[boost] Re: Any, variant and more, and Sobol sequences

From: Geoff Leyland (geoff.leyland_at_[hidden])
Date: 2002-08-12 06:28:20

On Lundi, août 12, 2002, at 12:52 , "Dylan Cuthbert" <dylan_at_[hidden]> 
wrote:

> Sounds cool, just one question, why do you need .vector()[3], why not 
> return
> a template wrapper class for the variable that "operator TYPE"s down to
> whatever type it is, and at the same time has an operator[] that
> automatically casts the type and applies [] to it.

Good point!  I've got lots of ways of getting at the object.  There's a 
.be<something>() and a .ref<something>(), and a few names (.str(), 
.c_str(), .vector() and .map()) for common things.
I've also got a generic cast

template <class T> operator T(void);

which seemed a bit scary, 'cos I thought it would get in all over the 
place, but which actually works really well, and cuts down on a lot of 
.be<>()'s.

But the [] operators are an excellent idea.  Cuts down on more typing!

Incidentally, there's a big difference between the .be<> and the 
.ref<>.  I use a "storage_type" to store the thing:

eg: template <> class storage_type<const char *> { typedef string 
type; };

so that for example, all string types are strings, and all real types 
are doubles etc.  This means that when you do a be<>, you can cast back 
to the type you wanted (so a stored double can be cast to a float), but 
if you want a reference, you have to get exactly the type that's stored.

I'll have a bit of a tidy this afternoon or tonight, and put it 
somewhere people can have a look at it tomorrow.  Or in an "appropriate 
place" if someone can confirm where that might be...

cheers,
goof

--
Geoff Leyland
Laboratoire d'energetique industrielle
LENI-DGM-EPFL, CH-1015, Lausanne, Switzerland
Phone: [phone], Fax: [phone]