1.zip", "train/Town3_4.zip",
            "train/Town3_6.zip", "train/Town3_7.zip", "train/Town3_8.zip",
            "train/Town3_9.zip", "train/Town4_0.zip", "train/Town4_1.zip",
            "train/Town4_3.zip", "train/Town4_4.zip", "train/Town4_6.zip",
            "train/Town5_0.zip", "train/Town5_3.zip", "train/Town5_4.zip",
            "train/Town5_5.zip", "train/Town5_6.zip", "train/Town5_7.zip",
            "train/Town5_8.zip", "train/Town6_0.zip", "train/Town6_2.zip",
            "train/Town6_3.zip", "train/Town6_4.zip", "train/Town6_5.zip",
            "train/Town7_1.zip", "train/Town7_3.zip", "train/Town10_0.zip",
            "train/Town10_1.zip", "train/Town10_6.zip", "train/Town10_7.zip",
            "train/Town10_8.zip", "train/Town15_1.zip"
        ]

        val_zips = [
            "val/Town3_0.zip", "val/Town4_5.zip", "val/Town5_2.zip",
            "val/Town5_9.zip", "val/Town6_6.zip", "val/Town7_4.zip",
            "val/Town10_4.zip", "val/Town10_9.zip", "val/Town10_10.zip"
        ]

        test_zips = [
            "test/Town1_0.zip", "test/Town1_2.zip", "test/Town1_3.zip",
            "test/Town2_4.zip", "test/Town2_5.zip", "test/Town3_2.zip",
            "test/Town3_3.zip", "test/Town3_5.zip", "test/Town4_2.zip",
            "test/Town5_1.zip", "test/Town6_1.zip", "test/Town7_0.zip",
            "test/Town7_2.zip", "test/Town10_2.zip", "test/Town10_3.zip",
            "test/Town10_5.zip", "test/Town15_0.zip"
        ]    

        train_dirs = [dl_manager.extract(dl_manager.download(p)) for p in train_zips]
        val_dirs = [dl_manager.extract(dl_manager.download(p)) for p in val_zips]
        test_dirs = [dl_manager.extract(dl_manager.download(p)) for p in test_zips]

        return [
            SplitGenerator(name=Split.TRAIN, gen_kwargs={"town_dirs": train_dirs}),
            SplitGenerator(name=Split.VALIDATION, gen_kwargs={"town_dirs": val_dirs}),
            SplitGenerator(name=Split.TEST, gen_kwargs={"town_dirs": test_dirs}),
        ]

    def _generate_examples(self, town_dirs):
        id_ = 0
        agents = ["cav", "roadside", "tractor", "trailer"]
        camera_counts = {"cav": 4, "roadside": 1, "tractor": 5, "trailer": 5}
        lidar_counts = {"cav": 1, "roadside": 1, "tractor": 2, "trailer": 2}

        for town_dir in town_dirs:
            # 如果自动解压后是有一级文件夹，进入它
            subfolders = [f for f in os.listdir(town_dir) if os.path.isdir(os.path.join(town_dir, f))]
            if len(subfolders) == 1:
                town_dir = os.path.join(town_dir, subfolders[0])

            town_name = os.path.basename(town_dir.rstrip("/\\"))

            # 用cav目录判断帧ID
            cav_dir = os.path.join(town_dir, "cav")
            if not os.path.exists(cav_dir):
                continue

            frame_ids = sorted([
                fname.replace(".yaml", "")
                for fname in os.listdir(cav_dir)
                if fname.endswith(".yaml")
            ])

            for frame_id in frame_ids:
                example = {
                    "town": town_name,
                    "frame_id": frame_id,
                    "agent_data": {}
                }

                for agent in agents:
                    agent_dir = os.path.join(town_dir, agent)
                    data = {
                        "camera_images": [],
                        "lidar_files": [],
                        "yaml_file": ""
                    }

                    for i in range(camera_counts[agent]):
                        cam_path = os.path.join(agent_dir, f"{frame_id}_camera{i}.jpg")
                        if os.path.exists(cam_path):
                            data["camera_images"].append(cam_path)

                    for i in range(lidar_counts[agent]):
                        lidar_path = os.path.join(agent_dir, f"{frame_id}_lidar{i}.pcd")
                        if os.path.exists(lidar_path):
                            data["lidar_files"].append(lidar_path)

                    yaml_path = os.path.join(agent_dir, f"{frame_id}.yaml")
                    if os.path.exists(yaml_path):
                        data["yaml_file"] = yaml_path

                    example["agent_data"][agent] = data

                yield id_, example
                id_ += 1