entry: convert hard link absolute paths to relative ones

[gdesmott]

When unpacking a docker image, some hard links may be pointing to, say,
/bin/gunzip. Those were failing the validate_inside_dst() check,
preventing to unpack the image.

Fix this by re-using the same logic as we are already using for actual
files, removing the leading '/' from the link target path.

We do this only if the absolute link is not an ancestor of the target
dir as we want to support this use case according to the existing
absolute_hardlink() test.

[alexcrichton]

Looks good to me! I think tests may be failing now though :(

[gdesmott]

All tests are passing here (Linux) and I don't have access to a macos system to debug this I'm afraid. :\

[gdesmott]

I got access to a macos box from a colleague. Here is the problem:

[src/entry.rs:485] &dest_canon = "/private/var/folders/v3/fb_smfzj3bn41k7jj7vfqqh80000gr/T/tarampqNh"
[src/entry.rs:488] src.components() = Components(
    [
        RootDir,
        Normal(
            "var",
        ),
        Normal(
            "folders",
        ),
        Normal(
            "v3",
        ),
        Normal(
            "fb_smfzj3bn41k7jj7vfqqh80000gr",
        ),
        Normal(
            "T",
        ),
        Normal(
            "tarampqNh",
        ),
        Normal(
            "foo",
        ),
    ],
)
[src/entry.rs:492] &src = "var/folders/v3/fb_smfzj3bn41k7jj7vfqqh80000gr/T/tarampqNh/foo"

As you can see Path::components is parsing /private instead of / as root. Is that normal or a bug in the parsing? (I have 0 experience with macos)

[alexcrichton]

Sorry I'm not sure, I've never seen this behavior before.

[gdesmott]

Sorry I got confused while debugging. The actual issue is test's Tempdir being created as /var/folders/... which is canonicalized to /private/var/folders/... as /var is a symlink to /private/var on macos.
The unpack dest dir was canonicalized but the not the target dir, hence the test failing.

I solved it by canonicalizing the tempdir as well.

[alexcrichton]

I think windows may now be failing?

[gdesmott]

god, working with paths isn't fun on Windows. canonicalize() returning the extended syntax version of the path makes everything much harder. Anyway, I think I managed to make it work .

[gdesmott]

That's really strange, the test is passing on my Windows box so I'm afraid there is not much more I can do. :\