Empty-set safety

[ojii]

aiodynamo used to be empty-string-safe when DynamoDB didn't support empty string values. This mercifully got fixed in DynamoDB, but DynamoDB still doesn't accept empty sets (NS, SS, BS). There's an argument to be made that aiodynamo should handle this automatically, converting F("some_set").set(set()) to F("some_set").remove(), possibly with a warning.

[dimaqq]

Is that what amazing dunamo db does? Let's say if there's an item with a set with 1 element and later that element is removed.

[ojii]

Is that what amazing dunamo db does? Let's say if there's an item with a set with 1 element and later that element is removed.

this test fails for the .set(...) case, but succeeds for the .delete(...) case. Meaning, DynamoDB is seemingly okay with empty sets so long as they have previously had values in them, just not creating new ones. Absolutely baffling.

@pytest.mark.parametrize("op", [F("set").set(set()), F("set").delete({"a"})])
async def test_empty_set(client: Client, table: TableName, op):
    await client.put_item(table, {"h": "h", "r": "r", "set": {"a"}})
    await client.update_item(table, {"h": "h", "r": "r"}, op)

EDIT: Actually, F("set").delete({"a"}) causes the key to be removed from the item 😱

[ojii]

🤔 this gets even more complicated because of F.set_if_not_exists which doesn't have an equivalent for empty sets from what I can see (eg "remove_if_not_exists")