Add MethodOverridingInitializer extension

[lnestor]

Describe the change

This PR adds a new extension - MethodOverridingInitializer. This extension allows existing hash methods to be overridden if a hash passed in to the constructor contains such keys. For example:

class OverridingHash < Hash
  include Hashie::Extensions::MethodOverridingInitializer
end

hash = OverridingHash.new(zip: 'a-dee-doo-dah')
hash.zip #=> 'a-dee-doo-dah'
hash.__zip #=> [[[zip a-dee-doo-dah]]]

Issue

References #338

Design Decisions

I made this its own extension in case people want to restrict which methods can be overwritten. Using this extension only, they can pass in a hash of only the methods they want overwritten and prevent overwriting other methods.

Possible Alternatives

This could be included in the extension MethodAccessWithOverride instead of being it's own extension.

[lnestor]

Some thoughts for discussion if need be.

[lnestor]

I made this its own extension in case people want to restrict which methods can be overwritten. Using this extension only, they can pass in a hash of only the methods they want overwritten and prevent overwriting other methods.

[lnestor]

This causes a warning to be shown when running the specs. If that's an issue it can easily be removed. I added this spec because the already_overridden? isn't needed in this extension, so I was testing the case of duplicate keys.

[michaelherold]

Great idea! A couple things I would like to see:

• A reduction of the duplicated code. Since the redefine_method is the same as in the MethodOverridingWriter, it should be extracted into a helper module so we don't have it defined in multiple places.
• We could ship this in a minor revision (3.6) as a new extension, but we shouldn't change the way that MethodAccessWithOverride works without a bump to 4.0. I think we could add a log message to the included hook for MethodAccessWithOverride if we want to make it a default in 4.0.

One additional thing that I'm thinking about here is that this somewhat replicates a subset of the behavior of MergeInitializer. I wonder if there is a clean way we can make it as an extension to MergeInitializer. I think we would have to add some sort of hook method that gets called for each key in MergeInitializer.

What are your thoughts about an approach like that?

[lnestor]

I like that idea. We could maybe make it an optional named argument. So, for a normal hash we would do something like MyHash.new(initial_hash, override: true), and for a clash you could do MyClash.new(initial_hash, parent, override: true). Then if you don't want to override, we keep the original way the method was called with MyHash.new(initial_hash).

Although MergeInitializer doesn't relate to method access of properties, so that is another thing to consider. If a person passes in the override: true argument to the constructor, that will create access methods, which is more of the spirit of MethodAccess. This new extension resides somewhere in the middle of MergeInitializer and MethodAccess.

[dblock]

Neat

[lnestor]

After implementing it both ways, I think it makes more sense to keep this under a MethodOverridingInitializer module. I don't really see this being used without MethodReader, but I can see people wanting to use MergeInitializer without overriding the methods.

[michaelherold]

This is looking good. I'll do the version bumping in the release commit, so I'm going to pop that off this branch, take the necessary changes from it, and merge outside of Github. You'll still get credit!

Thanks for the contribution.

[michaelherold]

Merged via cabb3e7