Fix edge attributes of bipartite edges

The breaking changes of igraph version 1.4.0 regarding the classes of edge
attributes revealed an actual bug in the edge attributes of bipartite edges in
coronet:

We need to make sure that we only add the edge attributes for the edges that are
really added to the network. As, for example, edges to the untracked artifact
are not added to our bipartite network, we also need to make sure to remove the
attributes of the corresponding edges (otherwise this could lead to wrong
attributes). Therefore, we add a (rather complex) check for which edges (i.e.,
vertex sequences) are present in the final bipartite network and extract the
corresponding information for the edge attributes accordingly.

This addresses se-sic#236.

Signed-off-by: Thomas Bock <bockthom@cs.uni-saarland.de>

util-networks.R

@@ -1298,7 +1298,7 @@ add.edges.for.bipartite.relation = function(net, bipartite.relations, network.co
                 igraph::V(net)[d, vert] # get two vertices from source network:  c(author, artifact)
             })
             return(new.edges)
-        }, names(net1.to.net2), net1.to.net2)
+        }, names(net1.to.net2), net1.to.net2, SIMPLIFY = FALSE)
 
         ## initialize edge attributes
         allowed.edge.attributes = network.conf$get.value("edge.attributes")
@@ -1309,6 +1309,31 @@ add.edges.for.bipartite.relation = function(net, bipartite.relations, network.co
         extra.edge.attributes.df = parallel::mclapply(net1.to.net2, function(a.df) {
             cols.which = allowed.edge.attributes %in% colnames(a.df)
             return(a.df[, allowed.edge.attributes[cols.which], drop = FALSE])
+        extra.edge.attributes.df = parallel::mcmapply(vertex.sequence = vertex.sequence.for.edges, a.df = net1.to.net2,
+                                                      SIMPLIFY = FALSE, function(vertex.sequence, a.df) {
+
+            ## return empty data.frame if vertex sequence is empty
+            if (length(unlist(vertex.sequence)) == 0){
+                return(data.frame())
+            }
+
+            ## get the artifacts from the vertex sequence (which are the even elements of the sequence vector)
+            vertex.names.in.sequence = names(unlist(vertex.sequence))
+            artifacts.in.sequence = vertex.names.in.sequence[seq(2, length(vertex.names.in.sequence), 2)]
+
+            ## get the edges that will be constructed from the artifacts,
+            ## to get only the edge attributes for edges that will be present in the final network
+            ## (i.e., ignore edges to removed artifacts, such as the empty artifact that has been removed above)
+            constructed.edges = a.df[a.df[["data.vertices"]] %in% artifacts.in.sequence, , drop = FALSE]
+
+            ## return empty data.frame if there will be no edges in the end
+            if (nrow(constructed.edges) < 1) {
+                return(data.frame())
+            }
+
+            ## select the allowed attributes from the edge data.frame's columns
+            cols.which = allowed.edge.attributes %in% colnames(constructed.edges)
+            return(constructed.edges[ , allowed.edge.attributes[cols.which], drop = FALSE])
         })
         extra.edge.attributes.df = plyr::rbind.fill(extra.edge.attributes.df)
         extra.edge.attributes = as.list(extra.edge.attributes.df)