在加载 GDAL 包时使用 zipfile 时出现 Python 段错误

Question

我在尝试使用 zipfile 提取 ZIP 文件时遇到了一个奇怪的段错误错误。导入 osgeo.gdal 的模块模块(并非唯一，osgeo.ogr 产生相同的错误——也许它与使用 SWIG 的两个模块有关？)。
我能够将崩溃缩小到 osgeo.gdal 的特定导入，请参见下面的示例。使用 GDB 运行 Python 解释器似乎表明崩溃源于无效的 free()。在 zlib 中的某个地方调用，但不幸的是我缺乏在那里进一步挖掘的经验。
有人猜测这里可能出了什么问题吗？或者也许一些关于如何最好地调试这样的问题的指针(呵呵)？
万分感谢!
编辑:经过更多调查，我发现针对 Conda 提供的 GDAL 包( https://github.com/conda-forge/gdal-feedstock/issues/365 )报告了类似的问题。将 GDAL 模块导入移动到导入列表末尾的建议解决方法修复了段错误，就是这样!不过，我很感激任何可以为我指明调试此类问题的正确方向的答案:)
违规代码，(几乎)最小的例子

# >>> Comment this import to make the segfault disappear
from osgeo import gdal
# <<<

import io
import tempfile
import zipfile
import urllib3
import faulthandler

class GadmDownloader:
    BASE_URL_TEMPLATE = 'https://biogeo.ucdavis.edu/data/gadm3.6/gpkg/gadm36_{}_gpkg.zip'

    def __init__(self):
        self.http = urllib3.PoolManager()

    def download_country(self, country, output_dir = None):
        url = self.BASE_URL_TEMPLATE.format(country.upper())

        resp: urllib3.HTTPResponse = self.http.request('GET', url)
        if resp.status != 200:
            raise RuntimeError(f'Download failed: {resp.status}')

        faulthandler.enable()
        with zipfile.ZipFile(io.BytesIO(resp.data), 'r') as zipf:
            gpkg = [name for name in zipf.namelist()
                    if name.endswith('.gpkg')]
            zipf.extract(gpkg[0], output_dir)
        return gpkg[0]


def main():
    gdal.UseExceptions()

    dl = GadmDownloader()
    with tempfile.TemporaryDirectory() as tempdir:
        path = dl.download_country('gmb', tempdir)
        print(f'>> Path: {path}')

if __name__ == "__main__":
    main()

输出

$ python -X dev gadm.py                                                                                                                                                                                               (minikube:default)
/usr/lib/python3.8/site-packages/osgeo/__init__.py:15: DeprecationWarning: the imp module is deprecated in favour of importlib; see the module's documentation for alternative uses
  import imp
Fatal Python error: Segmentation fault

Current thread 0x00007ffb15579740 (most recent call first):
  File "/usr/lib/python3.8/zipfile.py", line 724 in _get_decompressor
  File "/usr/lib/python3.8/zipfile.py", line 819 in __init__
  File "/usr/lib/python3.8/zipfile.py", line 1571 in open
  File "/usr/lib/python3.8/zipfile.py", line 1700 in _extract_member
  File "/usr/lib/python3.8/zipfile.py", line 1630 in extract
  File "gadm.py", line 29 in download_country
  File "gadm.py", line 38 in main
  File "gadm.py", line 42 in <module>
[1]    200128 segmentation fault (core dumped)  python -X dev -d -X showmalloc gadm.py

