From: George Ogata <g_ogata@...>
Date: 2002-07-08T20:43:57+09:00
Subject: Re: is there a better string.each?

Hi,

Yukihiro Matsumoto wrote:

> Hi,
> 
> In message "Re: is there a better string.each?"
>     on 02/07/08, Tom Sawyer <transami@transami.net> writes:
> 
> |the only point i am trying to make is that in so far as both account for
> |order and list, i want their exposed methodologies to be the same.
> |that's it. (whihc is why .each should work differently)
> 
>   * a string is not an array.
> 
>   * but still, a string can be seen as an array of characters
>     sometimes, hence str[0] returns the first character (a byte in the
>     current implementation) in the string.
> 
>   * text processing model of Ruby is besed on lines, not characters.
>     That's why I made "each" to be line oriented, not character
>     oriented.  I took usefulness over consisitency here.
> 
> So, if you want String consistent with Array, you need to express
> either:
> 
>   * why it is important over usefulness
>   * or line-oriented "each" is not usefull at all
> 
> good enough to bring incompatibility.
> 
> matz.

Well, "usefulness" depends on the application.  And it seems that 
sometimes, strings are better seen as arrays (not Arrays!) of characters, 
and at other times, as arrays of lines (and at yet other times, as arrays 
of words or paragraphs).

Since there doesn't seem to be a universally happy medium, perhaps the 
problem lies in assuming there is one.  I think the method "each" is the 
problem: it's ambiguous ("each what?").

Couldn't we have separate methods (as suggested earlier) #bytes, #chars (or 
#characters), #words (maybe), #lines, #pars (or #paragraphs) for String?  
We could have identical methods for the IO class, so IOs and Strings can be 
used interchangeably with these methods.  These methods could be both used 
to iterate over the object (when called with a block), or used to retrieve 
an Array (not array!) of the elements we're interested in (when called 
without a block).  E.g.:

s = "abc\ndef\nghi"
s.lines                            >> ["abc", "def", "ghi"]
a = []
s.lines do {|s| a << (s + 'xyz')}  >> nil (or maybe something else)
s                                  >> ["abcxyz", "defxyz", "ghixyz"]

Thus, io.lines &proc has the same effect as io.lines.each &proc (though 
would hopefully require less memory when say, reading in a large file, 
since the former doesn't need to read the whole thing into a gigantic Array 
first).

Now, what about #each?  Well, we've made it essentially useless for calling 
directly (since the others are more readable and unambiguous (agree?)).  
But there's still the interaction of String and IO with the Enumerable 
mixin.  What about this?:

We have methods corresponding to the iterator/collector methods to set the 
one that #each points to:  #useBytes, #useChars (or #useCharacters), 
#useWords, #useLines, #usePars (or #useParagraphs).

These could be defined in their own mixin module:

Module StringProcessing
  def useBytes
    class << self
      alias each bytes
    end
  end

  def useChars
    class << self
      alias each chars
    end
  end

  def useWords
    class << self
      alias each words
    end
  end

  def useLines
    class << self
      alias each lines
    end
  end

  def usePars
    class << self
      alias each pars
    end
  end
end

These would change the behaviour of #each for the String or IO instance 
they're called.  E.g.:

s = "abc\ndef\nghi"
s.useChars
s.collect {|char| frob char}  # iterates over chars
s.useLines
s.collect {|line| frob line}  # now it iterates over lines

The default behaviour would be to iterate over lines, so as to be backward 
compatible.  Well, except for my little, implicit wish that the record 
separators would be removed automatically.  E.g., I'd rather 
"abc\ndef".lines would return ["abc", "def"] than ["abc\n", "def"].  But I 
guess that's another war...  Still, even if the record separators stay, 
this'd be a pretty flexible String/IO model, wouldn't it?

Thoughts?