Subject: Re: [boost] [any] Multimethod for boost::any and modified boost::any to fit multi method
From: Andrey Semashev (andrey.semashev_at_[hidden])
Date: 2009-02-13 10:54:08
Mathias Gaunard wrote:
> Andrey Semashev wrote:
>> Mathias Gaunard wrote:
>>
>>> A while ago, I offered something like this
>>>
>>> any m = 0;
>>> apply_visitor<int, double, A>(f, m);
>>>
>>> or alternatively,
>>> apply_visitor(f, m); with f specifying all overloads in some
>>> compile-time reflection structure.
>>>
>>> unfortunately, it cannot be made constant-time because std::type_info
>>> sucks.
>>
>> FWIW, I would be really interested in this feature even if it's not
>> constant time. May I hope it will appear in trunk one day?
>
> It can actually be made constant-time using the result of
> std::type_info::name for dispatching.
> That means, however, that you need to know as a compile-time string the
> names of the types to check for, and have implemented a compile-time
> name mangling algorithm (which is of course platform-specific).
That looks like too much of a burden for users, let alone the
portability issues. I think the implementation that provides the
interface you mentioned above would be more suitable, even if slower. I
don't think that people will often try to dispatch between hundreds or
even tens of types, where the difference would matter.