This discussion is more active on the ticket: https://issues.apache.org/jira/browse/IMPALA-3060
On Thu, Nov 30, 2017 at 9:35 PM, Jin Chul Kim <[email protected]> wrote: > Hi, > > Here are additional questions. > > 3. It seems there is no container for complex type in backend. I think we > need to introduce StructType container at be before handling IMPALA-3060. > What do you think about this? > > 4. If we can support the container for StructType, we can support not > implemented features such as specify a complex-typed column in a select > list, join between struct type. Unsupported features will be enabled > incrementally. Do you have a plan for this? > > Best regards, > Jinchul > > 2017-12-01 9:53 GMT+09:00 Jin Chul Kim <[email protected]>: > > > Hi, > > > > I would like to get your advice before implementing the feature > > https://issues.apache.org/jira/browse/IMPALA-3060 > > > > Let me start with struct type of complex types. Would you please answer > > the questions? Thanks. > > > > 1. Does StructType have getSlotSize? If your answer is yes, do you think > > the following code piece is fine to you? Here is a initial code for > > StructType. > > > > @Override > > public int getSlotSize() { > > int slotSize = 0; > > for (StructField f: fields_) slotSize += f.getType().getSlotSize(); > > slotSize += (slotSize % 8); // For 8 byte alignment > > return slotSize; > > } > > > > 2. I think we should consider materialized path for struct type, but it > > seems struct type might not be described in SlotDescriptor. > getMaterialzedPath() > > due to the guidance of the function: > > "Assembles the absolute materialized path to this slot starting from the > > schema root. > > The materialized path points to the first non-struct schema element along > > the path starting > > from the parent's tuple path to this slot's path." > > > > Can we determine materialized path for complex types? > > > > Best regards, > > Jinchul > > >
