Re: [boost] Re: any interest in a foreach looping construct?

From: Daniel Wallin (dalwan01_at_[hidden])
Date: 2003-10-12 06:12:09

At 23:40 2003-10-11, Eric Niebler wrote:
>Daniel Wallin wrote:
>>   1. It operates on containers rather than iterators.
>
>Yes, containers, arrays and null-terminated strings. What is your 
>objection exactly? If you are saying that iterating over a container is 
>wrong *in principle*, then I disagree -- it's often exactly what you want. 
>If you are saying that there should be a way to iterate over a 
>sub-sequence, then I agree.

I just prefer to always specify ranges explicitly, mostly for consistency. 
I also
dislike interfaces which tries to handle every possible use case.

   BOOST_FOREACH(int i, iter1, iter2)

Would be better IMO.

>>   2. It reevaluates the container expression on every iteration.
>
>True, but the "container expression" must be an lvalue, which makes it 
>difficult to use incorrectly. The container expression cannot be a 
>function call that returns a temporary, for instance.
>
>You may be wondering about in_range() above -- it is a function that 
>returns a temporary, yet it is being used as a container expression. 
>in_range() returns a special type, and BOOST_FOREACH accepts it even 
>though it is not an lvalue. However, the *arguments* to in_range() must be 
>iterator lvalues and not temporaries returned by functions.
>
>I understand your concerns, however.  There is still the danger for this:
>
>BOOST_FOREACH( int i, in_range( ++iter1, iter2 ) )
>                                 ^^^^^^^ oops!
>
>I'm not sure how to guard against that.

Yeah.. So when using in_range() you'd still need to name your iterators, 
and also
place them in the "wrong" scope. I'd prefer

   for (iter_type iter = x.begin(), end = x.end();
      iter != end; ++iter)
   {}

over

   iter_type begin = x.begin();
   iter_type end = x.end();

   BOOST_FOREACH(xxx i, in_range(begin, end))
   {}

any day.

---
Daniel Wallin