Re: [Boost-users] Re: Iterators and maps [was Re: Where should STLsuggestionsgo?]

From: Ben Hutchings (ben.hutchings_at_[hidden])
Date: 2004-12-14 14:29:14

Jim Lear wrote:
<snip>
> I'm a little overwhelmed by these discussions, so forgive me for my 
> ignorance or incapacity to communicate.  My intention isn't to create 
> container-independent code.

Ah, good.

 > My intention is to create code for maps
> (and vectors) that looks like what I think code for maps (and vectors) 
> would look like.  Iterators look to me like pointers to linked lists for 
> vectors, and like pairs of pointers for maps.

Iterators are kind of like pointers into arrays, but possibly with some 
limitations on how you can use them; random-access iterators have no 
such limitations, whereas input and output iterators only allow you to 
go through the sequence once, forward, one step at a time.

Iterators over maps are bidirectional iterators, which are somewhere 
in-between: they can then be moved forward and backward over the 
sequence however many times you want, but only one step at a time.  They 
do not behave like pairs of pointers but like pointers to pairs.

(The standard iterator categories unfortunately combine requirements on 
traversal operations with requirements on access operations; if you want 
to know exactly what the iterator categories mean, look elsewhere.)

 > There is nothing wrong
> with these semantics, except they seem odd compared to some other 
> languages semantics for associative arrays (e.g. awk).  However, 
> overloading the operator[] in the map class to take iterators as a 
> parameter would allow iterators to be treated like keys, if one desires, 
> which to me seems more natural.  This would still allow one to iterate 
> over keys, mapped values, or pairs.
 >
> So in the absolute most abhorently poor code example (let's just call it 
> "meta-code"), the map operator would behave like:
> 
>     data_type &operator[](const iterator &i) { return i->second; }

I think that code should actually work if you only replace "data_type" 
to "mapped_type", but I suggest you don't go editing your <map> header 
just yet!

> Am I just too ignorant to make any sense? :-)  Maybe I'll crawl back 
> into my hole. :-)

No, but I think your concept of what an iterator is may be faulty. 
Since an iterator "points" to a sequence element (or to the end of a 
sequence) by itself; there is no need to combine it with a container to 
access that element.  You seem to want a regularity which doesn't really 
make sense, hence my earlier guess that you wanted to write container- 
independent code.

Ben.