Re: [boost] Re: MPL containers and algorithms

From: Emily Winch (emily_at_[hidden])
Date: 2002-08-10 08:02:03

Douglas Gregor wrote:
> On Saturday 10 August 2002 01:45 am, Andrei Alexandrescu wrote:

[ snipped ]

> At one point you distilled your question down to "why do we need more than one 
> type of sequence?". I'm wondering if you considered Burton et al's use of 
> typevectors (heterogeneous value vectors, if you will) to be a useful 'other' 
> sequence type? (I described their use of the typevectors in a different 
> thread).

That sounds like a killer argument (and I'd love to know why it didn't 
occur to me before in this context). Treating a value sequence as a type 
sequence is definitely definitely a Useful Thing To Do. For printing out 
the types to check that they're the ones you expected: for asserting 
that all the types conform to the expected interface: for calculating 
the length of the sequence.

The alternative would be a value sequence that exposes its type sequence 
for people to mess with: but it just seems so _neat_ that a value 
sequence should act like a type sequence if you apply type-sequence 
algorithms to it.

I suppose one could argue for just having one sequence, a value 
sequence, and using it as a type sequence by not instantiating it: but 
that sounds like a nasty idea. Not least because it is simply not 
intuitive that instantiating a type_sequence<big-type, other-big-type> 
should drag two big objects into existence.

Emily