On Mon, 25 Jul 2005, skaller wrote:

> On Sun, 2005-07-24 at 12:14 -0700, Stephane Glondu wrote:
>> On Sunday 24 July 2005 11:48, skaller wrote:
>>>> I strongly disagree. Look at source code of buffer.ml: no Obj.magic.
>>>> What do you mean by "efficiently"?
>>>
>>> Buffer only works for characters.
>>
>> You can make it work for any datatype by using an 'a array option instead
>> of a string.
>
> That fails to be 'efficient'. Would you use
>
> char array option
>
> instead of the existing Buffer???

The problem is that there is no one perfect structure.  bool array is 
incredibly inefficient- it's much better to use a bitfield.  Etc.

As for having a dynamic (resizable) array as a standard library, I think 
I disagree with this idea.  A standard applicative tree-based array, 
maybe.  But a dynamic array, like a mutable doubly linked list, encourages 
imperitive programming- when I think imperitive programming should be 
*discouraged*.  There are reasons, and time, when imperitive programming 
is necessary.  But I think they are few and far between- much rarer than 
people suppose.  But imperitive semantics impose implicit coupling- when 
you pass a mutable data structure to a subroutine there is an implicit 
contract between the caller and callee- either to modify in a known way or 
to not modify the value at all.  This increases coupling and leads to 
maintainance problems.

Brian