Re: [boost] [hash] Extract module from functional + std::hash_combine

Subject: Re: [boost] [hash] Extract module from functional + std::hash_combine
From: Peter Dimov (lists_at_[hidden])
Date: 2017-12-20 15:08:17

Daniel James wrote:
> On 19 December 2017 at 15:48, Peter Dimov via Boost
> <boost_at_[hidden]> wrote:
> >
> > If we name the module 'hash' it's not hard to see that in the future it 
> > will be seen as the obvious home for implementations of MD5, SHA-*, and 
> > other hash algorithms (FNV-1a, cityhash...) which we do miss in Boost. 
> > So there will be pressure to add them, pull requests implementing them, 
> > and so on.
> >
> > The question is do we see this as a bad thing?
>
> Well I don't want this to create extra pressure on me, so I do. I'd rather 
> they were separate, unless we were expanding the boost::hash 
> implementation to support multiple hash algorithms.

After looking into this a bit, I think that it would be hard to expand our 
current boost::hash in a compatible manner, due to the way it's specified - 
boost::hash<X>()(x) is required to call hash_value(x), which is probably 
implemented by the user in terms of hash_combine, which in turn is required 
to call back to boost::hash<T>, and boost::hash can be specialized, so 
everything is detectable and has to work in precisely this manner.

It would probably be best to proceed as planned with the extraction of 
functional/hash as-is into its own 'hash' module and leave the future 
extensible framework separate (we could imaginatively name its module 
'hash2'.)