Hi everybody, while reviewing PR #2094 [1] I noticed that the field reference syntax for FieldAccessors is not compatible with the syntax supported for key definitions (ExpressionKeys) used in groupBy(), keyBy(), join().where().equalTo(), etc.
FieldAccessors are only used for build-in aggregations in the DataStream API (sum(), min(), max(), ...). In particular I identified the following inconsistencies: - FieldAccessors allow to address array cells. ExpressionKeys treat arrays as AtomicTypes (Array TypeInfos do not extend CompositeType). Hence, it is not possible to address array cells. - ExpressionKeys do only support Integer keys for tuples. An atomic type can only be addressed with "*". FieldAccessors allow to address AtomicTypes with 0 in addtion to "*". - ExpressionKeys support to address fields of Java tuples with "f2" and Scala tuple fields with "_3". FieldAccessors do not support the "f" or "_" prefix. I would like to propose to adapt the syntax of both mechanisms (ideally, both should use the same code for validation). IMO, the ExpressionKey syntax much more widely used and is well designed. Therefore, I would adopt it for FieldAccessors as well. However, that would mean to restrict the syntax of the FieldAccessors and might break existing code. What do others think? [1] https://github.com/apache/flink/pull/2094
