The purpose would be to write only the record batch to the device memory, while leaving the Schema on the CPU. Similarly, record batches can be reconstructed through pointer arithmetic (the IPC loading as in arrow/ipc/reader.h) without having to copy the data to main memory
Reporter: Wes McKinney / @wesm
Assignee: Wes McKinney / @wesm
Related issues:
Note: This issue was originally created as ARROW-1364. Please see the migration documentation for further details.