[boost] Re: char[] support in Boost.Range

From: Thorsten Ottosen (nesotto_at_[hidden])
Date: 2005-05-17 04:28:50

"Eric Niebler" <eric_at_[hidden]> wrote in message 
news:428926DB.2090108_at_boost-consulting.com...
|
| Thorsten Ottosen wrote:
| > "Eric Niebler" <eric_at_[hidden]> wrote in message

| > | vector<bool> creates all kinds of problems because generic code can't
| > | make assumptions about the behavior of vector<T>. vector<bool> is widely
| > | regarded as a Bad Move. Dave is saying that treating char[] different
| > | than, say, int[] is inviting the same sorts of problems. It will make it
| > | difficult to deal with T[] in generic code.
| > |
| > | I agree with Dave.
| >
| > are these "problems" as serious as vector<bool> ?
| >
|
|
| Yes. It's exactly the same situation. You are giving collection<X>
| different semantics than collection<Y>.

I guess my objection is to the use "exactly".

If I have

template< class T >
class my_vec
{
    std::vector<T> vec;
};

then I might need  some traits for dealing with the bool case.

If I have

template< class Range, class OutIter >
void copy( const Range&, OutIter );

then I just need to be able to say what a Range means.

I think it would be wrong for the algorithm to assume anything; that would
not give the caller a chance to decide.

And that  is why Pavol said, let's allow

copy( as_array( rng ), out );
copy( as_string( rng ), out );
copy( rng, out );

If we don't wan't to give string literals a special meaning in the range 
library, then
we up with the rather clumsy

sub_range<string> r = find( rng, as_string( "foo" ) );

compared to

sub_range<string> r = find( rng, "foo" );

It is not obvious why you ever want to include the 0 in a literal, but if you 
want,
there should be a way to do so. Currently the isn't any support in the range 
lib,
but there should be.

| BTW, I find your use of quotes rather "disparaging". If you think this
| situation is different, please give /technical/ reasons. Thanks.

Yeah, sorry. I guess I was equally annoyed by the claim that we had a new 
problem
simply by stating vector<bool> was a problem. I don't see that close an 
analogy
and so the argument made by Dave could be used to to argue for anything.

I don't think the decisions of the range library was a big issue during the 
review;
Peter Dimov was, I think, one who said something along your opinion.

Anyway, it wasn't (and isn't) obvious  to me that having special cases in the 
range library is a bad idea---
as long as you can still explicitly ask for the other generic version.

br

-Thorsten