Re: [xsl] Sequential numbers in pure xslt, breaking the no-side-effect rule

[Abel Braaksma <abel.online@xxxxxxxxx>]

David Carlisle wrote:
> > I don't have that luxury, 
> >     
>
> Sometimes (like here) it's tempting (or even from a performance point of
> view perhaps a requirement) to say you need an imperative assignment
> with side effects and guaranteed execution order, but in a strong sense
> that is not true. That's Turing's (standing on Church and Godel and
> others) main insight: Any machine (or programming language) that can
> compute anything intersting at all can compute exactly the same
> algorithms as every other one. So no algorithm _requires_ state to be
> mantained as a side effect, it's _always_ possible to rewrite the
> algorithm so that any state required is explictly passed from function
> to function. 

Nice to see Turing coming along here. You have a clear point and you are 
right, of course. I just don't often look at the language in such a way.

> Now it may be that in practical cases the amount of state
> that would be required to be passed is large and you don't want to pass
> it through a function call interface, but well, that's the job of the
> optimising complier to sort out.  Saying that the _end user_ has to
> rewrite the algorithm in an imperative style in order to get an
> effective computation is just so 1960's:-)

and man, do I love the sixties!

But what I really meant was: if you have some function library, it's 
just easier to have something like "get-uuid()", instead of 
"get-uuid(get-uuid(get-uuid()))". But I understand (I think) your point 
in how it can be done with functional programming. I've just lingered 
too long in the imperative world and sometimes I just can't resist to 
try to bridge the two.

Thanks for your insights,

Cheers,
-- Abel