Make os::make_absolute logic available for arbitrary Paths

[mitsuhiko]

Especially in web applications (or other apps that do not use the current working directory of the operating system) it would be useful to be able to use make_absolute against an arbitrary path. In Python that's provided by os.path.join. Unfortunately that also comes with negative security aspects (if the right hand side is absolute).

I would propose a Path::unsafe_join function that works like the Python one. As an extension it would also make sense to further improve how joining works for Windows.

Example implementation: https://gist.github.com/mitsuhiko/4974499

[mitsuhiko]

The current path joining logic is also broken for Windows. I started work on this in a separate pull request: #5023

[alexcrichton]

You probably have a much better idea of the scope of this than I do. How much of this is left to be done after your pull request?

[emberian]

I think path needs to be reworked a bit to do more than just filesystems, but I'm not sure how abstract it can get and not be useless.

[catamorphism]

clearing milestone - won't break backwards-compatibility

[lilyball]

My Path rewrite provides .join() functionality, but it doesn't do the relative testing that you seem to want. You can, however, check something like Path::new(".").is_ancestor_of(proposedRelativePath) to make sure that, not only is it relative, but it doesn't start with ../.

[lilyball]

The path stuff just landed in 40180cd. A relative-path-checking variant of path.join() can be written pretty easily, as I said above, but I don't think we need to provide such functionality in libstd.

[flaper87]

Visiting for triage.

Path has been rewritten and this sounds like doable in a saner way.

[l0kod]

cc #16507

[steveklabnik]

@alexcrichton how did this turn out with the Path RFC?

[alexcrichton]

We've removed make_absolute essentially and this would be an addition to the Path API (e.g. would require an RFC).

[l0kod]

Why does make_absolute was removed? It should be useful to be able to set a Path as always absolute.

Like said @mitsuhiko, the current join is dangerous and should be replaced with something else (e.g. splice).

[l0kod]

Normalization for paths is motioned in #22500.

However there seems to be nothing about unsafe join.
Something like AsPath::nest(&self, path: &AsPath) -> PathBuf could be a safe alternative to Path::join() and PathBuf::push().

[alexcrichton]

This function ended up being removed, so I'm going to close this in favor of a to-be-opened RFC issue to include this functionality.