Re: [ublas] Transition to default handling of possible lvalue aliases

From: David Abrahams (dave_at_[hidden])
Date: 2005-06-25 19:39:13

Ian McCulloch <ianmcc_at_[hidden]> writes:

> David Abrahams wrote:
>
>> Ian McCulloch <ianmcc_at_[hidden]> writes:
>> 
>>> The containers have some different behaviour, but then so do
>>> vector/deque/list.  That doesn't stop us writing generic code,
>> 
>> Wanna bet?  When was the last time you wrote a generic function
>> designed to work on "any container?"
>> 
>
> Heh.  Well, simple functions work fine (eg, forwarding to an iterator-based
> algorithm :-).  

That's iterators, which have a very well organized and well-defined
taxonomy of concepts.  There's no such thing for containers.

> But it would be much simpler in this case, because we have
> STATIC_ASSERT(is_noalias<T>) and enable_if<is_noalias<T> >.  If we
> could do that for containers, with metafunctions like
> is_iterator_invalidated_on_insert<T> and so on, writing complex
> generic functions for containers would be easier.

But still very, very messy.

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com