Re: A class template for an N-dimensional generic resizable array

From: Giovanni Bavestrelli (gibav_at_[hidden])
Date: 2000-12-01 17:12:53

--- In boost_at_[hidden], Jeremy Siek <jsiek_at_l...> wrote:
> 
> On Thu, 30 Nov 2000, Jeremy Siek wrote:
> > Hi Giovanni,
> > 
> > Hmm, my intuition is that when dereferencing the 
Array<int,2>::iterator
> > and Array<int,3>::iterator I should get a subarray, not an 
element. I
> > would rather have to explicitly request a "flattened" view to get 
what you
> > have here.
> 
> Just something to back up this intuition:
> 
> It is natural to expect that a[i] be equivalent to *(a.begin() + 
i). For
> this to be true for the multi-dim array, the iterators for higher
> dimensional slices need to dereference to give subarrays instead of 
the
> flattened sections.

I agree with this, but if the number of dimensions grow, wouldn't 
that force you to use complex sets of iterators? I still have to 
finish reading Andrei Alexandrescu's October 98 CUJ article, but my 
first impression is that with increasing N (number of dimensions) it 
becomes complex. And if you return another iterator as you suggest, 
how would you use, for example, std::max_element to find the maximum 
element in a three dimensional slice of a five dimensional array? 
Would it not be very complex? But you are right, I need to consider 
this alternative, and see if I can come up with something simple and 
that scales well to big number of dimensions. Any suggestion is 
welcome.

> Ciao,
> 
> Jeremy

Thanks, Jeremy!

Giovanni
 
> 
> --------------------------------------------------------------------
--
>  Jeremy Siek                        www: 
http://www.lsc.nd.edu/~jsiek/
>  Ph.D. Candidate                  email: jsiek_at_l...
>  Univ. of Notre Dame         work phone: [phone]
> --------------------------------------------------------------------
--