From: Ryan Pavlik <rpav@...>
Date: 2003-11-09T10:36:42+09:00
Subject: Re: Managing metadata about attribute types

On Sun, 9 Nov 2003 10:07:50 +0900
dblack@wobblini.net wrote:

> On Sun, 9 Nov 2003, Ryan Pavlik wrote:
<snip>
> > As in the previous post, this is possible, but there is no incentive
> > to do so.  If you want to break your own code, there are quicker and
> > easier ways.
> 
> You must not be a big fan of class methods :-)

I'm not sure what your point is here.  The only conceivable point I
can see is that class methods are singleton objects of type Class, and
that if I ask for a Class, I may get one, but it won't be the exact
API of Class if I use class methods.

You're probably just trying to be funny---which is fine---but the joke
does overlook a few points I've made, including the one you pasted
beneath.

Adding methods does not break type checking.  An extended object is
still the object you wanted.  (Overriding and replacing an existing
class method might break something, but this is my point in the above
quote---there's not a lot of incentive to break things.)

This is coupled with a previous point:  you will never ask
specifically for something you don't know about.  For instance, your
code might ask for a Foo, but if it doesn't know about a Bar, a
subclass of Foo, it will not ask for it.  Singletons fall under this;
they are still their original class, but with things added.

(Of course you _can_ remove or incompatibly redefine things.  This is
usually bad.  For instance, when CGI used to return singleton Strings
with #[] redefined in an incompatible manner without any information
on the fact.  This caused a lot of problems.)

Anyway, you will obviously never accept my position, and I've had too
much success with strongtyping to give it up, so add any closing
points you wish to make and let's wrap this one up so we can get back
to doing something productive... like writing ruby code. ;-)

-- 
Ryan Pavlik <rpav@mephle.com>

"You're about a thousand miles of *wrong*." - 8BT