V(g) not the same as 1:vcount

[maelle]

V(g) not the same as 1:vcount

[clpippel]

Instead of identical(index, V(graph)) use all(index == V(graph)).
Straightforward and easy to understand.

[clpippel]

@maelle,
I am afraid this is not going to work.
Because of the <weak reference> in V(g).
For example, consider:

g1 <- make_graph(c(1, 2, 3, 4))
index <- V(g1)
identical(index, V(g1))
# FALSE

# delete all vertices
g2 <- delete_vertices(g1, V(g1))
identical(V(g2), V(g2))
# FALSE

# This works for empty and non-empty sets V().
# For example.
all(V(g2) == seq_len(vcount(g2)))
# TRUE
all(index == seq_len(vcount(g1)))
# TRUE
all(index == V(g1))
#TRUE

Function identical() could be too strong.
Note thatidentical(1:4, c(1, 2, 3, 4)) returns FALSE.
Both parameters do not have the same class.

[maelle]

would equal work? feel free to suggest a change directly on the PR. I any case, thanks a ton!

[szhorvat]

My point about V(g) not being the same as 1:vcount was a general one. I did not have the time / chance to understand the issue you were working on in detail.

Maybe this perspective helps clarify what I meant:

distances(g, 1, V(g) will be faster than distances(g, 1, 1:vcount(g)), even though they are semantically equivalent. This is because V(g) indicates "all vertices" in a more efficient manner, and in a way that can be tested for in the C core, to help choose a better algorithm. Some algorithms are specialized for the "all vertices" case and the performance difference can be quite significant. From memory, some variants of transitivity() as well as igraph_ecc() should show a considerable difference.

[szhorvat]

Generally, I would recommend introducing a helper function to check if a vertex sequence vs includes all vertices in standard order.

I think all(vs == seq_len(vcount(g)) should work (as @clpippel says), but keep in mind that I don't know any R (!!). An optimization can/should be added to check attr(vs, 'is_all'). This is TRUE for V(g) specifically. All this assumes that vs is indeed of class igraph.vs.