Python迭代类__next__未触发，返回生成器对象问题排查

问题现象 #

自定义可迭代类VSIFile用于遍历VSI文件的ROI区域，实现了标准的__iter__（返回自身）和__next__方法，但执行遍历代码时，输出的是生成器对象而非预期的numpy数组。调试确认__next__方法从未被触发，仅__iter__方法执行。测试简单数字计数迭代器类运行正常，但当前类出现异常，示例输出：

<generator object VSIFile.__next__ at 0x000001AA3C4AFC80>

关联代码 #

vsi_file.py #

from typing import Tuple
import javabridge
import bioformats
from tqdm import tqdm
from cv2 import resize

javabridge.start_vm(class_path=bioformats.JARS)


class VSIFile:

    def __init__(self, vsi_file: str,
                 roi_size: Tuple[int, int] = (1024, 1024),
                 target_size: Tuple[int, int] = (256, 256),
                 use_pbar: bool = True):

        self.file_path = vsi_file
        self.roi_size = roi_size
        self.target_size = target_size
        self.slide = None
        self.shape = None
        self.max_x_idx = None
        self.max_y_idx = None
        self.num_rois = None
        self.skip = [1, 2, 5, 11, 22, 45, 72]

        if use_pbar:
            self.pbar = tqdm()
        else:
            self.pbar = None

    def __enter__(self):
        self.slide = bioformats.ImageReader(self.file_path)
        self.shape = self.slide.rdr.getSizeY(), self.slide.rdr.getSizeX(), 3
        self.max_x_idx = self.shape[1] // self.roi_size[1]
        self.max_y_idx = self.shape[0] // self.roi_size[0]
        self.num_rois = self.max_x_idx * self.max_y_idx

        if self.pbar is not None:
            self.pbar.total = self.num_rois
            self.pbar.refresh()

        return self

    def __exit__(self, exc_type, exc_val, exc_tb):
        self.slide.close()

        if self.pbar is not None:
            self.pbar.close()

    def __del__(self):
        self.slide.close()

        if self.pbar is not None:
            self.pbar.close()

    def __iter__(self):
        return self

    def __next__(self):
        while self.idx in self.skip:
            if self.idx == self.max_x_idx * self.max_y_idx:
                if self.pbar is not None:
                    self.pbar.close()
                raise StopIteration

            self.idx += 1

            if self.pbar is not None:
                self.pbar.update(1)

        if self.idx == self.max_x_idx * self.max_y_idx:
            if self.pbar is not None:
                self.pbar.close()
            raise StopIteration

        y = (self.idx // self.max_x_idx) * self.roi_size[0]
        x = (self.idx % self.max_x_idx) * self.roi_size[1]

        roi = self.get_roi(x, y, self.roi_size[0], self.roi_size[1])
        roi = resize(roi, self.target_size) if self.target_size else roi
        yield roi

        self.idx += 1

        if self.pbar is not None:
            self.pbar.update(1)

process_vsi.py（相关片段） #

with VSIFile(os.path.abspath(os.path.join(data_dir, file))) as vsi:
  for roi in vsi:
    print(roi)

问题根源与修复方案 #

核心问题1：__next__误用yield #

迭代器的__next__方法要求直接返回单个元素，当使用yield时，该方法会变成生成器函数，调用时直接返回生成器对象，而非预期的numpy数组，这也是遍历得到生成器的原因。

核心问题2：未初始化迭代索引self.idx #

代码中使用了self.idx但未在任何地方初始化，会导致运行时抛出AttributeError，需要在__enter__方法中添加初始化语句。

修复后的关键代码 #

1. 在__enter__方法末尾添加索引初始化：

def __enter__(self):
    # 原有代码...
    self.idx = 0  # 添加这行初始化迭代索引
    return self

2. 修改__next__方法，移除yield，改为直接返回roi，并调整后续逻辑：

def __next__(self):
    while self.idx in self.skip:
        if self.idx == self.max_x_idx * self.max_y_idx:
            if self.pbar is not None:
                self.pbar.close()
            raise StopIteration

        self.idx += 1

        if self.pbar is not None:
            self.pbar.update(1)

    if self.idx == self.max_x_idx * self.max_y_idx:
        if self.pbar is not None:
            self.pbar.close()
        raise StopIteration

    y = (self.idx // self.max_x_idx) * self.roi_size[0]
    x = (self.idx % self.max_x_idx) * self.roi_size[1]

    roi = self.get_roi(x, y, self.roi_size[0], self.roi_size[1])
    roi = resize(roi, self.target_size) if self.target_size else roi

    self.idx += 1
    if self.pbar is not None:
        self.pbar.update(1)
    
    return roi  # 替换yield为return

验证 #

修复后再次执行process_vsi.py的遍历代码，将输出预期的numpy数组，__next__方法会被正常触发，完成ROI的迭代输出。

内容的提问来源于stack exchange，提问作者Jack Avante