`largest_connected_component()` crashes with ZeroDivisionError on empty graphs

[thehalleyyoung]

largest_connected_component() crashes with ZeroDivisionError on empty graphs

Repository: microsoft/DeepGNN

Description

The largest_connected_component() function in src/python/deepgnn/graph_engine/snark/preprocess/sampler/metric.py does not guard against empty graphs (0 nodes), causing two crashes:

def largest_connected_component(g: nx.digraph) -> float:
    """Return the scaled size of the largest strongly connected component."""
    largest_strongly_connected_component = g.subgraph(
        max(nx.strongly_connected_components(g), key=len)
        #^^^ ValueError: max() arg is an empty sequence (when g has 0 nodes)
    )
    return nx.number_of_nodes(
        largest_strongly_connected_component
    ) / nx.number_of_nodes(g)
    # ^^^ ZeroDivisionError (when g has 0 nodes)

1. max() raises ValueError on the empty iterator from strongly_connected_components() of an empty graph
2. nx.number_of_nodes(g) returns 0 → ZeroDivisionError

Empty graphs are a realistic edge case in graph preprocessing pipelines — e.g., when a filtered subgraph has no matching nodes, or when processing an empty graph partition.

The sibling function densification() in the same file has the same issue with math.log(..., nx.number_of_nodes(g)).

Suggested fix

def largest_connected_component(g: nx.digraph) -> float:
    """Return the scaled size of the largest strongly connected component."""
    n = nx.number_of_nodes(g)
    if n == 0:
        return 0.0
    largest_strongly_connected_component = g.subgraph(
        max(nx.strongly_connected_components(g), key=len)
    )
    return nx.number_of_nodes(largest_strongly_connected_component) / n

The same guard should be added to densification(), diameter(), and max_adjacency() in this file.

How this was found

This was identified via static analysis using a3-python, which flagged the unguarded division as a DSE-confirmed reachable DIV_ZERO.