induced.subgraph and subgraph.edges don't adjust layout if layout is matrix

[ahmontgo]

I've been plotting multiple subgraphs while attempting to retain layout values, and came across an interesting bug. If g$layout is a matrix, vertex deletion doesn't modify it, which produces odd results. For example:

g <- make_ring(10)
g$layout <- layout_in_circle(g)
plot(induced.subgraph(g,c(1:9)))

The "ghost" node 10 is still plotted, picking up the label of the first vertex to boot!

Workaround is to manually remove the node from layout:
g2 <- induced.subgraph(g,1:9)
g2$layout <- g2$layout[1:9,]
plot(g2)

Similar but slightly more complex manual removal needs to occur when subgraph.edges is used:
g3 <- subgraph.edges(g,1:8)
g3$layout <- g3$layout[V(g)[inc(1:8)],]
plot(g3)

Expected behavior would be for plot() to ignore any "extra" vertices in layout. But depending on internal representation, there may be no way of doing so. In that case, subgraph.edges and induced.subgraph should probably modify any g$layout if g$layout is a matrix.

I'd offer a patch, but my C is very shaky.

Thanks!

[ntamas]

Not planned for the time being for sake of backwards compatibility. If you want to adjust the layout when vertices are deleted, you can use vertex attributes named x and y instead of a graph-level layout attribute and this will treat vertex deletion correctly. These attributes are also recognized by layout_nicely().

[szhorvat]

I understand that there is no capacity for dealing with this now, but it seems to me that the behaviour is bad enough to be called a "bug", which should ideally be tracked.

What the correct behaviour would be is debatable. I'd say that the entire design of dealing with layouts is problematic. This issue would provide a space to discuss that.

[ntamas]

I'd probably just remove the special handling of the layout graph attribute from layout_nicely() (which is the only place where this attribute is treated as a layout matrix). Having a vertex-specific quantity stored as a graph attribute is asking for trouble.

[szhorvat]

Yes, I agree, and that is of course breaking, so it's probably for 2.0 only. My point is that either there should be a new issue for that or this one should be kept open so we do not forget about this problem that occurs with some usage patterns.

[schochastics]

fixed in #1880