From: ahoward <ahoward@...>
Date: 2003-02-01T01:46:25+09:00
Subject: Re: String#+

On Sat, 1 Feb 2003, Chris Pine wrote:

<snip>
> I always assumed there was some good reason for it (even if it was a bit
> confusing, and probably the source of fully half of my newbie mistakes, and
> which requires a disproportionate amount of my tutorial to explain...).
> However, if there is a good reason for it, I still don't see what it is.
> (Mind you, I wouldn't be surprised if there was one, but at this point I'd
> also be surprised if there wasn't one.)
<snip>

there is a good reason IMHO!

here is a real eg. i encountered :

in ruby 1.6.8

  class CGI
    class Session
      def []= key, value
	...
	@db[ key ] = value.to_s
	...
      end
    end
  end

now, ONLY strings can be stored as keys or values of sessions.  no big deal
you say?  well, say you wanted to write CGI::Session::Pstore, which would
allow a session to cache arbitrary objects, not only strings like
CGI::Session::FileStore or CGI::Session::MemoryStore.

(actual code from ruby lib - which i had to modify)

  class CGI
    class Session
      def []=(key, val)
	unless @write_lock
	  @write_lock = true
	end
	unless @data
	  @data = @dbman.restore
	end
	#@data[key] = String(val)	  # original code
	@data[key] = val
      end
  end

now you move on and write CGI::Session::Pstore like

  http://raa.ruby-lang.org/list.rhtml?name=cgi-sess-pstore

but you worry that, in the next release the 'little modification' you made to
CGI::Session will adversely affect CGI::Session::FileStore and
CGI::Session::MemoryStore.

as it stands - 1.8 does _not_ cast it's value to string, and so i will be able
to remove my hack into CGI::Session.

moral :

  if you cast arguments inside a method you are preventing
  subclassing/extension and promoting white-box programming.

this holds in all languages, for example, if you wrote a C function which did
this

  int
  function (x)
    int x;
  {
    unsigned char uc = (unsigned char) x;
    ...
  }

users of this method would be perplexed as to why the argument went in as an
int, and all values > 255 seem to be handled identically?!  so you start
looking into the library code which breaks the whole 'black-box' idea libraies
should manifest, etc., etc, down a road fraught with peril...

-a



-- 

 ====================================
 | Ara Howard
 | NOAA Forecast Systems Laboratory
 | Information and Technology Services
 | Data Systems Group
 | R/FST 325 Broadway
 | Boulder, CO 80305-3328
 | Email: ahoward@fsl.noaa.gov
 | Phone:  303-497-7238
 | Fax:    303-497-7259
 ====================================