Re: Cross-implementation metadata specification, IPC details

2016-03-06 Thread Chao Sun
Thanks Wes. This makes sense. Best, Chao On Sun, Mar 6, 2016 at 2:01 PM, Wes McKinney wrote: > Chao -- yes, I believe you're right. > > In an IPC message, if the node is required or has null_count == 0, > then the null bitmap can be omitted from the payload, but otherwise > it's there (this is

Re: Cross-implementation metadata specification, IPC details

2016-03-06 Thread Wes McKinney
Chao -- yes, I believe you're right. In an IPC message, if the node is required or has null_count == 0, then the null bitmap can be omitted from the payload, but otherwise it's there (this is similar to the Hive HS2 V6 protocol). One alternative to this is that rather than omitting the bitmap, it

Re: Cross-implementation metadata specification, IPC details

2016-03-06 Thread Chao Sun
Hi, Sorry if this is a n00b question. For the example that Jacques used in the previous thread, how does it work if the struct is nullable - shouldn't there be a is_null array between 1 and 2? For instance, with the example: list elem #0: { , null, } list elem #1: { } list elem #2: null list e

Re: Cross-implementation metadata specification, IPC details

2016-03-01 Thread Wes McKinney
Inline responses On Tue, Mar 1, 2016 at 8:24 AM, Jacques Nadeau wrote: > Wes, thanks for starting this conversation. > > Couple thoughts: > > For metadata, we have two models existing (one in the ValueVectors approach > and one in Parquet). It seems like we should start from one of those and > th

Re: Cross-implementation metadata specification, IPC details

2016-03-01 Thread Jacques Nadeau
Wes, thanks for starting this conversation. Couple thoughts: For metadata, we have two models existing (one in the ValueVectors approach and one in Parquet). It seems like we should start from one of those and then shape as appropriate. It seems like we have a richer physical capability that the