Re: [Boost-users] [fusion] Why no multi{map,set} ?

From: Joel de Guzman (joel_at_[hidden])
Date: 2008-08-14 10:25:37

John C. Femiani wrote:
> Steven Watanabe wrote:
>> AMDG
>>
>> John C. Femiani wrote:
>>> I was just wondering why fusion (and mpl) don't include multi-maps or 
>>> multi-sets. Are they unnecessary at compile time? Are they too 
>>> trivial? Do they conceptually not work?
>>
>> The easiest way to implement a multimap is in terms of an ordinary map.
>> In a fusion/mpl map the key is a type
>> and lookup uses overload resolution.  This means that there
>> is necessarily a 1-1 correspondence between keys and values.
> I would expect a multimap to contain pairs, and to provide an iterator 
> range or sequence of like keys via an 'equal_range' function (like 
> unordered_multimap).
> 
> I  also just noticed there is no 'equal_range' query metafunction, 
> although there is a 'lower_bound' and 'upper_bound'.
>> The obvious way to implement a multimap is to make the
>> value a container.  Because of the function nature of
>> fusion/mpl  this would be like
>>
>> template<class M, class K, class V>
>> struct multimap_insert_existing {
>>    typedef typename insert<typename erase_key<M, K>::type, pair<K, 
>> typename push_back<typename at<M, K>::type, V>::type> > type;
>> };
>>
> <more code>
> 
> So it is feasible, but the authors just wanted to keep the scope limited?
> 
> That works for me; I've been toying with that idea anyhow but I wanted 
> to make sure I was not going waste my time.

As for Fusion, contribution is always welcome :-)

Regards,
-- 
Joel de Guzman
http://www.boostpro.com
http://spirit.sf.net