The string passed as the cause is currently unused, so we don't need to
actually store it.

Proposing that we toss Metalua into the boneyard. Reasons:

1) It is hard-coded to look for Lua & Luac binaries, which are
obviously now pointing at Lua (5.2). I tried tweaking those hard-coded
lines with inreplace and had some success, but it still broke on
finding Luac. It complies ‘successfully’, but it won’t do very much.

2) There’s been no stable Metalua release for 5 years.

3) The HEAD build hasn’t worked in at least 8 months because the github
has moved away from compile scripts in favour of becoming Luarocks.

4) Honestly, interest in Metalua seems to be pretty low. If nobody has
noticed the HEAD build has been dead for 8 months…?

I feel like we’d be better scrapping this and pointing people at the
new Luarocks available for Metalua, which are on versions 0.7.2-1
compared to our shipped 0.5-rc2.

As we know, files cannot be moved across filesystems atomically. In that
case, FileUtils.mv will make a copy. But if we create the temp file in
the same directory as the target, we can avoid this and use File.rename
directly.

Additionally, the rename should be the absolute last step, so that the
original file is preserved if altering ownership and permissions fails.

This was added in d2ecfb9 / Homebrew/homebrew#12784 to fix a bug repotted in Homebrew/homebrew#12779, but
at that time we were invoking the editor via `system` rather than
`exec`.

Now we are using exec, and running it through an interactive shell seems
to cause other problems, for example, emacs does not suspend/resume
properly.

Fixes Homebrew/homebrew#32328.

You can use any number of "/" operators, just as you can "+".

Closes Homebrew/homebrew#31451.
Closes Homebrew/homebrew#32223.

Avoid problems like Homebrew/homebrew#32367.

This was added in 3a7a3619f7016c75c3e53e944221bf7f6354a63b to make the
tests pass, but this method is now stubbed out in the tests, so we can
assume $? will be non-nil.

Fixes Homebrew/homebrew#32410.

Closes Homebrew/homebrew#32394.

Long term we should probably rename the installer's "f" attribute to
"formula".

And don't uninstall them in-between. This should fix the issues where
if e.g. OpenSSL changes in a pull request then it's only the dependents
after it alphabetically would be built against the new version.

If ARGV.named.empty?, then ARGV.formulae.map(&:name) == [].

None of these subclasses share any behavior other than what is inherited
from RuntimeError, so we can just get rid of the superclass.

This class is never instantiated directly, and all subclasses
initializers have different signatures and do not rely on this.