From: Kevin Ballard <kballard@...>
Date: 2005-10-21T22:52:00+09:00
Subject: Re: Unit testing an each function

Brian Schr�der wrote:
> Hello Kevin,
>
> it is clear that reject without a block does not work, it was more a
> question of visual parsing. Reject without a block has no meaning,
> because we need to tell what to reject. And telling this is done with
> a block. Your quite intelligent magic gives IMHO a wrong meaning to
> the dot operator. You seem not to call .even? on the result of a
> reject.
>
> Well effectively you do, but I read chained methods as doing a chain
> of succeding operations on a chain of objects, and that is not what I
> read when I read (1..100).reject.even? Therefore I proposed
> (1..100).reject :even?, because there it is made clearer that even is
> given to reject to do something.
>
> Nonetheless go ahed and package it as a gem. I just wanted to voice
> what was disturbing me with this syntax.

I understand your disturbance, but the syntax you propose makes the
primary use of this idiom impossible, which is

  [(1..5),(10..15)].each.each { |i| puts i }

What would you suggest, making

  [(1..5),(10..15)].each(:each) { |i| puts i }

behave that way? That looks even more wrong then what my code is doing.
And what if you have even further nesting?

  [[(1..3), (5..8)], [(4..7), (2..6)]].each.each.each { |i| puts i }

Would that be

  [[(1..3), (5..8)], [(4..7), (2..6)]].each(:each, :each) { |i| puts i
}

in your syntax? That's even *harder* to read.

Do you see any way to provide the proper behaviour without using the
syntax my code supports?