CODEWITHSUNDEEP
c#.netgenericstype-inference
user
user

Reputation: 7333

Allow templates to be inferred

Say I'm using an external package for storing graphs. A BidirectionalGraph takes two templates: a vertex and an edge type:

var graph = new BidirectionalGraph<Vertex, Edge<Vertex>>();

Unfortunately, this graph package doesn't allow you to get the edges radiating into a vertex in a single line. Instead, you have to provide an IEnumerable, which it will populate with the results. This can disrupt a good coding rhythm by making tasks like "loop through all vertices that are successors of vertex x" take far too much code.

I wanted to use .NET's extensions to add a one-line solution to the graph class:

public static class GraphExtensions
{
    public static IEnumerable<TEdge> IncomingEdges<TGraphSubtype, TVertex, TEdge>(this TGraphSubtype graph, TVertex n)
        where TGraphSubtype : BidirectionalGraph<TVertex, TEdge>
        where TEdge : IEdge<TVertex>
    {
        IEnumerable<TEdge> inputEdgesForVertex;
        graph.TryGetInEdges(n, out inputEdgesForVertex);
        return inputEdgesForVertex;
    }
}

But when I call graph.IncomingEdges(vertex), for some reason C# (.NET version 4.5) can't infer the template arguments, so I have to say:

graph.IncomingEdges<GraphThatInheritsFromBidirectionalGraph<VertexType,EdgeType>,VertexType,EdgeType>(vertex). Not really a great improvement.

First, why can't the template types be estimated? I have a feeling it has to do with inheritance, but don't understand. I'm used to using C++, and for some reason feel that gcc could infer the template types.

Second, if this can't be prevented, is the correct design choice to make a graph class for actual use, which inherits from BidirectionalGraph? It seems a waste to have to rewrite the constructors, but I'm sure you'd agree that calling the method with explicit template types is inelegant.

EDIT:

Strangely, the equivalent specification (below) does allow automatic inference of template types. So, even though it solves my initial problem (adding this functionality to the graph), I'd still really like to understand.

public static class GraphExtensions
{
        public static IEnumerable<TEdge> IncomingEdges<TVertex, TEdge>(this BidirectionalGraph<TVertex,TEdge> graph, TVertex n)
            where TEdge : IEdge<TVertex>
        {
            IEnumerable<TEdge> inputEdgesForVertex;
            graph.TryGetInEdges(n, out inputEdgesForVertex);
            return inputEdgesForVertex;
        }
}

Upvotes: 7

Views: 95

Answers (1)

Eren Ers&#246;nmez
Eren Ers&#246;nmez

Reputation: 39085

The first version of your extension method is able to infer TGraphType and TVertex but not TEgde, as it would require inferring the TEdge from the type constraint:

where TGraphSubtype : BidirectionalGraph<TVertex, TEdge>

which C# compiler does not do (it does not infer generic type parameters from type constraints). I honestly don't know if there is a technical reason behind this or it just wasn't implemented.

Your updated version, on the other hand, includes BidirectionalGraph<TVertex, TEdge> as a parameter, so for example when you call the extension method on a class like:

class AGraph: BidirectionalGraph<AVertex, AnEdge> { ... }
...
var aGraph = new AGraph();
aGraph.IncomingEdges(vertex);

the compiler is able to examine the type AGraph and see that there is a unique type BidirectionalGraph<AVertex, AnEdge> in its inheritance hierarchy, so it is able to infer TVertex and TEdge.

Note that if the parameter type were IGraph<TVertex, TEdge> (instead of BidirectionalGraph<TVertex, TEdge>) and AGraph implemented multiple constructed types of that generic interface, e.g.:

class AGraph: IGraph<AVertex, AnEdge>, 
              IGraph<AnotherVertex, AnotherEdge> { ... }

then type inference would fail once again because it can't tell if, for example, TVertex is AVertex or AnotherVertex.

Upvotes: 1

Related Questions