tustvold commented on code in PR #5242:
URL: https://github.com/apache/arrow-datafusion/pull/5242#discussion_r1103772779
##########
datafusion/core/src/physical_plan/sorts/sort.rs:
##########
@@ -38,13 +38,14 @@ use crate::physical_plan::{
RecordBatchStream, SendableRecordBatchStream, Statistics,
};
use crate::prelude::SessionConfig;
-use arrow::array::{make_array, Array, ArrayRef, MutableArrayData};
+use arrow::array::{make_array, Array, ArrayRef, MutableArrayData, UInt32Array};
pub use arrow::compute::SortOptions;
use arrow::compute::{concat, lexsort_to_indices, take, SortColumn,
TakeOptions};
use arrow::datatypes::SchemaRef;
use arrow::error::ArrowError;
use arrow::ipc::reader::FileReader;
use arrow::record_batch::RecordBatch;
+use arrow::row::{OwnedRow, Row, RowConverter, SortField};
Review Comment:
Using OwnedRow is likely to seriously hurt performance, as it will allocate
per row.
What do you think about getting the v1 in that doesn't preserve the row
encoding, and then we can iterate on v2 in a separate PR. I suspect we will
want to preserve rows, and a list of sorted indices.
--
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]
