On Iteration

[Lutger]

Andrei Alexandrescu wrote:

> I consider changing a bit D's range model following the better
> understanding reflected in this article:
> 
> http://erdani.com/publications/on-iteration.html
> 
> If you have any thoughts and if you can help with the implementation,
> please let us know.
> 
> 
> Andrei

I thoroughly enjoyed the article, thank you in particular for making this 
material accessible to non computer scientists like myself. 

More questions than thoughts if I may, do feel free to ignore them, these 
are just things I wonder about:

- how to do ranges over a tree?
My first thought was that a tree would define preorder / inorder / postorder 
ranges, and then perhaps visitors for more complex algorithms. But in the 
'iterators must go' keynote you mentioned implementing traversal over trees 
by popFront() taking a parameter indicating the branch to go to. Can you or 
anybody else shed more light on how this will work out? What about graphs?

- ranges over immutable data structures
I tried making an immutable linked list and stack (not as simple as I 
thought) with range support. Ranges must be mutable though. So I take it 
that ranges are not usually supposed to be shared, rather they be 
consumable, local views into immutable data structures instead? 

This leads to a broader question: is it possible to model some support for 
concurrency in ranges? So that algorithms can possibly choose a particular 
implementation based on the concurrent properties of a range? Perhaps this 
relates to the separation between iteration and access as well.

- why is a UTF-string iterator bidirectional and why is that unexpected?

- Is there already a standard way to implement a default range for a 
container? 

I have tons more questions, but I'll try to work them out for myself first 
or post them in .learn.