PartitionedGraphs extension for DataGraphs and interface overhaul.

[jack-dunham]

This PR introduces a change to the DataGraphs interface. Data on an AbstractDataGraph can be queried with the functions getindex, setindex!, isassigned and unsetindex!. To customize the behavior of these function for a custom subtype of AbstractGraph, one should overload the following:

underlying_graph(::AbstractDataGraph) = not_implemented()

is_vertex_assigned(::AbstractDataGraph, vertex) = not_implemented()
is_edge_assigned(::AbstractDataGraph, edge) = not_implemented()

get_vertex_data(::AbstractDataGraph, vertex) = not_implemented()
get_edge_data(::AbstractDataGraph, edge) = not_implemented()

set_vertex_data!(::AbstractDataGraph, data, vertex) = not_implemented()
set_edge_data!(::AbstractDataGraph, data, edge) = not_implemented()

This a bit more verbose than just having a function vertex_data etc that returns a mutable dictionary, but I think it is ultimately more generic. The return value of vertex_data etc is now constructed on the fly, so doing setindex!(vertex_data(g), val, vertex) will not modify the underlying graph.

The benefit of this change is that one no longer require the data on the vertices and edges to be stored as a mutable field; it can be generated on the fly. This then results in constant behavior when defining views of vertex/edge data. One can still use a mutable dictionary to store the data for a given concrete AbstractDataGraph subtype.

DataGraphsPartitionedGraphsExt

This package extension (currently existing as a library in DataGraphs while PartitionedGraphs is a library of NamedGraphs) defines notions of getting data of partitioned and quotient graphs. By default, all index types that can be converted to collections of indices (e.g. QuotientVertex, QuotientVertices, QuotientEdgeEdges) returns subgraphs when using the getindex syntax, inline with the behavior of NamedGraphs. If one wishes to define separate data on the quotient graph on their custom partitioned data graph (this could be separate memory entirely, or data derived from the underlying graph), then one must overload the following:

# Turn off subgraphing for quotient vertices by defining:
NamedGraphs.to_graph_index(::MyGraph, qv::QuotientVertex) = qv

# Customize the behavior of this method:
DataGraphs.get_index_data(graph::MyGraph, qv::QuotientVertex) = ...

# And this method
DataGraphs.is_graph_index_assigned(graph::MyGraph, qv::QuotientVertex) = ...

and similarly for the QuotientEdge methods (if desired). One can also overload:

DataGraphs.set_index_data!(graph::MyGraph, val, qv::QuotientVertex) = ...

to define notions of setting/unsetting data on the quotient vertices. Again, similar methods can be defined for QuotientEdge.

Other Changes

• AbstractDataGraph now subtypes AbstractNamedGraph. This greatly reduces required method overloads.
• Added VertexDataView and EdgeDataView objects that behave like dictionaries with respect to the graph data.
• Added a library extension for the PartitionedGraphs package.

Closes #57, closes #58, closes #59, closes #60, closes #61, closes #62.

[mtfishman]

(I closed and reopened the PR to trigger the tests since NamedGraphs.jl v0.9 is now registered.)

[jack-dunham]

Ive accidentally committed a notebook. Let me fix the history.

[mtfishman]

Ive accidentally committed a notebook. Let me fix the history.

I think the history of the main branch should be ok as long as it is deleted by the time the PR is merged since we always squash and merge so the main branch should only see the final PR state as a commit in the history.

[jack-dunham]

Doing the rebase is good git-fu practice for me...

[mtfishman]

Thanks again! Having a carefully thought-out interface here will be very helpful as a basis for a lot of future work.