Github user HyukjinKwon commented on the issue: https://github.com/apache/spark/pull/20295 For https://github.com/apache/spark/pull/20295#issuecomment-360297123, I am fine without new serialization protocol actually. I didn't have a strong preference there because I wasn't sure if it's worth - complexity vs actual gain vaguely and seems that's now clarified there. I am okay with the current approach. @BryanCutler, I think the intention here is to follow other few APIs and `gapply` in R. I guess you meant the length and metadata stuff by "an optional kwargs to each pandas_udf to deal with 0-param udfs" if I remember correctly. I think that's slightly different because here the motivation is to provide consistent support similarly with other APIs vs the `kwargs` thing sounds pretty few concept to me.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org