From: Paul Brannan <paul@...>
Date: 2002-02-14T00:18:46+09:00
Subject: Re: qualms about respond_to? idiom

On Wed, Feb 13, 2002 at 11:48:07PM +0900, David Alan Black wrote:
> I actually like the look and feel of this better:
> 
>   begin
>     obj.method
>   rescue NameError
>     # stuff
>   end

There are at least two problems with this:
  1) I may need to do lots of setting up before I ever get around to
     calling obj.method.
  2) obj.method may inadvertently call another method that throws a
     NameError; thus, getting a NameError does not necessarily mean
     that obj does not respond to method.

I don't like respond_to? either, because just because an object has a
method with a particular name doesn't mean that method has the same
semantics as I expect it to.  Including a mixin to indicate whether an
object is *intended* to both implement a particular set of methods and
to implement them with particular semantics seems more reasonable.  The
downside is that you can't extend an arbitrary object just to make it 
usable with a particular function; instead you must either subclass or
use a delegate class.

An example of where respond_to? can be a problem is in each(); with an
Array, it iterates over the array and yields each element in the array.
With a hash, it iterates over the hash and yields each key,value pair.
To get hash-like semantics out of an array, I must use
each_with_index(), but I can't just figure this out simply by knowing
that Array and Hash both respond to each().

Paul