chuck-dbos
commented
1 month ago Looks good! I wish there was a more general fix (this wouldn't catch the specific problem that was originally reported, where the bad input was actually a dict that contained functions) but it's a good start that we can refine.
Been thinking about that a bit. An object with some fields and helper functions is kinda fine. A dict with functions, that we could catch I think. We could iterate lists and dicts and make sure there was nothing weird in them.
For #111
Detect functions passed as arguments / return values to DBOS functions, and stop them.
Put a little more type info on WorkflowInput, and split the serialization cases a little.
Generally, we may not want "unpicklable" items in the serializer, but Kafka currently uses it.
Editorial comment, if this code is catching something, I wonder if the end developer is using mypy or similar. Hate to preach, but they're making a big mistake if they don't... are we adequately stressing this?