DynamicNode faked/neutered Arrays are unintelligible frankenbeasts

[atz]

DynamicNode tries too hard to be invisible and fails.

    expect(apo.roleMetadata.collection_manager.include?('foo@bar.com')).to be_truthy
    expect(apo.roleMetadata.collection_manager.respond_to?(:include?)).to be_truthy

The former succeeds, the latter fails.

  1) Hydrus::AdminPolicyObject should be able to create an APO object, whose APO is the Ur-APO
     Failure/Error: expect(apo.roleMetadata.collection_manager.respond_to?(:include?)).to be_truthy
       expected: truthy value
            got: false

So the object does not think it respond_to?(:include), despite the fact it just did respond to it. Cute, eh? This is on a 1.8.x release.

[atz]

I think this is the missing logic:
https://robots.thoughtbot.com/always-define-respond-to-missing-when-overriding

We need respond_to_missing?. We cannot really replicate all the 4-part conditional logic of method_missing:

• it would have massive cost, since mm actually includes a constructor,
• even if we could afford it, the constructor is dependent on having arguments that respond_to_missing? does not receive.

But, we can at least say we care about the methods defined by the val object. And we don't have to say where those calls would be serviced, just that yes they are not defined, but we will respond to them... somehow or other.

@cbeer, @mjgiarlo: thoughts?

[atz]

In practice, method_missing will respond 4 different ways:

• ALL assignment operations are accepted/attempted as new nodes,
• ANY operation with multiple arguments is accepted/attempted as a new node (w/ index),
• With an auto-constructed sub DynamicNode object,
• By handing off to val. This is the only route that will return a sensible NoMethodError.

respond_to_missing? won't have those args, so we cannot handle cases 2 and 3. But we can at least do 1 and 4.

Note: the specialized constructor methods invoked by method_missing really should be private.

[mjgiarlo]

@atz I don't see any reason not to encourage you!