Fix hook install in submodules

[sj26]

Joining the repo root doesn't account for .git files which have a gitdir: ... redirect, but Overcommit::Utils.git_dir compensates for that.

Before

my-submodule (master) % overcommit --install
Installing hooks into .../my-repo/vendor/my-submodule
.../lib/ruby/2.4.0/fileutils.rb:230:in `mkdir': File exists @ dir_s_mkdir - .../my-repo/vendor/my-submodule/.git (Errno::EEXIST)
	from .../lib/ruby/2.4.0/fileutils.rb:230:in `fu_mkdir'
	from .../lib/ruby/2.4.0/fileutils.rb:208:in `block (2 levels) in mkdir_p'
	from .../lib/ruby/2.4.0/fileutils.rb:206:in `reverse_each'
	from .../lib/ruby/2.4.0/fileutils.rb:206:in `block in mkdir_p'
	from .../lib/ruby/2.4.0/fileutils.rb:191:in `each'
	from .../lib/ruby/2.4.0/fileutils.rb:191:in `mkdir_p'
	from .../gems/overcommit-0.41.0/lib/overcommit/installer.rb:80:in `ensure_directory'
	from .../gems/overcommit-0.41.0/lib/overcommit/installer.rb:33:in `install'
	from .../gems/overcommit-0.41.0/lib/overcommit/installer.rb:22:in `run'
	from .../gems/overcommit-0.41.0/lib/overcommit/cli.rb:122:in `block in install_or_uninstall'
	from .../gems/overcommit-0.41.0/lib/overcommit/cli.rb:120:in `each'
	from .../gems/overcommit-0.41.0/lib/overcommit/cli.rb:120:in `install_or_uninstall'
	from .../gems/overcommit-0.41.0/lib/overcommit/cli.rb:22:in `run'
	from .../gems/overcommit-0.41.0/bin/overcommit:49:in `<top (required)>'
	from .../bin/overcommit:23:in `load'
	from .../bin/overcommit:23:in `<main>'

After

my-submodule (master) % overcommit --install
Installing hooks into .../my-repo/vendor/my-submodule
Successfully installed hooks into .../my-repo/vendor/my-submodule

👍

[sj26]

I added a spec here, too.

[trotzig]

Thanks! Code looks good, but it seems like this isn't working correctly on windows machines. Do you have any ideas on what could be wrong?

[sj26]

Windows doesn't seem to understand File.expand_path. Or, rather, ruby doesn't recognize "C:/" as a directory root. I'll try File.realpath.

[sj26]

Oh, or I specified an absolute path instead of a realtive one and git didn't understand that. Still trying.

[sj26]

So I think I fixed the Windows problem introduced by the specs here, but there's a seemingly-unrelated problem with Ruby 2.1 and parallelization?

[trotzig]

Yeah, the windows tests are known to be flaky. I'm assuming this is not your fault. Sorry for the delay here.