Re: [boost] AlRangeExandrescu?

Subject: Re: [boost] AlRangeExandrescu?
From: Andrei Alexandrescu (andrei_at_[hidden])
Date: 2009-07-24 20:20:06

Stewart, Robert wrote:
> Andrei is far better able to defend his notion of Ranges than I,
> given the time he's spent working on the idea.  However, I don't
> yet accept as axiomatic that all ranges can be expressed as a
> pair of iterators.
> 
> The bottom line is if ranges can be made to very reasonably
> support all use cases, their other benefits (ease of
> specification, implementation, and composition) make supplanting
> iterators with ranges highly desirable.

Good points (also those I snipped too, I just have a high-level 
response). I'm pretty sure most of what ranges do, iterators can be made 
to do with enough effort and attention because iterators are lower 
level. The problem is, the effort in defining and manipulating such 
abstractions is just too high. (And programmers voted with their code: 
there just aren't that many awesome iterators out there.) For one, the 
designs feel off-key - for example, std::istream_iterator, with its 
useless "end" iterator, just hurts. Or if you want to express a random 
number generator or a mathematical series, how naturally are they 
modeled with iterators? And then how easy is it to manipulate all of the 
above?

One secondary issue is that designs with iterators are so unduly 
complex, we got used to the idea that a container offers ONE iterator 
(and one const_iterator). This is just odd if you look at it with a 
fresh eye. How about multidimensional containers? Should there really be 
only one way to span them? Ranges eliminate enough of the unneeded 
complexity to allow you to define several ranges that work on a 
container type.

Conversely, there are designs that are more efficient with iterators. I 
think those are few and far apart enough to not render me worried.

Andrei