Remove performance degradation caused by DelegateClass

[JonRowe]

I think DelegateClass is the cause of the performance degradation you're seeing @myronmarston, give this a try. I tried doing this without directly specifying the methods but that had a similar performance failing.

[coveralls]

Coverage increased (+0%) when pulling 3b679e7 on refactor_metadata_no_delegate into 929dda2 on refactor_metadata.

[coveralls]

Coverage increased (+0%) when pulling 6d5d7f3 on refactor_metadata_no_delegate into 929dda2 on refactor_metadata.

[myronmarston]

I have a few concerns here:

• Several of these methods take blocks and you aren't properly forwarding them. (e.g. fetch and delete).
• Why only this set of methods? Hash has many other methods, I believe.

[JonRowe]

This was what was required to implement the MetaData functionality. Do we want it to behave entirely like a hash, or is it acceptable to provide only our own api? I'm open to restoring the block forwarding though.

[myronmarston]

I think it's worth looking into doing something like:

require 'forwardable'

class Metadata
  extend Forwardable
  def_delegators :@hash, *Hash.instance_methods
end

Maybe see how that benchmarks?

Alternately, something else I've been mulling is the idea of just using a raw hash, w/o wrapping it or subclassing it or anything...and using it's default_proc to rather than overriding store, and have it delegate to an outside MetadataCalculator (or whatever) object.

Not sure it's worth the effort, though :(. Ultimately, what I care about most is getting rid of the runtime extends that were being used as that causes method caches to be blown away and creates a perf hit.

[coveralls]

Coverage increased (+0%) when pulling 01b7b5b on refactor_metadata_no_delegate into 5fe8ea9 on refactor_metadata.

[JonRowe]

I added Forwardable to my benchmark/delegate_v_defined.rb benchmark, for me it's about half the time of delegated, but still 2x more than composed.

[coveralls]

Coverage increased (+0%) when pulling dda6d1e on refactor_metadata_no_delegate into 5b25672 on refactor_metadata.

[myronmarston]

@JonRowe -- I think I've decided against this (even though I initially got this ball rolling). Thanks for giving it a shot, though.