`brew bottle` already complains and this is a niche case.

References Homebrew/homebrew#21374.

Closes Homebrew/homebrew#21506.
Closes Homebrew/homebrew#21508.

The formula locks used by the installer and commands like link and
unlink are backed by open files and flock(). The open file descriptors
are thus leaked to any subprocesses. This can result in weird behavior
in programs spawned from formula that do not expect to inherit these
descriptors.

Fix this by setting close-on-exec on the lock file descriptors.

Fixes Homebrew/homebrew#21486.

Since xcrun doesn't work on CLT-only systems. Querying HOMEBREW_PREFIX
first before xcrun is much faster for that case.

Fixes Homebrew/homebrew#21417.

Closes Homebrew/homebrew#21390.

Closes Homebrew/homebrew#21385.

Signed-off-by: Adam Vandenberg <flangy@gmail.com>

eql? should not depend on the hash value as hash values of uneql objects
can collide, but eql values may only collide for objects that are
actually eql.

Further, python dependencies are uniquely identified by the combination
of the name and imports attributes, so there is no reason to involved
the expensive binary computation for simple equality checks.

Fixes Homebrew/homebrew#20840.

This was worded in a confusing way, and made an assumption about future
OS X versions that we don't know to be true.

Also, it's probably wrong to state that the CLT location is dependent on
the OS X version; surely there will be a newer CLT for 10.8 that gets
installed in the same location as on 10.9. But we don't know any better
yet.

I'm tired of not remembering if start_with?/end_with? are portable, so
just add them both if they're not defined.

It is supposed to accept a variable number of prefixes, and also to
check if they are convertible to strings. This matches behavior
documented in RubySpec.