Something I'm not too sure about is the DDS event pub. I've removed the rename event for directory conversions because I think it doesn't really fit, but we should probably still publish some event(s?) - even if not the existing one.

Thoughts?

Thank you for your patch!

I'm not quite sure if I want to support this feature:

• It checks the file length equal to 0 to decide if the file is empty, but I'm not certain if all filesystems have the same behavior, such as FUSE or remote filesystems.
• The definition of an empty file is unclear. If a file contains nothing but a trailing newline, should it be considered empty? This is common on *nix, and text editors may insert a newline automatically.
• By appending a /, you can rename an empty file to a directory, but I'm not sure how the reverse should be achieved, namely, how to turn an empty directory into a file? This feels incomplete.
• The OS doesn't provide an API to rename a file as a directory, so this has to be implemented as stat + remove + mkdir, which is not an atomic operation and introduces a TOCTOU race condition. That is, if another process modifies the filesystem between the stat and remove, it may lead to data loss. On remote filesystems, this problem could be exacerbated by network instability, which is unsafe.

No worries, I figured the PR might not get accepted since you didn't comment on the issue, as well as the non-atomicity problem. And that's completely reasonable.

Just for the sake of discussion:

• I would be surprised if any mainstream filesystem considered anything other than files with a content size of 0 as empty.
• For this use case I would define empty as a file with no content whatsoever, whitespace or not. The feature provides a convenient way to correct files that get created by mistake when a directory was intended, by forgetting the trailing slash - and those are truncated afaict.
• Fair. However, I imagine the reverse functionality wouldn't be that useful haha.
• Again, fair. Unlikely given the use case but I understand if this level of data loss is unacceptable.