Re: [boost] [GSoC] Request for Feedback on Boost.Bloom Filter Project

Subject: Re: [boost] [GSoC] Request for Feedback on Boost.Bloom Filter Project
From: Phil Endecott (spam_from_boost_dev_at_[hidden])
Date: 2011-06-27 19:39:01

Vicente Botet wrote:
> Phil Endecott-48 wrote:
>> Alejandro Cabrera write:
>>> Arash Partow wrote:
>>>> Looks ok, but one important question - Why is the BF typed? Its not
>>>> necessary and in fact there are many use-cases where one might want to
>>>> insert and/or test membership for a range of different types using the
>>>> same BF - all that those types require are that they be hashable.

> Could you give a concrete example of insertion of different types? any type?

You mean a practical application?  No.

>>> Though not necessary, I believe having a typed BF is safer. If there are
>>> users that want to eschew type safety to allow operations on multiple
>>> types,
>>> I believe this can be accomplished either by using a Bloom filter that
>>> accepts insertions of type void *.
>> 
>> Hmm, it's not clear to me how exactly you would do that with void*.  I 
>> think Arash's suggestion is better, and you can then build 
>> strongly-typed filters on top of the "any" version like this:
>> 
>> class bloom_filter_any {
>>    template <typename T>
>>    void insert(const T& val) { .... }
>>    ....
>> };
>> 
>
> could you elaborate how the hash functions will be applied to any T?, Should
> this imply dynamic polymorphism?

No, it doesn't require dynamic polymorphism, just compile-type 
polymorphism.  The filter's hash functions must accept the types that 
are actually used in calls to insert() etc.

Regards,  Phil.