On 8/10/2011 2:51 PM, Eric Niebler wrote:
On 8/9/2011 11:17 PM, Joel de Guzman wrote:
On 8/10/2011 12:56 PM, Eric Niebler wrote:
A sequence created by push_back is a Forward Sequence, which means you can't call pop_back on it. I think that's a serious shortcoming. Fusion needs something like MPL's Back Extensible and Front Extensible concepts.
http://www.boost.org/doc/libs/1_47_0/libs/mpl/doc/refmanual/back-extensible-...
Unfortunately, this would likely require a redesign of the whole (push|pop)_(front|back) mechanism.
Man, this is a major bummer, alright. One plausible clean way I can think of is:
1) Make single_view random_access (if it's not already; IIRC No).
I implemented this not too long ago.
2) Make joint_view assume the least sequence category of its subsumed sequences. There's already some work done related to this where joint_view inherits the 'associative'-ness of its held sequences.
I don't think that's good enough. Take my example above. In, push_back(nil(), 1), the traversal type of nil is Forward, and so too would be the resulting view. It probably makes sense to see how MPL manages it.
Fusion diverges from MPL in its design in that sequence extension in Fusion always results to an 'extended' view (e.g. joint_view), instead of MPL's sequence-type-preserving operations. That is primarily for the sake of runtime efficiency (e.g. you do not want to have to build a new vector when you push_back an element). For that reason, I do not think it is a good idea to manage it like MPL. That example of yours [ push_back(nil(), 1) ] can also be made to work by making 'nil' random-access. In essence, we should give all sequences the highest potential traversal category that we can. Regards, -- Joel de Guzman http://www.boostpro.com http://boost-spirit.com