Fix widget install and uninstall using hardlinks.

[miguelfrde]

Fixes #2206
Related to ongoing discussion on #2264
Related to #2258 (merge conflicts?)

@rolandwalker This fixes the widget problem without showing a popup dialog to install.

File.link seems to do the job of hard linking directories.

[rolandwalker]

File.link seems to do the job of hard linking directories.

Wow, cool find. This is innovative, and I am enthusiastic about helping and testing. However, because it is innovative, we should expect to take it slow.

File-hardlinks are perfectly ordinary; every experienced sysadmin uses them. Dir-hardlinks are exotic, people have been taught they don't exist. Some different rules apply. Therefore, to start with, I think they should be represented by two clearly different classes in Ruby.

Do you mind if I PR to your personal branch?

Also, @Peeja has been thinking about hardlinks in #2258. Perhaps he will join us here.

[Peeja]

I may be missing something, but I don't think Ruby's being fancy here. I think it's just handing down the link(2) call to the OS, and if it supports directory hardlinks then it won't fail.

[miguelfrde]

Do you mind if I PR to your personal branch?

@rolandwalker I don't mind.

@Peeja I agree with you

[phinze]

Hello good people!

Thanks so much for the work you're putting in to solve this problem! 👔

I have to admit; the hardlinks sort of make me nervous. If widgets are always self contained should we consider just doing a ditto into the directory and an rm -rf on uninstall?

Basically I'm talking about an Artifact::Copied type.

Perhaps that will be simpler and more reliable?

Though if others are convinced hardlinks are worth it I'm willing to be persuaded. 😀

[phinze]

Okay @rolandwalker has talked me into the merits of hardlinks (or file hardlinks at least) for solving problems like this.

Provided dir-hardlinks are properly supported in OS X, and since this would be limited to a small subset of Casks, I'd be willing to give this a shot.

What's the latest on this? @rolandwalker did you end up PR'ing that branch?

[rolandwalker]

I remain interested but have not pursued it yet.

The fact that Ruby does dir-hardlinks without blinking is cool. Fonts using file-hardlinks has been working great.

However, using dir-hardlinks in production requires plenty of due diligence:

• I don't think stat can return the same data structure as for file hardlinks. Currently we count number-of-hardlinks as a way of detecting whether a hardlink is managed by us. Will that work the same?
• Must cover concerns on whether the user can delete dir-hardlinks using familiar tools such as Finder, command-line rmdir, etc. If not, that's a showstopper.
• basically: explore all the many edge cases.

But widgets really don't work with symlinks. I have even tried duplicating the directory-tree of a widget and symlinking in each piece of content. Even that does not work.

So, please don't close!

[wamatt]

Just throwing this out there, after researching dir hard links.

Considerations/potential pitfalls

• Non OSX volumes
• Filevault issues
• Finder issues (Finder Cannot Find Files that have hardlinks selkhateeb/hardlink#4)

I agree with the point earlier that we dealing with an exotic/esoteric feature
http://unix.stackexchange.com/questions/75225/forcibly-create-directory-hard-links

Apple also crippled LN to take away the directory functionality. (probably for good reason).
http://unix.stackexchange.com/questions/75225/forcibly-create-directory-hard-links

While it's certainly a super cool hack, how wise would it be, making it a mainstream option? Perhaps marking it 'experimental', would be more appropriate?

[rolandwalker]

@wamatt yes, and "experimental" option is the current plan.

[fuzzy76]

I guess this can be closed because of #2206

[rolandwalker]

If it's not too inconvenient, I'd like to leave this open, because I still want to do something like

$ brew cask --method=hardlink activate

[alebcay]

Widget support has been removed as of #8751.