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

From: gbavestrelli_at_[hidden]
Date: 2000-12-06 17:23:08

--- In boost_at_[hidden], Beman Dawes <beman_at_e...> wrote:
> At 12:30 PM 12/4/2000 -0500, Jeremy Siek wrote:
> 
>  >Yes, I'm very familiar with PETE, however I'd vote against using
>  >expression templates (ET) for the boost multi-dim array library. 
The
>  >reason is that there are already good ET-based array libraries: 
Pooma and
>  >Blitz. However, there is a need out there for a more portable, 
simpler,
>  >and faster to compiler array library. I think the boost array 
library
>  >should fill this need.
> 
> Q. Why bother when Blitz and Pooma are already well established?
> A. What Jeremy said above.
> 
> This question, phrased a hundred differ ways, will come up over-and-
over 
> again.  You might as well save Jeremy's answer for the first 
paragraph of 
> the docs.  And then again for the last paragraph.
> 
> --Beman

I fully agree with Jeremy and Beman, although I am biased by having 
written the original class in question. I would add to the goal of  
simplicity, compilation speed and portability, the fact that the 
number of dimensions should remain unlimited (ok, POOMA is limited to 
7 dimensions and Blitz to 11, and that's a lot, but it's still 
limited), that it's generic (POOMA can hold only certain types, I'm 
not sure about Blitz), that it uses a series operator [] like built-
in arrays, instead of a single operator () for indexing, and the fact 
that it's small (in a way implied by its simplicity). Feature bloat 
is a danger, as Beman wrote, although there is space for improvement, 
some more core features, and perhaps further speed optimization. But 
this is not going to be THE multidimensional array for every use, 
especially for complex numerical applications. For these, POOMA and 
Blitz are more appropriate, and there's no need to redo them.

Giovanni Bavestrelli