What is the difference between dg.function.copy_u('h','m') and dg.function.copy_u('h','e')?

[TouranTouran]

I’m confused by the following two message functions:
message_func1 = dgl.function.copy_u(‘h’, ‘m’)
message_func2 = dgl.function.copy_u(‘h’, ‘e’)
I understand that, the first one will only generate mesages, while the second one will generate both messages AND edge features (called ‘e’). Am I right?

The document tells the message function will return messages OR edge features. So when return messages and when return edge feature? thanks.

def edge_udf(edges):
“”"
Parameters
----------
edges : EdgeBatch
A batch of edges.

Returns
-------
dict[str, tensor]
    The messages or edge features generated. It maps a message/feature name to the
    corresponding messages/features of all edges in the batch. The order of the
    messages/features is the same as the order of the edges in the input argument.
"""

[VoVAllen]

No, they are actually the same. Just in the reduce function, the message is stored in node.mailbox by node.mailbox['m'] or node.mailbox['e']

Basically the message is the same as the edge data. But in the update_all function, the intermediate message will be considered as cache, and will be removed after the full update_all process