I was a bit confused about the "delete on close" semantics on Windows and did some experiments on it. The result indicates that Dokan's current API doesn't accurately reflect its semantics.

• If I call NtCreateFile on a file with DELETE permission and with the FILE_DELETE_ON_CLOSE flag set (let's call it h1)
  • If there's no other open handles, the file will be deleted once h1 g

Talking about https://docs.pyfilesystem.org/en/latest/reference/walk.html#fs.walk.BoundWalker.dirs and maybe walk.files is affected as well.

The documentation talks about (emphasis by me):

dirs(path='/', **kwargs)
Walk a filesystem, yielding ABSOLUTE paths to directories.

and

filter_dirs (list, optional) – A list of patterns that will be used to match directori

It took me quite a while to discover this library. I was using "inotify" as search keyword on https://godoc.org/?q=inotify.

Consider adding it to the first paragraph of package-level godoc.

Great work and thanks for this library!

When you call eg. "lscp" without specifiing device it does look in /proc/mounts for first nilfs device and use this. But i think it should not just take first device. It should check wheter the current working directory is inside a NILFS mount and use according device. Also it should work if i specify directory instead of device. eg. lscp /mnt/nilfs.

BTW there is nilfs2 plugin for thunar that