Start trivial tests for dynamiccomponent

[catagits/CatalystX-DynamicComponent.git] / TODO

TODO, or the plan for making this stuff work right.

The goals:

1/ I want to be able to specify a Controller base class, like
   Catalyst::Controller::MessageDriven. That's where my
   message-handling actually lives, so I want to build controllers
   based on that, or whichever other class.

2/ We need a calling convention for the model implementations - in the
   absence of defined interfaces this should be simple unblessed data,
   but if we do have an interface, we should be able to put proper
   constraints on it.

3/ Given generated controllers where I expose a bunch of methods from
   the actual model classes, there needs to be some way to control
   which methods are dispatchable - "very public", if you like.

4/ I'd like to be able to write down an "interface", consisting of a
   set of messages my app is willing to handle, and have the framework
   (or the controller base class, probably) reject messages that
   aren't in that set.

5/ I'd also like to be able to say in that interface definition what
   types of object those methods expect as a payload.

6/ It'd be nice if exceptions from model methods were caught by the
   framework and converted into messages representing the error -
   i.e. to set an error status, and return the text of the exception.

All these things should be based on configuration. So, we've got
something to say what classes expose methods, which of those methods
are dispatchable, and which types are expected.

Possible Implementation

We can provide a number of controller base roles, for a number of
different message styles: YAML with type-tags, JSON bare hashes,
etc.

We can also provide a number of controller base roles, for a number of
policies for what model methods get reflected / are dispatchable etc..

These should be selectable by configuration, the default
controller class plus overrides for specific controllers.

The calling convention to model methods should be the "payload". That
might be a bare hashref for simple messages, or a blessed object of
some app-specific type if the message format contains first-class
objects, like tagged YAML.

The "interfaces" should be Roles, which don't provide any methods but
which "need" methods to be implemented. We then use the "needed"
methods as the set of dispatchable methods, given if we compile OK, we
know that the model class has all the required methods.
*** Do we require that the model class in question has already composed
the appropriate role, and just check that by introspection or do we just
add it ourselves? Adding it ourselves involves making an anon subclass
to avoid action at a distance, so the former if possibly preferable?

Types are implemented as type signatures on role methods, and we copy
them onto the actual methods on the model class at compile time.
**** I'm thinking this happens at role composition time..

Exception handling probably needs to be behaviour from the controller
base class, so that it can be specific to the type of message
serialization being used.