The old action_class code was doing some magical stuff with the provider
accessor in order to determine if the class was supposed to be a custom
resource or not and have the action_class autovivifying accessor return nil
in cases when the resource wasn't a custom resource and implementing
inheritance by walking back up the tree in ways that were difficult to
grok.
This removes the magic from the provider so that there's no longer any
accessor that magically short-circuits to nil if the resource is not
supposed to be a custom resource.
There is now a simple inherited API for `Chef::Resource.custom_resource?`
which just defines if the class is a custom resource or not. Since both
`action` and `action_class` call `define_action_class` they both wind up
setting this boolean on the class, which is then inherited to subclasses
automatically, which eliminates the need to walk up the hierarchy.
The superclass.respond_to?(:action_class) checks have also been rendered
unnecessary by removing the code that walked up the inheritance
hierarchy and also because Chef::Resource is never going to be a
custom resource itself, so will never call `define_action_class` so from
inside of `define_action_class` you can always rely on the superclass
being a resource and implementing `custom_resource?` and `action_class`.
The wiring for picking the provider is now moved explicitly to the
ProviderResolver -- even though custom resources hardcode a 1:1
resource-to-provider mapping. This reads much clearer to me than the
magical wiring to the provider accessor off of the instance.
The bug that this fixes was that the way the magical accessor
nil-or-action_class was implemented the old way of defining action
helpers with class_eval broke:
```ruby
action_class.class_eval << EOM
def a
"foo"
end
EOM
```
If that came before any action_class-with-a-block or action declaration
and the resource did not inherit from another custom resource then the
action_class would not be created and it would return nil, which was an
API which the magical wiring in the provider accessor required.
Signed-off-by: Lamont Granquist <lamont@scriptkiddie.org>