From: "Hal E. Fulton" <hal9000@...>
Date: 2003-07-16T21:59:28+09:00
Subject: Re: Keyword "with"

----- Original Message -----
From: "Brian Candler" <B.Candler@pobox.com>
To: "ruby-talk ML" <ruby-talk@ruby-lang.org>
Sent: Wednesday, July 16, 2003 7:03 AM
Subject: Re: Keyword "with"


> On Wed, Jul 16, 2003 at 08:27:04PM +0900, Mark J. Reed wrote:
> > On Wed, Jul 16, 2003 at 06:41:11PM +0900, Mauricio Fern�ndez wrote:
> > > > module Kernel
> > > >   def with(obj, &blk)
> > > >     obj.instance_eval &blk
> > > >   end
> > > > end
> >
> > This has apparently been proposed before.  The resistance to
> > including it may simply come from the fact that, from a general
> > design perspective, instance_eval is a Bad Thing(TM).  It violates
> > encapsulation and removes an object's ability to manage its own data.

As regards the encapsulation issue -- it depends on
how the real 'with' was implemented. If it were a
real language construct and not just a call of
instance_eval, then stuff inside the block could be
interpreted as calls to accessors and we could disallow
direct access to instance vars.

Then it just becomes a shorthand -- which is the whole
point after all, I think.

> Possibly. Also, doing
>
>   def bar
>     some_long_variable_name.instance_eval {
>       ... code ...
>     }
>   end
>
> means that 'code' loses access to the instance variables of the enclosing
> object, so it's not as useful as you might think.

That part can be solved even using the instance_eval
version of 'with' -- just pass in arguments as they
are needed. A little clunky, of course.

  def with(obj,*args,&blk)
    obj.instance_eval &blk
  end

  @foo, @bar = 34,45
  with(some_obj,@foo,@bar) do |x,y|
    @foo = x + y
  end

(Note that the block at the end is a special case --
normally a starred argument can only go at the end.)

The version of this I'm used to is a trifle different --
I added 'with' to Object so that I could always do
  obj.with { ... }

Same thing, of course.

I think a big reason 'with' has not been accepted is that
Matz remains unconvinced. It's rather a Pascalism, and I
recall he said he never liked Pascal.

What's more, I don't think it was ever an RCR.

Hal

--
Hal Fulton
hal9000@hypermetrics.com