Hi --

On Sat, 8 Feb 2003, Yukihiro Matsumoto wrote:

> Hi,
>
> In message "Re: class of $1, $2 in 1.8.0"
>     on 03/02/07, dblack / candle.superlink.net <dblack / candle.superlink.net> writes:
>
> |Doesn't that kind of cascading (i.e., subclassing and overriding) take
> |place all the time?  It seems very natural to me.  Also, until 1.8.0
> |this wasn't a problem; it's only now that the class of the objects has
> |changed.  I'm still not sure what the reason is for the change.  It
> |can't be just to catch people like me who override methods :-)
>
> Could you describe your error with concrete code?

Sure.  I'll give a simplified (but working) example, so as to avoid
the whole underlying scanf architecture:

  class SpecializedString < String
    def to_i(s)
      # specialized overriding of to_i
    end
  end

  s = SpecializedString.new("12345")
  m = /(\d\d)/.match(s)
  n = m[1].to_i
  p n * 10

With 1.6.8:

  120

With 1.8.0:

  matches.rb:22:in `to_i': wrong number of arguments(0 for 1)
    (ArgumentError)

(because m[1] is a SpecializedString object, so the specialized to_i
gets called).

Note also that if the implementation of SpecializedString changes,
then the behavior of #match also changes:

  class SpecializedString
    def initialize(s)
      @string = s
    end
    def to_str
      @string
    end
  end

Now String#to_i gets called in 1.8.0 as well as 1.6.8.

My thinking is: if $1,$2... are canonicalized to String, the behavior
wouldn't change, and a new SpecializedString can always be created
from a String.


David

-- 
David Alan Black
home: dblack / candle.superlink.net
work: blackdav / shu.edu
Web:  http://pirate.shu.edu/~blackdav