Re: null terminated strings

[Ken Anderson]

At 01:06 PM 1/16/2004 -0800, Tom Lord wrote:

>    > From: address@hidden (Paul Jarc)
>
>    > FWIW, I think that (preferably copy-on-write) shared substrings are
>    > valuable enough for performance 
>
>mutation-effects-both shared substrings are valuable as a feature of
>(extended) Scheme.
>
>A common idiom in string-manipulating programs is to manipulate and
>pass around triples (STRING START END).
>
>It's well worthwhile to make such triples a first-class type -- the
>alternative is to have to write lots of string manipulation functions
>in a style where they take three actual parameters (ideally with two
>of those being optional) to represent a single conceptual parameter.
>
>And if you have that abstract data type, since it is compatible with
>the RnRS requirements for the STRING? type, it may as well be a subset
>of the STRING? type.
>
>copy-on-write shared substrings are a performance feature --
>mutation-effects-both shared substrings are an improvement to Scheme.

Can you describe how this is an improvement to Scheme?  I don't think i've ever 
wanted to do this.   In Java, which does copy-on-write i often find myself 
carefully copying the substrings so they don't share structure.  This is 
because of things like:

- i don't know how long the underlying string (char array actuall) is.  It can 
be much longer than the last line i've read.
- many fields are a relatively small set of values so interning them helps.
- some fields become something besides strings, such as numbers.

Java only has one kind of string, which is fairly heavy weight.  For example, 
the string "" takes 36 bytes:
> (describe "")

 is an instance of java.lang.String

  // from java.lang.String
  value: address@hidden
  offset: 0
  count: 0
  hash: 0

So, it might help to have several representations for strings.  In Gule, of 
course, you can play even more games because you have C underneath.

 From the applications i've seen, interning strings has been more useful