Is it legal to use primitive values as unknown?

[yamarten]

The lexicon validator forces the value of unknown type to be object, but there is no such description in the lexicon specification. From the specification, you can also refer to definitions with primitive types such as string, just like ref.

Is this a unknown specification or a validator-specific constraint?

[bnewbold]

This is a great question! I think this is a bit under-specified and we aren't sure about the details.

For example:

• can you distinguish between a null value and nullable value?
• are there limits on the data encoded in an unknown? eg, if it is an array, can it have a billion elements? I think this is mostly a generic data question, but it becomes more obvious with unknown fields
• can you have an array of unknown, with different types in each place?
• if an object, does it need to have $type included? I think not. can it have a $type though? do implementations need to distinguish between a generic object in an unknown and something like a blob ($type: blob)?
• what are the schema evolution rules? I think once something is unknown that probably can't change; but can something change from an object (or other type) to unknown? (I also think not)

Sorry to not have a more helpful answer. Pragmatically, right now I would try to stick to only using object in unknown fields, and we'll try to clarify this soon.

UPDATE: here is a tracking issue in the specs repo: bluesky-social/atproto-website#319