Help with returning value from func with reference

[kahunacohen]

Firstly, I'd like to compliment the rust community for welcoming people like me who have programmed in languages like Python, JavaScript etc and not C, C++ etc. It's an attitude like this that will help push Rust forward.

Also I'd like to thank whomever write the compiler, as its messages are super helpful.

That said, I have the need to extract the file name of a path that is not known in advance. My understanding is that a string not known in advance (e.g. gotten from program args, a config file etc.) must be a String type, as that's stored on the heap, not the stack and can't be determined at compile-time. Is that right?

The problem is, in the below function I return a struct that indirectly contains a reference to a string and thus get a compile-time error : cannot return value referencing local data f.path`.

Here's an excerpt of the code in question:

...
#[derive(Debug)]
struct SyncData<'a> {
    access_token: &'a String,
    file_name: Box<&'a OsStr>,
    file_modified: DateTime<Utc>
}
fn get_sync_data(access_token: &String, f: config::GistFile) -> Result<SyncData, Box<dyn Error>> {
    // Get file name from GistFile struct's path prop, which is a String.
   // for now ignore error handling and unwrap.
    let fname = Path::new(&f.path).file_name().unwrap();
    Ok(SyncData {
        file_name: Box::new(fname),
        access_token: access_token,
        file_modified: fs::metadata(fname)?.modified()?.into(),
    })
}

My understanding is that this error occurs because f.path is dropped at the end of the get_sync_data function, but the SyncData "instance" returned from the function contains a reference to the dropped f.path variable. Rust complains because of what in C could be a dangling pointer. Is that right, and if so how do I go about addressing this?

[2e71828]

file_name should probably be an OsString. It's the owned counterpart to OsStr in the same way String is the owned version of str.

[alice]

Yes. You also do not need the Box around the OsStr. Just use an OsString.

[kahunacohen]

I'm not sure how that solves my problem. It seems Path::new expects a reference to a string. If I change the struct so that file_name is an OsString and try to get a Path by converting f.path which is a String to an OsString I can't construct the Path because it expects a ref. If I use a ref, I get the original error because of a dangling pointer....

[2e71828]

I meant something like this:

#[derive(Debug)]
struct SyncData<'a> {
    access_token: &'a str,
    file_name: OsString,
    file_modified: DateTime<Utc>
}

fn get_sync_data(access_token: &str, f: GistFile) -> Result<SyncData, Box<dyn Error>> {
    // Get file name from GistFile struct's path prop, which is a String.
   // for now ignore error handling and unwrap.
    let fname = OsString::from(Path::new(&f.path).file_name().unwrap());
    Ok(SyncData {
        access_token: access_token,
        file_modified: fs::metadata(&fname)?.modified()?.into(),
        file_name: fname,
    })
}

Playground

(I also replaced &String with &str because it's strictly more general. You can still pass an &String there due to auto-dereferencing.)

[kahunacohen]

This is great and works. Thanks for helping.

Instead of unwrapping, now I tried an early return using ok_or and taking advantage of the boxed error kind of like this:

...
Path::new(&f.path).file_name().ok_or("invalid file name")?;
...

If I understand correctly this returns early if there's a problem getting the file name and converts None to Error, correct?

How could I check behavior of this? How would I create a file name on OSx that would trigger this error?

[alice]

If you read the documentation, you will see that it:

Returns None if the path terminates in ..