Setting <UnicodeSetAttribute>.set([]) (empty list) throws "Nonetype is not iterable" error.

[hoIIer]

During an update operation a UnicodeSet field may have it's value changed back to empty e.g. []

The following code fails:

(Pdb) actions = []
(Pdb) actions.append(attr.set([]))
(Pdb) record.update(actions)

*** TypeError: 'NoneType' object is not iterable

The following two examples work:

(Pdb) actions = []
(Pdb) actions.append(attr.set(None))
(Pdb) record.update(actions)

(Pdb) actions = []
(Pdb) actions.append(attr.set(['foo', 'bar']))
(Pdb) record.update(actions)

Is this a bug?

[hoIIer]

@ikonst reopening as I ran into this again, is it a bug?

Feels very annoying to not be able to reset a UnicodeSetAttribute to empty in an update operation.

instance.update(actions=[model.Thing.set_or_list.set({}|[])]) -> *** TypeError: 'NoneType' object is not iterable

[ikonst]

I wonder what causes this error.

For what it's worth, though, underlying dynamodb does not support empty sets:
https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/HowItWorks.NamingRulesDataTypes.html#HowItWorks.DataTypes
so it's unclear how it would behave, e.g.

• same as unsetting (assigning None and then you read None back)
• give an error, but a clearer one

[hoIIer]

I guess the annoying part is by removing it, you have to ensure any uses of the value explicitly test for None versus knowing it would always be an instance of list or set

[ikonst]

I agree. Since the underlying database cannot store an empty set for us, we have to pick an alternative representation. An attribute being unset in the DynamoDB item manifests as None in the Pythonic model, so we cannot use unset as the representation, since it'll read back as None.

We could decide on a rule e.g. an empty set serializes as {"__MAGIC_PYNAMODB_VALUE_UNSET__"}, but that would be awkward.

Another compromise would be:

• UnicodeSetAttribute(null=True) serializes set() as unset, and reads back as None
• UnicodeSetAttribute(null=False) serializes set() as unset, and reads back as set()

[hoIIer]

@ikonst thank you!

[erovira]

Ran into this. Thanks for the fix!