Boost logo

Boost :

Subject: Re: [boost] [BGL] bundled properties and their restrictions (particularly default constructibility).
From: Geoff Hilton (geoff.hilton_at_[hidden])
Date: 2008-10-27 10:58:34


Geoff Hilton wrote:
> Hi, my graph uses bundled properties for both vertices and edges. Both
> bundled properties have containers that must be of a certain size when
> they're actually used. For testing purposes these containers were of
> fixed size (underlying container is currently a boost::array<..>.
>
> However the real world dictates that the size of these containers pretty
> much must be determined at run-time as opposed to compile-time because
> the graph is built using an external source such as a file or database
> for example.
>
> I know the documentation requires that bundled properties be all of
> default constructible, assignable and copy constructible[1][2] because
> of how the containers (as specified by selectors) store them. I'm unsure
> as to whether or how this may effect algorithms provided by the BGL, so
> far the only ones I've used are dijkstra_shortest_path and
> dag_shortest_path.
>
> Given that a custom container selector may be created and push/erase
> must be overloaded for the given custom container (for my purposes, this
> would be to specify a custom allocator as in the example)[3], how safe
> would it be to say that the container selector entirely dictates the
> restrictions in question?
>
> If the answer is "very", presuming we weren't to change these
> restrictions for existing provided selectors would it be possible for an
> explicit exception to be made for custom containers and their selectors
> (created using boost::container_gen<>)? This exception might say that
> the restrictions in question are then dictated by the custom container.
> I figure this shouldn't affect existing user code since the current
> restrictions are already at their most strict.
>
> Thanks very much,
> Geoff
>
> [1]
> http://www.boost.org/doc/libs/1_36_0/libs/graph/doc/using_adjacency_list.html#sec:adjacency-list-properties
>
> [2] Doug Gregor 23/03/2005, this list.
> [3]
> http://www.boost.org/doc/libs/1_36_0/libs/graph/doc/using_adjacency_list.html#sec:custom-storage

This is just a bump of sorts, I'm still very much in need of a reply.
Note, in [2], I meant the boost.user list, not boost.devel.

Thanks,
Geoff


Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk