I noticed that in #81261 all of the stdlib module names were explicitly listed, however as of Python 3.10 the stdlib now has a mechanism for this. https://github.com/python/cpython/issues/87121
I figured it was better to use `sys.stdlib_module_names` going forward for 3.10+ instead of having to maintain this file for every new Python release. For docs see:
https://docs.python.org/3/library/sys.html#sys.stdlib_module_names
I did a symmetric difference to determine what the effective change would be. I verified that everything listed in this file ins included in sys.stdlib_module_names. However, there are files in sys.stdlib_module_names that are not included in the previous hard coded definition. Namely these are:
```
frozenset({'__future__',
'_abc',
'_aix_support',
'_asyncio',
'_bisect',
'_blake2',
'_bootsubprocess',
'_bz2',
'_codecs',
'_codecs_cn',
'_codecs_hk',
'_codecs_iso2022',
'_codecs_jp',
'_codecs_kr',
'_codecs_tw',
'_collections',
'_collections_abc',
'_compat_pickle',
'_compression',
'_contextvars',
'_crypt',
'_csv',
'_ctypes',
'_curses',
'_curses_panel',
'_datetime',
'_dbm',
'_decimal',
'_elementtree',
'_frozen_importlib',
'_frozen_importlib_external',
'_functools',
'_gdbm',
'_hashlib',
'_heapq',
'_imp',
'_io',
'_json',
'_locale',
'_lsprof',
'_lzma',
'_markupbase',
'_md5',
'_msi',
'_multibytecodec',
'_multiprocessing',
'_opcode',
'_operator',
'_osx_support',
'_overlapped',
'_pickle',
'_posixshmem',
'_posixsubprocess',
'_py_abc',
'_pydecimal',
'_pyio',
'_queue',
'_random',
'_scproxy',
'_sha1',
'_sha256',
'_sha3',
'_sha512',
'_signal',
'_sitebuiltins',
'_socket',
'_sqlite3',
'_sre',
'_ssl',
'_stat',
'_statistics',
'_string',
'_strptime',
'_struct',
'_symtable',
'_threading_local',
'_tkinter',
'_tracemalloc',
'_uuid',
'_warnings',
'_weakref',
'_weakrefset',
'_winapi',
'_zoneinfo',
'antigravity',
'genericpath',
'idlelib',
'nt',
'nturl2path',
'opcode',
'pydoc_data',
'pyexpat',
'this'})
```
I'm not sure if excluding these matters. I wouldn't think it would, but if it does and it is better to explicitly update this file each time, then feel free to close this.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/81520
Approved by: https://github.com/malfet
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/51694
We implicitly extern standard library modules. Our method of determining
whether a module is in the standard library is a little unreliable. In
particular, I'm seeing lots of flaky errors on windows/mac CI when I
start doing more complicated packaging tests.
I looked into the best ways to do this, turns out there's no reliable
way, so tools that need to do this generally just parse the Python docs
for a listing and save it. I took `isort`'s lists and called it a day.
Test Plan: Imported from OSS
Reviewed By: zdevito
Differential Revision: D26243751
Pulled By: suo
fbshipit-source-id: 48c685cd45ae847fe986bcb9f39106e0c3361cdc
