3.12.0: test_bad_lock_file[current_directory-UnixFileLock] fails

[Marcel Telka]

During upgrade of the filelock package for OpenIndiana from version 3.11.0 to version 3.12.0 I noticed that one test fails (with version 3.11.0 all tests passed):

______________ test_bad_lock_file[current_directory-UnixFileLock] ______________

lock_type = <class 'filelock._unix.UnixFileLock'>
expected_error = <class 'IsADirectoryError'>, match = 'Is a directory'
bad_lock_file = '.'

    @pytest.mark.parametrize("lock_type", [FileLock, SoftFileLock])
    @pytest.mark.parametrize(
        ("expected_error", "match", "bad_lock_file"),
        [   
            pytest.param(FileNotFoundError, "No such file or directory:", "a/b", id="non_existent_directory"),
            pytest.param(FileNotFoundError, "No such file or directory:", "", id="blank_filename"),
            pytest.param(ValueError, "embedded null (byte|character)", "\0", id="null_byte"),
            pytest.param(
                PermissionError if sys.platform == "win32" else IsADirectoryError,
                "Permission denied:" if sys.platform == "win32" else "Is a directory",
                ".",
                id="current_directory",
            ),
        ]
        + [pytest.param(OSError, "Invalid argument", i, id=f"invalid_{i}", marks=WindowsOnly) for i in '<>:"|?*\a']
        + [pytest.param(PermissionError, "Permission denied:", i, id=f"permission_{i}", marks=WindowsOnly) for i in "/\\"],
    )
    @pytest.mark.timeout(5)  # timeout in case of infinite loop
    def test_bad_lock_file(
        lock_type: type[BaseFileLock],
        expected_error: type[Exception],
        match: str,
        bad_lock_file: str,
    ) -> None:
        lock = lock_type(bad_lock_file)

        with pytest.raises(expected_error, match=match):
>           lock.acquire()

tests/test_filelock.py:145:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
../prototype/i386/usr/lib/python3.9/vendor-packages/filelock/_api.py:217: in acquire
    self._acquire()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = <filelock._unix.UnixFileLock object at 0x7fffa83972e0>

    def _acquire(self) -> None:
        open_flags = os.O_RDWR | os.O_CREAT | os.O_TRUNC
>       fd = os.open(self.lock_file, open_flags, self._context.mode)
E       OSError: [Errno 22] Invalid argument: '.'

../prototype/i386/usr/lib/python3.9/vendor-packages/filelock/_unix.py:36: OSError

[Matthew Schweiss]

I see the current fix is to omit this test which will work, since this is an edgecase of the test, not of filelock.

I'd like a little more information though so we can see about a permanent fix. Sorry I am not familiar with OpenIndiana / SunOS.

What is sys.platform for this OS?

For this OS, I assume the expected error for trying to open "." as if it were a file is OSError: [Errno 22] Invalid argument: '.'?

os.open('.', os.O_RDWR | os.O_CREAT | os.O_TRUNC, 0o644)

[Marcel Telka]

$ python -c 'import sys; print(sys.platform)'
sunos5
$ python -c "import os; os.open('.', os.O_RDWR | os.O_CREAT | os.O_TRUNC, 0o644)"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
OSError: [Errno 22] Invalid argument: '.'
$

[Matthew Schweiss]

This is interesting. Based on the Oracle page, I would have thought the error message would be EISDIR. If you can explain this, I'd love to know.

[Matthew Schweiss]

Can you try another thing for me? Want to make sure I understand how this is behaving.

os.open(os.path.curdir, os.O_WRONLY | os.O_CREAT | os.O_TRUNC, stat.S_IRUSR | stat.S_IWUSR | stat.S_IRGRP | stat.S_IROTH)

open(os.path.curdir)

print(os.path.curdir)

[Matthew Schweiss]

Considering the reason for this test case was because of a Windows specific error (#231), I think it may be reasonable to change the check so "if non-windows, make sure the result isn't EPERM".

[Bernát Gábor]

This should be done now, the CI passes at least with 3.12.