Re: [boost] [fusion] mpl::clear doesn't work on fusion views

Subject: Re: [boost] [fusion] mpl::clear doesn't work on fusion views
From: Joel de Guzman (joel_at_[hidden])
Date: 2011-07-30 06:24:20

On 7/30/2011 5:45 AM, Eric Niebler wrote:
> On 7/29/2011 1:36 PM, Jeffrey Lee Hellrung, Jr. wrote:
>> On Fri, Jul 29, 2011 at 12:52 PM, Eric Niebler <eric_at_[hidden]> wrote:
>>
>>> Today I ran into a limitation in Fusion/MPL interoperability. I couldn't
>>> use an mpl algorithm on a Fusion sequence because mpl::clear wasn't
>>> defined on it. The sequence was a joint_view. I see that
>>> fusion::detail::clear is only defined for Fusion cons, map, set, vector
>>> and deque. Can we get it defined for all Fusion sequences? I think for
>>> everything else, an empty Fusion vector would do. That is all
>>> fusion::clear does, btw.
>>>
>>> (Aside: it seems strange to me that mpl::clear of a fusion::set gives an
>>> empty set, but that fusion::clear of an empty set gives an empty vector.
>>> This should probably be made consistent.)
>>
>> I'm confused.  Why would fusion::detail::clear (which I presume is
>> associated with fusion::clear) have anything to do with mpl::clear?
> 
> Because all Fusion sequences are also MPL sequences.
> 
>> Also, mpl::clear requires an MPL Extensible Sequence, which...I'm guessing
>> fusion::joint_view doesn't satisfy...?  So either make fusion::joint_view an
>> MPL Extensible Sequence, or relax the requirements on mpl::clear...?  I'm
>> not sure either make sense, really.
> 
> Ah. Certainly a fusion::joint_view is not extensible. But that's a weak
> rationale because it's not how Fusion operates. A push_back on a Fusion
> vector yields a joint_view, not a Fusion vector. Clearing any Fusion
> sequence could just yield an empty vector, like fusion::clear does (with
> the caveat stated above).
> 
>> I guess there's some reason you don't want to or can't copy the types in the
>> joint_view into an mpl::vector and operate on that instead?
> 
> Because I'm lazy. And because Fusion can do this. And it's a reasonable
> thing to expect. And the patch would be about 3 lines, which I'm happy
> to provide.

The issue is that fusion has 2 levels, one is the MPL level, which involve
mpl interoperability and is only about types. The other is the type/data
level. fusion::clear is not the same as mpl::clear. The former is non
sequence-type preserving (by design) while the latter is (always, also by
design). In the MPL level, we have to do as MPL sequences do. E.g. you cannot
clear an mpl::joint_view. Surely, MPL can do that as well (return an
mpl::vector<>, but it doesn't, by design.

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