Version 1.13.7

Fix x64_mingw_24. Typo in PR #5170

(cherry picked from commit 6e0dcf48ee70314e42a65dcfd1f81c1494646d91)

Support :mri_24 platform

Current ruby trunk is versioned as "2.4.0".

% ruby -ve 'p RUBY_VERSION'
ruby 2.4.0dev (2016-11-15 trunk 56792) [x86_64-darwin15]
"2.4.0"

We'd like to run tests on some gems against ruby-trunk or 2.4 pre-releases, so here's a support for MRI 2.4.

(This is Ruby 2.4 version of #3865)

(cherry picked from commit aa9414e78f527ee2066d21783abb7d52a55f68a9)

4854 gemnotfound error

Fixes broken test

Wanting to help out, I was on the outlook for something easy to fix in bundler. I stumbled upon this issue #4854, which was already tackled by @b-ggs but somehow not merged yet. I went on to check whether I could do something. Here's what I've come up with:

@b-ggs' PR included everything to solve the issue, but introduced the use of `Bundler.app_cache` in the resolver. This works out just fine until you run `bundle plugin install`, which does not seem to require a Gemfile. `Bundler.app_cache` however calls `root`,  which then again tries to find a Gemfile. This obviously fails and raises a `GemfileNotFound` exception. The error message  "Could not locate Gemfile or .bundle/ directory" starts very similar to the one expected in the failing test ("Could not find"). This however, is a completely different error.

The solution is now is to rescue the GemfileNotFound error when it comes up to here.

All credit for solving the issue goes to @b-ggs and his PR here: https://github.com/bundler/bundler/pull/4865

Use correct hook name in comment

I checked `Bundler::Source::Path::Installer#post_install` - it's definitely `pre_install` and not `post_install` twice though I wasn't sure given the method name 😄

Specs for multiple platform bug fix in outdated.

This supersedes PR #4734, which fixed a bug but didn't have proper spec
coverage. Since that PR was submitted, a different commit (291a807c in
PR #5176) fixed a similar bug at the same time. That caused a merge
conflict in #4734. It was unknown if the same fix fixed both bugs,
since there were no successfully implemented specs. These specs show
that the bug was properly fixed.

These specs were mostly already implemented by @b-ggs, just needed a
little additional guidance.

This supersedes PR #4734, which fixed a bug but didn't have proper spec
coverage. Since that PR was submitted, a different commit (291a807c in
PR #5176) fixed a similar bug at the same time. That caused a merge
conflict in #4734. It was unknown if the same fix fixed both bugs,
since there were no successfully implemented specs. These specs show
that the bug was properly fixed.

These specs were mostly already implemented by @b-ggs, just needed a
little additional guidance.

Improve -rbundler performance

Along with https://github.com/rubygems/rubygems/pull/1801, speeds up `-rbundler` by ~5x

Profiled via

```ruby
# frozen_string_literal: true

REQUIRE_TIMES = {} # rubocop:disable Style/MutableConstant
module Kernel
  require "benchmark"
  alias_method :require_before_req, :require
  def require(path)
    ret = nil
    Benchmark.realtime do
      ret = require_before_req(path)
    end.tap do |t|
      if ret
        from = begin
                 caller[2]
               rescue
                 nil
               end
        ::REQUIRE_TIMES["#{path} : #{from}"] ||= t
      end
    end
    ret
  end
end

at_exit { puts REQUIRE_TIMES.to_a.sort_by(&:last).reverse.map {|k, v| "#{k} =>\n\t#{v}" } }
```