Thanks!

Can you change this to use the G(n,m) model and add a comment that spells out that the method samples DAGs with the same number of vertices and edges uniformly?

I'll push this change soon.

But a question: Does Sugiyama not do proper edge routing in Python, like it does in R? I see results like this:

I am expecting something like this instead:

Does Sugiyama not do proper edge routing in Python, like it does in R?

No it doesn't. More precisely: it can be called as layout_sugiyama(), in which case it returns the layout matrix but does not return the extended graph, for sake of compatibility with other layout_...() methods that also return the layout only, or it can be called as layout_sugiyama(return_extended_graph=True), in which case it returns the extended graph with the dummy vertices and the layout, but then you need to take care of setting up the plotting so that the additional vertices are hidden and only the edges are routed between them (and even then you won't have nice curved edges).

Not sure what the best approach would be when dealing with this in the tutorial. Something like this should do the trick:

layout, extd_graph = g.layout("sugiyama", return_extended_graph=True)
ig.plot(
    extd_graph,
    vertex_size = [20] * g.vcount() + [0] * (extd_graph.vcount() - g.vcount())
    # the rest of arguments should be as they are in the current example
)

but then we'd still need to hide the arrowheads on all the edges that point to a dummy vertex, otherwise we end up with extra arrowheads. So it's a bit too much code for my taste for a simple tutorial example.
It shifts the focus from generating the DAG to plotting.

In C/igraph 0.11 we should develop conventions for returning edge routing information. The extended graph approach might not be the best fit for other algorithms. Once we have this in place, there can be a convention layout_ functions in Python as well. It would be nice to implement this for the new Python interface.

Indeed when we say "graph layout" casually we kind of mean "vertex layout", didn't think about that before.

I agree with @szhorvat that some form of support for routing would be nice to have after the refactor. That would also need a special infrastructure on the plotting side, to automate the issues highlighted above