distances() ignores 'mode' when algorithm='johnson'

[szhorvat]

Describe the bug

The distances() function ignores the mode parameter when algorithm='johnson' because igraph_distances_johnson() does not support setting the mode: igraph/igraph#2193

This can lead to confusion and invalid results.

I suggest that when algorithm='johnson' and mode != 'all' for undirected or mode != 'out' for directed an error should be thrown.

To reproduce

> g<-make_graph(c(1,2,2,1,2,3))
> distances(g)
     [,1] [,2] [,3]
[1,]    0    1    2
[2,]    1    0    1
[3,]    2    1    0
> distances(g, algorithm='dijkstra')
     [,1] [,2] [,3]
[1,]    0    1    2
[2,]    1    0    1
[3,]    2    1    0
> distances(g, algorithm='johnson')
     [,1] [,2] [,3]
[1,]    0    1    2
[2,]    1    0    1
[3,]  Inf  Inf    0

Notice that since the default mode is 'all', the graph is treated as undirected by both the default algorithm and Dijkstra. But with Johnson it is unexpectedly treated as directed.

Version information

• R/igraph version: 1.3.4

[szhorvat]

BTW I find it very strange that the default mode is all and not out. I would expect the function to respect the directedness of the graph by default. But this is not something we can change unfortunately.

[ntamas]

7a34b43 now throws an error when Johnson's algorithm would be invoked with mode="in" or mode="all" (and forces the mode to be "out" for the undirected case).