Hi Owen, Just to clarify. So in my parser as I am mapping fields from Avro to ORC equivalent, any time I encounter a union type on the Avro side I map to struct type on the ORC side and the N-1 one out of the N fields is to rid of the null?
Best, Ryan On Fri, Jul 31, 2020 at 10:52 AM Owen O'Malley <owen.omal...@gmail.com> wrote: > Ryan, > I did just look at that code in Spark last week. The problem as you > correctly surmised is that Spark > doesn't have a uniontype. I think we probably need a fix that converts the > uniontype into a struct for > Spark. In such a translation, you would have fields for each variant of the > union and N-1 of the N fields > would be null for each row. > > .. Owen > > On Thu, Jul 30, 2020 at 9:19 AM Ryan Schachte <coderyanschac...@gmail.com> > wrote: > > > I am writing ORC binaries in Java and they deserialize perfectly with the > > Apache ORC jar on the docs that I've used to validate the data. The > schemas > > looks good, etc. > > > > When reading this information via Spark, we are encountering failures - > in > > particular > > > > mismatched input '<' expecting '>'(line 1, pos 6569) > > taxPercent:uniontype<int,float>, > > > > Does Spark support uniontypes like this? Just curious what some > > plausible work arounds could be. > > Thanks. > > >