GDB 回溯

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff7e3aa50 in free () from /usr/lib/libc.so.6
(gdb) bt
#0  0x00007ffff7e3aa50 in free () from /usr/lib/libc.so.6
#1  0x00007ffff485ae0f in inflateReset2 () from /usr/lib/libcfitsio.so.9
#2  0x00007ffff39ec1a4 in inflateInit2_ () from /usr/lib/libz.so.1
#3  0x00007ffff74a7880 in ?? () from /usr/lib/python3.8/lib-dynload/zlib.cpython-38-x86_64-linux-gnu.so
#4  0x00007ffff7b466ad in ?? () from /usr/lib/libpython3.8.so.1.0
#5  0x00007ffff7b43097 in _PyEval_EvalFrameDefault () from /usr/lib/libpython3.8.so.1.0
#6  0x00007ffff7b4f138 in _PyFunction_Vectorcall () from /usr/lib/libpython3.8.so.1.0
#7  0x00007ffff7b3e77c in _PyEval_EvalFrameDefault () from /usr/lib/libpython3.8.so.1.0
#8  0x00007ffff7b3d0c4 in _PyEval_EvalCodeWithName () from /usr/lib/libpython3.8.so.1.0
#9  0x00007ffff7b47095 in _PyObject_FastCallDict () from /usr/lib/libpython3.8.so.1.0
#10 0x00007ffff7b59c2f in ?? () from /usr/lib/libpython3.8.so.1.0
#11 0x00007ffff7b47c90 in _PyObject_MakeTpCall () from /usr/lib/libpython3.8.so.1.0
#12 0x00007ffff7b42fc4 in _PyEval_EvalFrameDefault () from /usr/lib/libpython3.8.so.1.0
#13 0x00007ffff7b3d7e2 in _PyEval_EvalCodeWithName () from /usr/lib/libpython3.8.so.1.0
#14 0x00007ffff7b5e982 in ?? () from /usr/lib/libpython3.8.so.1.0
#15 0x00007ffff7b3f553 in _PyEval_EvalFrameDefault () from /usr/lib/libpython3.8.so.1.0
#16 0x00007ffff7b3d7e2 in _PyEval_EvalCodeWithName () from /usr/lib/libpython3.8.so.1.0
#17 0x00007ffff7b4f1cd in _PyFunction_Vectorcall () from /usr/lib/libpython3.8.so.1.0
#18 0x00007ffff7b3eb41 in _PyEval_EvalFrameDefault () from /usr/lib/libpython3.8.so.1.0
#19 0x00007ffff7b3d0c4 in _PyEval_EvalCodeWithName () from /usr/lib/libpython3.8.so.1.0
#20 0x00007ffff7b4f1cd in _PyFunction_Vectorcall () from /usr/lib/libpython3.8.so.1.0
#21 0x00007ffff7b3eb41 in _PyEval_EvalFrameDefault () from /usr/lib/libpython3.8.so.1.0
#22 0x00007ffff7b3d0c4 in _PyEval_EvalCodeWithName () from /usr/lib/libpython3.8.so.1.0
#23 0x00007ffff7b4f1cd in _PyFunction_Vectorcall () from /usr/lib/libpython3.8.so.1.0
#24 0x00007ffff7b3eb41 in _PyEval_EvalFrameDefault () from /usr/lib/libpython3.8.so.1.0
#25 0x00007ffff7b4f138 in _PyFunction_Vectorcall () from /usr/lib/libpython3.8.so.1.0
#26 0x00007ffff7b3e77c in _PyEval_EvalFrameDefault () from /usr/lib/libpython3.8.so.1.0
#27 0x00007ffff7b3d0c4 in _PyEval_EvalCodeWithName () from /usr/lib/libpython3.8.so.1.0
#28 0x00007ffff7bee323 in PyEval_EvalCode () from /usr/lib/libpython3.8.so.1.0
#29 0x00007ffff7bf9b98 in ?? () from /usr/lib/libpython3.8.so.1.0
#30 0x00007ffff7bf3db3 in ?? () from /usr/lib/libpython3.8.so.1.0
#31 0x00007ffff7ab2a0b in PyRun_FileExFlags () from /usr/lib/libpython3.8.so.1.0
#32 0x00007ffff7ab2482 in PyRun_SimpleFileExFlags () from /usr/lib/libpython3.8.so.1.0
#33 0x00007ffff7c06a1a in Py_RunMain () from /usr/lib/libpython3.8.so.1.0
#34 0x00007ffff7be2bd9 in Py_BytesMain () from /usr/lib/libpython3.8.so.1.0
#35 0x00007ffff7dd6002 in __libc_start_main () from /usr/lib/libc.so.6
#36 0x000055555555504e in _start ()

系统事实

Linux 5.7.10-1-MANJARO #1 SMP PREEMPT 2020 年 7 月 22 日星期三 15:23:40 UTC x86_64 GNU/Linux

Arch/Manjaro Linux

Python 3.8.5

GDAL 3.0.4

Answer 1

以相反的字母顺序执行 Python 导入解决了这个问题。