fd cannot search files under a RAM disk

[pgp]

Describe the bug you encountered:

when performing search (e.g. fd somepattern) under a ramdisk (using ImDisk as RAM disk utility), the following error is returned immediately:

[fd error] Cannot retrieve current directory (has it been deleted?)

Instead, the builtin windows prompt command:
dir /S /P somepattern
works correctly.

What version of fd are you using?
8.2.1

Which operating system / distribution are you on?
Windows 8.1 x64

[sergeevabc]

Also, it does not work with virtual disks created by apps encrypting folders (e.g. cppcryptfs).

[sharkdp]

Thank you for reporting this.

This is due to this check in main.rs:

    let current_directory = Path::new(".");
    if !filesystem::is_dir(current_directory) {
        return Err(anyhow!(
            "Could not retrieve current directory (has it been deleted?)."
        ));
    }

We probably shouldn't check whether . is a directory... just if it exists(?). Apparently on these special file systems, . is not a directory? Can someone confirm this?

[pgp]

I did a quick check using the os.stat command within python, and the st_mode integer is identical for folders under the RAM disk, and folders on the main hard drive. Also, the rest of the os.stat output does not present notable differences, so the issue might be located somewhere (or it could be due to a bad implementation of filesystem::is_dir)...

I'll try to run the filesystem::is_dir snippet under the two kinds of directories as well, and let you know.

[pgp]

Well, it seems that the culprit is the path.canonicalize().is_ok() part in the is_dir implementation, which returns false in the RAM disk case. The other check within the parenthesis (path.file_name().is_some()) returns false for both filesystems, at least when providing the current directory "."

[tavianator]

Python has run into very similar issues with ImDisk before: https://bugs.python.org/issue26658

I'm guessing that canonicalize() is trying to do the same thing as Python used to, i.e. figure out the drive letter associated with the mount point. That seems surprisingly aggressive to me -- I expect it to resolve symlinks, but not other junction/reparse points. Might be worth reporting as a Rust bug.

[bl-ue]

Having the same issue. I'm using ImDisk for a virtual R:\ drive and fd can't do a thing in it.

[sharkdp]

I took another look at this.

Related issues on the Rust tracker: rust-lang/rust#59117 (and rust-lang/rust#74327)

We might be able to use the normpath crate to fix this: https://docs.rs/normpath/0.3.0/normpath/trait.PathExt.html#tymethod.normalize