test_subprocess is failing on Windows.

C:\py\cpython3.7>./python -m test -uall -v -m test_nonexisting_with_pipes test_subprocess
Running Debug|Win32 interpreter...
== CPython 3.7.0b1+ (heads/3.7:1a0239e, Feb 4 2018, 16:19:37) [MSC v.1911 32 bit (Intel)]
== Windows-10-10.0.16299-SP0 little-endian
== cwd: C:\py\cpython3.7\build\test_python_11092
== CPU count: 2
== encodings: locale=cp1251, FS=utf-8
Run tests sequentially
0:00:00 [1/1] test_subprocess
test_nonexisting_with_pipes (test.test_subprocess.ProcessTestCase) ... FAIL
test_nonexisting_with_pipes (test.test_subprocess.ProcessTestCaseNoPoll) ... skipped 'Test needs selectors.PollSelector'

======================================================================
FAIL: test_nonexisting_with_pipes (test.test_subprocess.ProcessTestCase)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\py\cpython3.7\lib\test\test_subprocess.py", line 1194, in test_nonexisting_with_pipes
    self.assertEqual(stderr, "")
AssertionError: 'Traceback (most recent call last):\n  Fil[923 chars]le\n' != ''
Diff is 965 characters long. Set self.maxDiff to None to see it.

----------------------------------------------------------------------
Ran 2 tests in 0.171s

FAILED (failures=1, skipped=1)
test test_subprocess failed
test_subprocess failed

1 test failed:
    test_subprocess

Total duration: 391 ms
Tests result: FAILURE


Here stderr is:

Traceback (most recent call last):
  File "C:\py\cpython3.7\lib\subprocess.py", line 1101, in _execute_child
    args = os.fsdecode(args)  # os.PathLike -> str
  File "C:\py\cpython3.7\\lib\os.py", line 821, in fsdecode
    filename = fspath(filename)  # Does type-checking of `filename`.
TypeError: expected str, bytes or os.PathLike object, not list

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<string>", line 16, in <module>
  File "C:\py\cpython3.7\lib\subprocess.py", line 756, in __init__
    restore_signals, start_new_session)
  File "C:\py\cpython3.7\lib\subprocess.py", line 1104, in _execute_child
    args[0] = os.fsdecode(args[0])  # os.PathLike -> str
  File "C:\py\cpython3.7\\lib\os.py", line 821, in fsdecode
    filename = fspath(filename)  # Does type-checking of `filename`.
TypeError: expected str, bytes or os.PathLike object, not tuple

In _execute_child args is passed to os.fsdecode() unless it is a string. In this case args is a list. os.fsdecode() doesn't accept a list.

The regression was introduced in issue31961.

This is strange, because _execute_child calls os.fsdecode with `args` as the argument, which may be a list. os.fsdecode calls fspath. Now, the python docstring of _fspath, as defined in Lib/os.py on line 1031, clearly states that it will raise a TypeError if the argument is not of type bytes, str or is a os.PathLike object, and that's probably why I wrote the initial code the way I did (catching TypeError from os.fsdecode).

Doesn't the try-except block actually catch this TypeError? I don't understand off the top of my head why my code doesn't catch this exception..

Also, isn't there continuous integration testing? Everything passed on the PR, so where does this come from?

Wait a minute. The failing test is test_nonexisting_with_pipes, and it fails because args[0] is a tuple - how can that be? Nobody is supposed to pass cmd=sequence-where-first-element-is-a-tuple!

Is everything all right with the test itself?

Turns out this is a trivial typo in the test, so I stole the issue from Greg and pushed a PR.

I fixed this independently earlier today, tied to bpo-30121 (the issue that introduced the test bug) because I didn't find this one.  See PR5758, PR5759 (3.7), and PR5760 (3.6).

Sounds good to me