[boost] Re: boost::filter_iterator Model

From: David Abrahams (dave_at_[hidden])
Date: 2004-12-11 00:38:27

David Abrahams wrote:
> Bob Bell wrote:
> 
>> Now consider a typical reverse loop, where we decrement backwards
>> through [b, e):
>> 
>> while (b != e) { --e; // use *e; }
>> 
>> No "begin sentinel" is required to make this loop behave as
>> expected. When b is created, it will point to the first element of
>> the filtered pseudo-sequence; this means that eventually, e will
>> become equal to b, and the loop will terminate.
> 
> Not if the filter predicate returns false for *b.
> 
>> In fact, it seems that without a begin sentinel, uses of b and e
>> are just as well-defined (or undefined) as equivalent ordinary
>> iterators. For example, for an ordinary sequence [b, e), --b is
>> well-defined only if there is something dereferencable there. For a
>> filter iterator lacking a begin sentinel, the same thing is true.
>> 
>> What do you think?
> 
> I think you overlooked something important.

Well, let me think about this again.  Maybe you're right.

The issue would be that you can't predict where the beginning of the
filtered sequence lies... but that's not true.  Imagine you have a
std::list<int> l and you build filter iterators b and e over [l.begin(),
l.end()).  Since b and e form a valid range, I see no reason you can't
iterate backwards from e to b.

Okay, I buy it.

-- 
Dave Abrahams
Boost Consulting
http://www.boost-consulting.com