proposed fix for issue #67

[jahrules]

This is a proposed fix for #67

As suggested in the issue; we're checking the OSError for NotImplemented (which is errno 38).

OSError 38 is not fatal; so this results in the reported issue of an indeterminate loop.

I don't see a clear way to "fall back" from a UnixFileLock to a SoftFileLock; so my proposal is to raise a fatal error with a clear message to the end user.

I also don't see a good way to create a testcase for this; as raising an OSError code 38 requires a condition that can't be easily simulated. I'm open to ideas for this if I'm missing something obvious.

[TheMatt2]

I also don't see a good way to create a testcase for this; as raising an OSError code 38 requires a condition that can't be easily simulated.

I think there is actually a relatively straight forward way to test this condition.
You can stub fcntl.flock to test the function.

(This is psuedo code for it, I haven't tested it)

import errno
import fcntl
import filelock

def dummy_flock(fd, operation):
    raise OSError(errno.ENOSYS, "flock not supported")

def test_flock_not_implemented():
    _fcntl_flock = fcntl.flock
    try:
        fcntl.flock = dummy_flock
        with pytest.raises(NotImplementedError):
            with filelock.FileLock("/path/to/test/file"): pass
    finally:
        fcntl.flock = _fcntl_flock

[gaborbernat]

Can you try adding the suggested test and changelog entry?

[jahrules]

I also don't see a good way to create a testcase for this; as raising an OSError code 38 requires a condition that can't be easily simulated.

I think there is actually a relatively straight forward way to test this condition. You can stub fcntl.flock to test the function.

(This is psuedo code for it, I haven't tested it)

import errno
import fcntl
import filelock

def dummy_flock(fd, operation):
    raise OSError(errno.ENOSYS)

def test_flock_not_implemented():
    _fcntl_flock = fcntl.flock
    try:
        fcntl.flock = dummy_flock
        with pytest.raises(NotImplementedError):
            with filelock.FileLock("/path/to/test/file"): pass
    finally:
        fcntl.flock = _fcntl_flock

This worked very well thanks.

I write a lot of functional and regression type tests, so my unit-test-foo is a bit rusty.

[jahrules]

After a lot of fighting with the Unix type checker I finally got it to pass. (frustration)
the Windows type checker is still failing because it sees the flock imports even though the testcase is only set to run for Unix.

I'm not sure how to tell the windows type checker to ignore these...

[TheMatt2]

I'm not sure how to tell the windows type checker to ignore these...

It seems the way to get the type checker to ignore these is to change

@pytest.mark.skipif(sys.platform == "win32", reason="Windows filesystems do not support flock")
def test_flock_not_implemented_unix(tmp_path: Path) -> None:

to

def test_flock_not_implemented_unix(tmp_path: Path) -> None:
    if sys.platform == "win32":
        pytest.skip("Windows filesystems do not support flock")

Will make the type checker happy on windows. That way you don't need to have ignore statements that Ubuntu is now complaining about!

(I must admit, Github Copilot actually gets the credit for figuring this out, which is just a little scary)

[jahrules]

I'm not sure how to tell the windows type checker to ignore these...

It seems the way to get the type checker to ignore these is to change

@pytest.mark.skipif(sys.platform == "win32", reason="Windows filesystems do not support flock")
def test_flock_not_implemented_unix(tmp_path: Path) -> None:

to

def test_flock_not_implemented_unix(tmp_path: Path) -> None:
    if sys.platform == "win32":
        pytest.skip("Windows filesystems do not support flock")

Will make the type checker happy on windows. That way you don't need to have ignore statements that Ubuntu is now complaining about!

(I must admit, Github Copilot actually gets the credit for figuring this out, which is just a little scary)

too bad mypy couldnt just detect the skipif decorator. Talking to some people; apparently in certain circumstances it can... idk. Im just happy it works now.

[gaborbernat]

too bad mypy couldnt just detect the skipif decorator. Talking to some people; apparently in certain circumstances it can... idk. Im just happy it works now.

Check my feedback fix, that avoids the issue.

[TheMatt2]

Please use err.errno == errno.ENOSYS instead of hard coding 38.

On common linux it looks like ENOSYS is 38. On MacOS ENOSYS is 78.

[TheMatt2]

fixed

[TheMatt2]

typo fix.
Perhaps "Filesystem does not support file locking. Please use SoftFileLock instead."?