I had to change how Box2d is serialized/deserialized to make it work (otherwise the elements are written out of order)
This probably means that other shapes are like also wrong right now, but I wanted to solicit feedback before testing/changing those
Right now you still need to give it an explicit StructType schema so the resulting dataset is readable by spark and it knows when to convert the Rikai types into native types. It would be nice to add an option to just infer the types based on a sample of rows in the DataFrame.
Open issues: