From: Christoph <chr_news@...>
Date: 2002-04-30T19:45:15+09:00
Subject: Re: [RFC] Ranges - enable (10..-10)

"David Alan Black" <dblack@candle.superlink.net> wrote in
....
> In [ruby-talk:8993], matz said:
>
>   The truth are:
>
>     * A range enumeration cannot go back, it's upward only.
>
>     * But A range with end point less than start point is valid.
>
>     * Thus 1..-1 is a completely valid range object.
>
>     * Array and String internally convert negative end point into
>       size+n, where n is a negative number.
>
>     * So range n..-m is interpreted as n..(size-m), thus works fine (if
>       n <= (size-m)).
>
> I can't claim I've totally grasped why a range is upward by
> definition... but apparently it is, so that's *my* problem :-)  You'll

You might find the following example somewhat enlighting?

---

class Class
  alias succ superclass
end

p (Fixnum..Object).to_ary # => [Fixnum, Integer, Numeric, Object]

begin
    (Fixnum..UnboundMethod)  # Fixnum and UnboundMethod are not comparable
rescue ArgumentError => mes
    p mes                    # => #<ArgumentError: bad value for range>
end

---

Given the fact that  empty Ranges ala (1..-2)  and (1...-2) are valid
construct, it
would makes to extend the Range construction and declare


(Fixnum..UnboundMethod).to_ary == [Fixnum, Integer, Numeric, Object]
(Fixnum..Float).to_ary == [Fixnum, Integer, Numeric]

# and

(Fixnum...UnboundMethod).to_ary == []
(Fixnum...Float).to_ary == []

# or (?)

(Fixnum...UnboundMethod).to_ary ==  [Fixnum, Integer, Numeric]
(Fixnum...Float).to_ary ==  [Fixnum, Integer]


Note I am not at all lobbying for such a behavioral change.

/Christoph