Re: [boost] Re: MPL containers and algorithms

From: Paul Mensonides (pmenso57_at_[hidden])
Date: 2002-08-01 18:43:45

----- Original Message -----
From: "David Abrahams" <dave_at_[hidden]>

> I think this is a slightly hasty analysis. In general we need to know a bit
> about the internals of the compiler to say what goes on. For example,
> insertion in a dot-style list may only require copying the head part of the
> list:

Okay, maybe a little. :)

> > Given that this is the case, the most efficient solution would be to
> always use
> > a tuple-like (i.e. vector-like) container that can hold a huge number of
> > elements.
>
> I'm not sure. When things get really large, push_front on a cons-style list
> could easily be faster than copying all those pointers.

When things get really large, you are entering the realm of impracticality.  Not
everything should be done within the language itself.

> > On the other hand, the most extensible solution would be to use a
> > cons-style list.  One or the other is fine, but both is unnecessary.
>
> I agree with you and Andrei that some numbers would really help with the
> analysis. However, I think even with the numbers I'd be reluctant to give
> up the flexibility at this stage, especially if it's not costing us much.

Yeah, I think only time will tell.

> There are other reasons to be able to work with different sequences. For
> example, if someone already has a program full of Loki typelists and wants
> to use an MPL algorithm, they can do it. A program that manipulates
> boost::tuple can treat the tuple's types as an MPL sequence.

This is a good point, IMO, but I also think it only goes so far.  I think it is
fair to say that rolling-your-own is typically frowned on when something already
exists.

> I realize that
> these all have cons-style structures, but their types /are/ expressed
> differently from mpl type_list, and from one another. Not being hardcoded
> to use a single type list structure still has great advantages, and as
> Emily said I appreciate the flexibility to experiment with other
> structures.

I'm just not sure about the real benefit, in the fully generic sense, of
compressing a more complicated non-sequencial structure into a sequence.  As I
said, only time will tell, and I agree, the MPL does provide the foundation to
experiment with this sort of thing.

Paul Mensonides