Tracking Issue for `path_trailing_sep`

[clarfonthey]

Feature gate: #![feature(path_trailing_sep)]

This is a tracking issue for the Path::has_trailing_sep and related methods, which enforce whether a Path or PathBuf ends in a trailing separator.

Public API

impl Path {
    fn has_trailing_sep(&self) -> bool;
    fn with_trailing_sep(&self) -> Cow<'_, Path>;
    fn trim_trailing_sep(&self) -> &Path;
}
impl PathBuf {
    fn set_trailing_sep(&mut self, suffix: bool);
    fn push_trailing_sep(&mut self);
    fn pop_trailing_sep(&mut self);
}

Steps / History

(Remember to update the S-tracking-* label when checking boxes.)

• ACP: ACP: PathBuf::has_dir_suffix libs-team#335
• Implementation: Add Path::has_trailing_sep and related methods #142506
• Final comment period (FCP)¹
• Stabilization PR

Unresolved Questions

• Should we offer the set_trailing_sep API or push_trailing_sep/pop_trailing_sep separately? Right now, both are offered to see which is preferred.
• Is trim_trailing_sep necessary? It wasn't in the original ACP.
• trim_trailing_sep will leave / as /, but has_trailing_sep returns true for /. That seems like a possible bug?

Footnotes

1. https://std-dev-guide.rust-lang.org/feature-lifecycle/stabilization.html ↩

[valarauca]

Shouldn't check box 2 be marked as complete?

[yoanlcq]

Hi, maybe this is known already, but it looks like the trailing separator might not preserved by some other API calls, which is counter-intuitive.

I've seen this with strip_prefix() at least, but there might be others. (Playground link) :

#![feature(path_trailing_sep)]

use std::path::{PathBuf, MAIN_SEPARATOR};

fn main() {
    let mut ab = PathBuf::from("a/b");
    ab.set_trailing_sep(true);
    assert!(ab.to_string_lossy().ends_with(MAIN_SEPARATOR)); // Will succeed
    
    let b = ab.strip_prefix("a").unwrap();
    assert!(b.to_string_lossy().ends_with(MAIN_SEPARATOR)); // Will panic
}

[ChrisDenton]

Yes, that's one reason for these functions. Components ignores trailing separators and since most functions are specified in terms of components that means they don't see the trailing separator.

The trailing_sep functions give a workaround so you can test if a trailing separator exists before you operate on the path and add it back if so after you finish.

This is not ideal but we'd need to redesign the Path API in order to do better.

[clarfonthey]

That said, it's quite surprising that strip_prefix will affect the suffix at all. I would expect this to work, personally.

[ChrisDenton]

I mean, I think components directly or indirectly leads to surprising behaviour in a number of cases. Can we fix it in certain cases without it being a breaking change™? We should look at if that's possible. But also we'd need to ensure any inconsistency doesn't lead to more problems (hopefully not for strip_prefix).

Tbh I would like a new Path API that fixes this category of problems along with other issues.