domoritz commented on PR #35780: URL: https://github.com/apache/arrow/pull/35780#issuecomment-1846069737
It looks like what happens is that when you put differently chunked vectors into a table, the record batch boundaries are created at the chunk gaps from all vectors. So if you have two vectors with chunk lengths of 2,3 and 3,2, you will get record batches of length 2,1,2. I would hope that the data buffers are views on the existing buffers. The only issue I can see with this approach is that in the case of variable width vectors (like utf8)---where we have a data buffer where we might reference the same position multiple times from the offset buffer---we would need to copy the data buffer in serialized record batches. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
