Consistency of `show_dot_graph` and `get_num_nodes`

[majosm]

A question came up while revisiting #489 with @kajalpatelinfo: should show_dot_graph be consistent with get_num_nodes, in the sense that the number of nodes displayed in the graph matches what get_num_nodes computes? It looks like show_dot_graph's code is based on CopyMapper, which ignores duplicates, but currently get_num_nodes uses id(expr) for a hash key, meaning it counts duplicates separately.

I'm tempted to think they should behave the same, at least by default. It might actually be useful to give them both an option to include/exclude duplicates? Thoughts @inducer?

[inducer]

I think get_num_nodes should be adapted to only count non-duplicates. I'm skeptical that including duplicates (at least the naive way) is useful functionality, as it could very easily have exponential complexity. As for non-naive, maybe that's useful? But it should probably not be the default.

[majosm]

I think get_num_nodes should be adapted to only count non-duplicates. I'm skeptical that including duplicates (at least the naive way) is useful functionality, as it could very easily have exponential complexity. As for non-naive, maybe that's useful? But it should probably not be the default.

What's naive/non-naive here? No cache vs. cache-by-id? Or something else?

[inducer]

Naive = no cache.

[kajalpatelinfo]

Okay, thank you both!