From: "Mauricio Fernández" <batsman.geo@...>
Date: 2004-08-24T19:47:52+09:00
Subject: Re: *with* block?

On Tue, Aug 24, 2004 at 07:24:07PM +0900, T. Onoma wrote:
> On Tuesday 24 August 2004 05:11 am, Mauricio Fern�ndez wrote:
> > you can make it 'setter-safe' (i.e. go through the normal attribute
> > setters instead of bypassing them and assigning to the iv.s directly)
> > trivially with Kernel#local_variables.
> 
> clever
> 
> > You'll have to make sure you are introducing new variables (not reusing
> > the ones from the outer scope) though, since otherwise there's no way
> > to know which locals you used in the with block, short of using
> > set_trace_func or doing  my_obj.with %{  ..... }.
> 
> > And this won't work with Rite.
> 
> why's that?

In Rite,

def foo
  [0].each{ a = 1 }
  p a  #=> 1
end

will work, which means that variables introduced in the block will
belong to the method, and the only way to create block local vars will
be using   local{|your,locals,here|   ..... } or adding them to the
parameter list, so you'd have to write the with construction
as follows:

  foo.with do |x,y|
     x = 1
	 y = 2
	 _
  end

Otherwise it would work only once in a given scope, cause when running
the second instance of  whatever.with{...}, the locals introduced in
the previous block would already be there and it would be impossible to
know what has been assigned in the block, unless you do some ugly hack
like assigning some NotSetValue to all those locals before eval'ing the
block and then verifying which ones have been set to something else.
But this wouldn't be thread safe (if you want several threads within
the same method/scope, that is)...

> p.s. are we going to get &{ ... } shorthand for lambda?

I'd search the RCRs to see if this has been proposed formally in the
past, but rcrchive.net seems to be down...

-- 
Running Debian GNU/Linux Sid (unstable)
batsman dot geo at yahoo dot com