ActiveSupport's Hash#reverse_merge doesn't get along with Hashie::Mash

[ajvondrak]

I don't know how much of a concern ActiveSupport compatibility is for you, but I thought that this behavior was surprising:

[1] pry(main)> require 'hashie'
=> true
[2] pry(main)> require 'active_support/all'
=> true
[3] pry(main)> params = Hashie::Mash.new(a: 1, b: 2)
=> {"a"=>1, "b"=>2}
[4] pry(main)> params[:a]
=> 1
[5] pry(main)> params['a']
=> 1
[6] pry(main)> params.merge(a: 2)
=> {"a"=>2, "b"=>2}
[7] pry(main)> params.reverse_merge(a: 2)
=> {:a=>2, "a"=>1, "b"=>2}

This is of course a consequence of ActiveSupport defining reverse_merge as

def reverse_merge(other_hash)
  other_hash.merge(self)
end

Here, other_hash is of class Hash, so of course Hash#merge doesn't play with Hashie::Mash's key conversion. Not sure how you'd fix it efficiently (Hash#merge being an MRI builtin), but maybe monkeypatch in a Hashie::Mash#reverse_merge like

def reverse_merge(other_hash)
  other_hash.each do |key, value|
    self[key] ||= value
  end
end

Doesn't capture block_given? behavior or anything, though.

Another workaround for now is just

[8] pry(main)> params.merge(a: 2) { |key, old, new| old }
=> {"a"=>1, "b"=>2}
[9] pry(main)> params.merge(a: 2, c: 3) { |key, old, new| old }
=> {"a"=>1, "b"=>2, "c"=>3}

or doing the ||=s by hand.

[ajvondrak]

Looking at the source for Hashie::Mash now, I realize that you're already defining a custom #merge method, so...seems like #reverse_merge isn't too far of a stretch. Could even just do it like this:

def reverse_merge(other_hash)
  Hashie::Mash.new(other_hash).merge(self)
end

[dblock]

We generally like not to have surprises with ActiveSupport. So feel free to PR this, and thank you.