dolfin team mailing list archive
-
dolfin team
-
Mailing list archive
-
Message #09074
Re: GenericSparsityPattern
On Sat, Aug 16, 2008 at 07:16:31PM +0100, Garth N. Wells wrote:
>
> Anders Logg wrote:
> > On Fri, Aug 15, 2008 at 12:44:08PM +0100, Garth N. Wells wrote:
> >> I'm looking at some dof renumbering schemes which will operate on a
> >> graph of the matrix structure, and it would be convenient to rename
> >> SparsityPattern to something like 'AdjacencyGraph' (since it is just an
> >> adjacency graph) and move it to another directory (dolfin/graph).
> >> GenericSparsityPattern should probably also be renamed since it's member
> >> functions are for either inserting graph edges or returning information
> >> on the graph.
> >>
> >> Any naming suggestions/objections?
> >>
> >> Garth
> >
> > In what way(s) is the AdjacencyGraph different from the current Graph
> > class (which is essentially a CRS matrix).
> >
>
> The current Graph class requires a Mesh.
>
> > If it's just the weights, we could have a Graph base class and then
> > WeightedGraph as a subclass adding the weights.
> >
>
> We could have a (base) class Graph, a base class GenericGraphBuilder and
> specialised graph builder classes. I need to think though if the CSR
> format is appropriate for what I want.
>
> Some time ago, you tried using the Boost Graph Library before removing
> it. Was there a particular reason for not using it? I was planning to
> use it's renumbering features. It does support CSR graphs.
>
> Garth
I removed it because it wasn't used:
changeset: 3059:bad44cb9ad97
user: "Anders Logg <logg@xxxxxxxxx>"
date: Tue May 15 13:29:05 2007 +0200
summary: Remove AdjacencyGraph and dependency on boost graph lib
I'm not sure why we ended up not using it, but I vaguely remember I
wanted to use it to do Cuthill-Mckee reordering but found it very
difficult to decrypt the manual.
--
Anders
Attachment:
signature.asc
Description: Digital signature
References