Re: [boost] Re: Re: [graph] bellman_ford_shortest_paths interface

From: Doug Gregor (dgregor_at_[hidden])
Date: 2004-11-15 11:47:09

On Nov 15, 2004, at 11:02 AM, Vladimir Prus wrote:
>> You can store information within the vertices and have only the names
>> of a few of the vertices to start from.
>
> You'd need to make property map return +inf when property on any other
> vertex is accessed, then.

Correct.

>> For the properties, you keep a
>> global counter: each time you run the algorithm, you increment the
>> counter, thereby telling all vertices that their values are out of
>> date.
>
> Is this for implementation of the above? Else I don't understand what 
> you
> mean.

Sorry; yes, this is an efficient implementation of the above.

> May I again ask you if you have a real example? It still feels bad to
> optimize for generic case which might be never needed. It might be 
> possible
> to provide fully generic version with a slightly different name or use 
> a
> special parameter, or something, without complicating interface for the
> common case.

I don't know of a real example.

If we can dispatch cleanly to provide both functionality types, I don't 
see a problem with having both versions. Unfortunately, the common case 
(VertexAndEdgeListGraph) has the stranger syntax, and we're stuck with 
that until we move everything into namespace boost::graph.

        Doug