If two datasets are locally joined by integer fields and the fields are different widths in the left and right sides, the code generator will complain that the input distributions do not appear to be matched - even though the integers will actually be hashed to the same value.