Re: Coercions and custom type parameters

On Thu, Jul 24, 2008 at 16:16:28 -0400, Charles Alderman wrote:
> I guess that if you're trying to prevent this as an action at a distance, 
> 'deep_coerce' wouldn't really be acceptable either.  Wouldn't that just be 
> a more explicit warning of an action at a distance?

Declaring an attribute with a parameterized type:

	has foo => (
		isa => "ArrayRef[Foo]",
		coerce => 1,
	);

has a specific behavior right now, it enables only the coercions on
the type "ArrayRef[Foo]".

If this started coercing using 'Foo's coercions we break
compatibility, and deining a coercion on 'Foo' affects types that
parametrize on it.

deep_coerce is different because:

	a. it doesn't break existing code
	b. it's explicit about the fact that it's accepting coercions
	from the type parameter, so a developer would know to look
	there, instead of looking for coercions on 'ArrayRef[Foo]'.

> If moose type constraints become too complicated, I guess at some point, a 
> designer/developer just needs to turn the parametrized type into its own 
> class.  So, maybe I'm answering my own question here.

Custom parametrized types could still make sense though:

	MyCustomType[Foo]

is a useful construct for custom container types, for functors, etc
etc, so deep_coerce *is* a useful property to have

> At any rate, I'll look more at the mailing list archives...

I think this was only discussed on IRC actually

-- 
  Yuval Kogman <[email protected]>
http://nothingmuch.woobling.org  0xEBD27418