Github user srowen commented on the issue: https://github.com/apache/spark/pull/22063 For those following along, the problem right now is this: functions like `udf()` have always declared that their type params have `TypeTag`s, meaning the caller has to have runtime info about types. This supports schema inference. Although the internals of the `udf()` methods changed in this PR, the signature did not. `SchemaReflection.schemaFor` was called before. However, with this change, suddenly it seems like the `TypeTag`s are needed by the compiler, and many instances of `udf()` fail now because the call site does not have one. I worked around a few, but am having more trouble with others. I am still not clear why it is that only with this change do the `TypeTag`s seem to matter; they were always required?
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org