Add support for creating symlinks. about go-billy HOT 6 CLOSED

@jmalloc If we add Symlink to the billy.Filesystem interface, we'd need to bump version to v3. So my proposal would be adding Symlink function to implementations in v2, and defer exposing it in billy.Filesystem.

Mimicking os.Symlink as much as possible is definitely the way to go. PRs are welcome ;-)

from go-billy.

I have a question about the behaviour of the subdir filesystem and how best to implement Symlink() there. First some background; subdir.New() doesn't create the subdirFs.base sub-directory on the underlying filesystem. As best I can tell, this doesn't usually present any problem because:

• Filesystem.Create() creates all parent directories when the O_CREATE flag is set.
• There is no Filesystem.Mkdir() only MkdirAll(), and hence no way that directory creation can fail just because subdirFs.base does not exist.

This presents a small problem for Symlink(), which always fails because the base directory is never created. In light of that, should I:

1. Create subdirFs.base if it does not exist. This most closely mirrors the behaviour of the standard library insofar as Symlink() will fail if anything deeper than base does not exist.
2. Create all necessary parent directories in all FS implementations. This more closely mirrors the behaviour of Filesystem.Create().

I'll proceed with option (2) for now to best maintain consistency within Billy's API. Please let me know if some other behaviour is more desirable.

Note: Just like osfs.OS.Create(), any directories created will remain if symlink creation fails.

As an aside, my comment about not being able to create symlinks to non-existent files with os.Symlink() appears to be a blatant lie ;)

from go-billy.

@jmalloc SubdirFs should just delegate the Symlink call to the underlying filesystem if it does implement the Symlink function, or return an error otherwise.

Then the underlying filesystem should create directories as needed.

from go-billy.

@jmalloc Are you working on this? Feel free to create a partial PR to get further feedback, we can also help completing it.

from go-billy.

I am. I have implementations for theos and subdir filesystems, but I haven't yet found time to complete the memory-based implementation. I'm submit a PR with what I have so far when I get a chance.

from go-billy.

@jmalloc Thank you! We'll review it soon and figure out how to move it forward.

from go-billy.