Jefffrey commented on code in PR #22285:
URL: https://github.com/apache/datafusion/pull/22285#discussion_r3301980338
##########
datafusion/functions-nested/src/arrays_zip.rs:
##########
@@ -327,3 +332,282 @@ fn arrays_zip_inner(args: &[ArrayRef]) ->
Result<ArrayRef> {
Ok(Arc::new(result))
}
+
+fn arrays_zip_field_name(index: usize) -> String {
+ (index + 1).to_string()
+}
+
+fn arrays_zip_field_names(len: usize) -> Vec<String> {
+ (0..len).map(arrays_zip_field_name).collect()
+}
+
+/// Fast path for regular List inputs whose existing buffers already match the
+/// zipped output: all offsets and values lengths match, and null rows cover no
+/// values. This lets us reuse offsets and child values instead of rebuilding.
+fn try_perfect_list_zip(
+ args: &[ArrayRef],
+ field_names: &[String],
+) -> Result<Option<ArrayRef>> {
+ debug_assert_eq!(args.len(), field_names.len());
+
+ let mut list_arrays = Vec::with_capacity(args.len());
+ let mut struct_fields = Vec::with_capacity(args.len());
+
+ for (arg, field_name) in args.iter().zip(field_names) {
+ let arr = match arg.data_type() {
+ List(field) => {
+ struct_fields.push(Field::new(
+ field_name.clone(),
+ field.data_type().clone(),
+ true,
+ ));
+ as_list_array(arg)?
+ }
+ _ => return Ok(None),
+ };
+
+ list_arrays.push(arr);
+ }
+
+ let first = list_arrays[0];
+ let num_rows = first.len();
+ let offsets = first.offsets().clone();
+ let values_len = first.values().len();
+
+ // Reusing the child arrays is only valid when every list uses the exact
+ // same row boundaries and exposes the same total number of child values.
+ for arr in &list_arrays {
+ if arr.values().len() != values_len || arr.offsets() != &offsets {
+ return Ok(None);
+ }
+ }
+
+ let nulls = if list_arrays.iter().any(|arr| arr.null_count() != 0) {
+ let first_nulls = first.nulls();
+ if list_arrays.iter().all(|arr| arr.nulls() == first_nulls) {
+ NullBuffer::union_many(list_arrays.iter().map(|arr| arr.nulls()))
Review Comment:
This union is unnecessary since we check the nulls are already equal
##########
datafusion/functions-nested/src/arrays_zip.rs:
##########
@@ -327,3 +332,226 @@ fn arrays_zip_inner(args: &[ArrayRef]) ->
Result<ArrayRef> {
Ok(Arc::new(result))
}
+
+fn arrays_zip_field_name(index: usize) -> String {
+ (index + 1).to_string()
+}
+
+fn arrays_zip_field_names(len: usize) -> Vec<String> {
+ (0..len).map(arrays_zip_field_name).collect()
+}
+
+/// Fast path for regular List inputs whose existing buffers already match the
+/// zipped output: all offsets and values lengths match, and null rows cover no
+/// values. This lets us reuse offsets and child values instead of rebuilding.
+fn try_perfect_list_zip(
+ args: &[ArrayRef],
+ field_names: &[String],
+) -> Result<Option<ArrayRef>> {
+ debug_assert_eq!(args.len(), field_names.len());
+
+ let mut list_arrays = Vec::with_capacity(args.len());
+ let mut struct_fields = Vec::with_capacity(args.len());
+
+ for (arg, field_name) in args.iter().zip(field_names) {
+ let arr = match arg.data_type() {
+ List(field) => {
+ struct_fields.push(Field::new(
+ field_name.clone(),
+ field.data_type().clone(),
+ true,
+ ));
+ as_list_array(arg)?
+ }
+ _ => return Ok(None),
+ };
+
+ list_arrays.push(arr);
+ }
+
+ let first = list_arrays[0];
+ let num_rows = first.len();
+ let offsets = first.offsets().clone();
+ let values_len = first.values().len();
+
+ // Reusing the child arrays is only valid when every list uses the exact
+ // same row boundaries and exposes the same total number of child values.
+ for arr in &list_arrays {
+ if arr.len() != num_rows
+ || arr.values().len() != values_len
+ || arr.offsets() != &offsets
+ {
+ return Ok(None);
+ }
+ }
+
+ let nulls = if list_arrays.iter().any(|arr| arr.null_count() != 0) {
Review Comment:
Right, I missed that it is only null when all buffers are null. In that case
we could manually `OR` the null buffers together then, though not sure if that
is worth the trouble
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
