Re: [boost] Re: MPL containers and algorithms

From: Matthew Hurd (matt_at_[hidden])
Date: 2002-08-11 18:24:26

----- Original Message -----
From: "David Abrahams" <dave_at_[hidden]>
>
> Even if you don't buy the need for different sequences, the plain fact is
> that different sequences exist. It can be useful to be able to
interoperate
> with them.

Following from afar, the interoperability argument makes the most sense to
me.  If MPL can make meta-template libraries and their corresponding
algorithms work together easily when sequence types are different then you
seem to have a win.

Maybe the killer example(s) would be say MPL algorithms working with Loki,
or some other typelist, and even vice versa if possible, that would be hard
without MPL's abstractions.  Perhaps mapping lists of type tuples to simple
typelist for the purpose of manipulation would also fall into this category.

Could a tree such as that used in an expression template evaluator be easily
expressed with MPL and be suitably manipulated?  This might help the
argument especially if it could interoperate with something like POOMA's
PETE lib.  I'd see being able to specify, manipulate and subset expression
trees with their proxies and typelists in a similar style as a good thing
(TM).

Some tests come to mind:
    1.  Using the simple loki typelist with MPL - already shown i think
    2.  Using the Generator stuff in loki with some non-simple typelist via
MPL
    3.  Some expression tree stuff perhaps using PETE as an external
reference...

This would demonstrate usefulness beyond doubt to both of my lonely neurons.

Just a little deposit...

matt.

PS:  I definitely fall into Andrei's category 3 (maybe a little bit of 2...
is that a 2.75?)... just trying to inject something that might be useful.