Add Physical mode to realpath

[jaggededgedjustice]

This adds the 'Physical Mode' and 'Logical Mode' switches to realpath, which control when symlinks are resolved.

For other utilities that use the canonicalize function I have preserved the current behaviour.

[tertsdiepraam]

Nice work! Great to see this functionality become available for other utils as well. I just have some small suggestions and please update the doc comment on the canonizalize function to include the ResolveMode argument.

[tertsdiepraam]

Suggested change

	match fs::canonicalize(result) {
	Err(e) => return Err(e),
	Ok(path) => {
	result = path;
	}
	result = canonicalize(result)?;

[tertsdiepraam]

I think these options should override each other. See overrides_with

[tertsdiepraam]

I might be totally wrong here and it's a remnant of the existing implementation, but I feel like the concerns are not well separated between ResolveMode and CanonicalizeMode. I would expect something like:

enum ResolveMode {
    None,
    Logical,
    Physical,
}

enum MissingHandling {
    Normal,
    Existing,
    Missing,
}

But maybe that doesn't make sense for a reason I'm not seeing...

[jaggededgedjustice]

No, I think your suggestion is correct. The CanonicalizeMode is a remnant of the existing implementation and the options for handling symlinks should be grouped together.

[jaggededgedjustice]

I'm not sure why the windows tests are failing, as I don't have a windows box set up to test on. Any suggestions for potential fixes would be appreciated.

[tertsdiepraam]

I've previously tested some utils on wine, maybe that would work? I could also take a look at it later

[jaggededgedjustice]

Ok, I have discovered the cause for the test failures. When using the Physical mode, the path is passed to std::fs::canonicalize from the documentation

On Windows, this converts the path to use extended length path syntax, which allows your program to use longer path names, but means you can only join backslash-delimited paths to it, and it may be incompatible with other applications (if passed to the application on the command-line, or written to a file another application may read).

The \\?\ at the start of the path is the indicator for the 'extended length path syntax'. So we can either leave the function as is and change the tests so they pass on windows or strip out the \\?\ when running on windows and using physical mode.

[jaggededgedjustice]

Ok, I have a windows vm that I can test with and I made a change to remove the extended path length prefix. Now there's another failure, which also happens on my vm when I run cargo test but if I run the command directly then it works as it should, so I need to do more investigation.

[sylvestre]

Did you look how it performs again the gnu testsuite ?
I see "PASS: 171" when we are usually around 175

[jaggededgedjustice]

I had a look and the behaviour of readlink changed, I'll get it fixed.

[jaggededgedjustice]

The problem with readlink has been fixed now, so this should be good to merge.

[miDeb]

Thanks! The test failures are all because of a bug in a sort test that has since been fixed, so let's merge this!

[jfinkels]

Partly fixes issue #2253.