Currently, if you try to use BQ and materialize a feature that is a list (of numbers, strings, etc), Feast will crash because in BQ, the value type of the feature is a dictionary, such as

{'list': [{'item': 3}, {'item': 3}]}
In materialize, we convert the latest values retrieval job to a pyarrow table and then converts to ValueProtos to write. This calls

python_type_to_feast_value_type, which doesn't support dict types.