[boost] Re: string algorithms review

From: David Abrahams (dave_at_[hidden])
Date: 2003-10-24 12:23:58

Pavol Droba <droba_at_[hidden]> writes:

> On Thu, Oct 23, 2003 at 11:21:54PM -0400, David Abrahams wrote:
>> Pavol Droba <droba_at_[hidden]> writes:
>
> [snip]
>  
>> >> ?? We're not talking about making a copy, since the result differs
>> >> from the input.  In general, you pay the same price in functional
>> >> languages.
>> >> 
>> >
>> > So there is realy NO copy performed in Haskell when evaluating the function,
>> > it is a transformation. And this is much closer to inplace version,
>> > in C++ then to copy one.
>> 
>> Gimme a break, man!  In pure FP languages data is immutable, so
>> there's no way to make inplace changes.  In Haskell you have to go
>> into a monad to escape this restriction.
>>
>
> I know that this discusion is out of topic, quite a bit, but here is my
> explanation what I mean, when I say, that nothing is copied:
>
> Most of modern FP languages, perform the execution in terms of graph-reduction machine.
> High level syntax is precompiled to a extended version of lambda calculus
> and this is then executed.
>
> State of an execution is represented by a term, which is represented by a graph.
> This graph is being reduced as the lambda calculi reduction rules are
> applied.
>
> There is nothing like a stack or variables in this representation. So unless
> some subterm is bound to more then one expression, it can be safetly modified
> in-place. And thats usualy what is being done.

This is my last post on the matter:

     Yes, I know that's often what's being done.  However, we never
     said the subterm was only going to be bound to one expression.
     You just added that.  The two-expression binding case, of course,
     is exactly the one where immutability of data makes a difference
     in safety.

Over and out,
Dave

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com