From: "Hal E. Fulton" <hal9000@...>
Date: 2001-05-05T12:23:48+09:00
Subject: [ruby-talk:14699] local/nonlocal block vars

Well, I don't think I ever put in my (full) two cents...
and frankly I got lost in the details when this was
discussed before.

This was one of the longest threads on the list,
right up there with the "Ruby M*scot" discussion
and "101 misconceptions about OO" or whatever
that was.

So forgive me here...  

while horse.dead? do
  horse.beat
end


Here's my take on this. 

I've numbered the points so that you can shoot them
down more easily. :)

1. Sometimes we do want nonlocal block parameters.
2. Sometimes we don't.
3. Sometimes we do it accidentally. Oops!
4. IMO, the <> notation is ugly. 
5. Somebody suggested a "with" notation which was
not so bad.
6. There was talk about prefixes. Ehh. I like this 
somewhat...

Here's how I would do it.

7. Use a Symbol constant to represent a variable from
the outer scope. This has the advantages of the "prefix"
idea without reusing a prefix like * or &.

8. If you have a Symbol, you use a variable from the 
outer scope. If not, it's local.

9. Disadvantage: Break old code? Waaahhh! No,
I don't like this part either. 

10. And here is the icing on the cake. Let me stir a little 
additional controversy by adding this rule:  "If a Symbol
appears *first* in a block parameter list, the variable in
the outer scope springs into existence then." 

11. Why this rule? No need to assign nil prior to the 
block. Also because "importing" and "exporting" variables
from the block are somewhat separate issues.

Continuing Johann's example:

  # obj and i are *not* defined here

  data_objects.each_with_index {|:obj, :i|
    if not obj.valid?
      break
    else
      obj.do_something
    end
  }

  # But now they are...

  if not obj.valid?
    print "The object at index ", i, " was invalid.  Aborted."
  end

========
OK, that's probably at least two cents there. As Bill Shakespeare
said, "Cry havoc, and let slip the dogs of war!"

Hal