Resolve vendor directory symlinks

[FugiTech]

If vendor is a symlink to another directory (for example, __vendor), filepath.Walk will not descend into it, preventing --strip-vcs and --strip-vendor from working. Appending os.PathSeperator is enough to resolve the symlink.

I'm not sure if this is the ideal solution, but I copied the pattern from glide-vc which works as expected on symlink'd vendor directories.

[mattfarina]

Thanks for the PR. Before merging I'd like to understand what's going on here and why. Also, I want to make sure it works across nix/windows.

[FugiTech]

Sure! I made https://github.com/Fugiman/vendortest to try and make this easier to reproduce. I ran glide up -u -v -s in that repo to generate the current state, and then ran rm -rf vendor/**/.git to ensure submodules weren't committed.

This repo demonstrates the current behavior of "glide up -u -v -s does not strip nested vendor directories if vendor is a symlink", and should be able to be used to ensure the patch works on nix/windows (I only tested on mac).

---

As for why appending / to vendor works: Glide uses glide/path.StripVendor to remove nested vendor directories. That uses filepath.Walk to recurse through the vendor directory. filepath.Walk users os.Lstat to determine if a file is a directory, which does not resolve symlinks. As far as I can tell, what os.Lstat does varies by platform. However, for mac appending / causes os.Lstat to return a FileInfo where IsDir is true.

I could make this behavior more explicit by manually resolving the symlink before passing it into filepath.Walk if you want?

[mattfarina]

@Fugiman could you detect a symlink and resolve the location? The tailing / is something I don't really trust cross platform which we need to work. There is an IsLink function to detect symlinks. In the resolver there is also a function to return the base directory for a symlink. These may be helpful.

[FugiTech]

@mattfarina As requested, I now explicitly detect and resolve symlinks. I made the change to Vendor(), and then modified StripVcs() and StripVendor() to use that function.

I added another test to Vendor() to ensure the symlink behavior remains functional, and adjusted the scaffolding of the StripVcs() test to work with the format expected by Vendor().

Please let me know if you'd prefer to have the change in the Strip functions rather than Vendor and I'll move it. I also set it up to resolve nested symlinks, but I don't know if that is overkill or not.

[mattfarina]

@Fugiman thanks for the contribution!