node_link_data JSON serialization omits set-type node attributes

The node_link_data function in NetworkX does not inherently filter out set or frozenset attributes; rather, it returns a standard Python dictionary containing those objects. The "silent loss" typically occurs during the subsequent json.dumps() call if a custom encoder is used that skips non-serializable types, or if the data is passed through a pipeline that strips non-primitive types to avoid TypeError.

Analysis of Serialization Behavior

Because the JSON standard lacks a native set type, the standard Python json library cannot serialize sets. Under default conditions, json.dumps(node_link_data(G)) will raise a TypeError: Object of type set is not JSON serializable. If you are experiencing silent omission, it is likely that your environment uses a wrapper or a custom JSON encoder configured to ignore unknown types rather than raising an exception.

Version Consistency (NetworkX 2.x vs 3.x)

This behavior is consistent across NetworkX 2.x and 3.x. In both versions, node_link_data is a data transformation utility that converts a graph object into a dictionary format. It does not perform the actual JSON string conversion, meaning it does not enforce JSON type constraints during its execution.

Recommended Solutions

To preserve set-valued attributes for JSON exchange, you must convert them to a serializable format (like a list) before or during serialization.

Option 1: Pre-serialization Conversion

The most reliable method is to cast set attributes to lists before calling the serialization function. This ensures the data is compatible with all JSON encoders.

import networkx as nx
import json

# Convert sets to lists in node attributes
for n, attr in G.nodes(data=True):
    for key, value in attr.items():
        if isinstance(value, (set, frozenset)):
            G.nodes[n][key] = list(value) 

data = nx.node_link_data(G)
json_string = json.dumps(data)

Option 2: Custom JSON Encoder

If you cannot modify the graph object, use a custom encoder to handle sets on the fly. Note that this only solves serialization; upon json.loads(), these will return as lists, and you must manually cast them back to sets.

class SetEncoder(json.JSONEncoder):
    def default(self, obj):
        if isinstance(obj, (set, frozenset)):
            return list(obj)
        return super().default(obj)

json_string = json.dumps(nx.node_link_data(G), cls=SetEncoder)

Verification Step

To diagnose if the omission is happening inside NetworkX or during JSON encoding, run the following check:

data = nx.node_link_data(G)
# Check if the set still exists in the dictionary before json.dumps()
print(type(data['nodes'][0]['your_attribute_name']))

If the output is <class 'set'>, the issue resides in your JSON encoding layer, not in NetworkX.

Diagnostic Request: Are you using a specific JSON library (e.g., ujson, orjson) or a framework-specific serializer (e.g., Django, Flask) that might be suppressing TypeError exceptions?