viirya commented on code in PR #2680:
URL: https://github.com/apache/arrow-rs/pull/2680#discussion_r966750654
##########
arrow/src/compute/kernels/sort.rs:
##########
@@ -311,41 +312,58 @@ pub fn sort_to_indices(
)));
}
},
- DataType::Dictionary(key_type, value_type)
- if *value_type.as_ref() == DataType::Utf8 =>
- {
- match key_type.as_ref() {
- DataType::Int8 => {
- sort_string_dictionary::<Int8Type>(values, v, n, &options,
limit)
- }
- DataType::Int16 => {
- sort_string_dictionary::<Int16Type>(values, v, n,
&options, limit)
- }
- DataType::Int32 => {
- sort_string_dictionary::<Int32Type>(values, v, n,
&options, limit)
- }
- DataType::Int64 => {
- sort_string_dictionary::<Int64Type>(values, v, n,
&options, limit)
- }
- DataType::UInt8 => {
- sort_string_dictionary::<UInt8Type>(values, v, n,
&options, limit)
- }
- DataType::UInt16 => {
- sort_string_dictionary::<UInt16Type>(values, v, n,
&options, limit)
- }
- DataType::UInt32 => {
- sort_string_dictionary::<UInt32Type>(values, v, n,
&options, limit)
- }
- DataType::UInt64 => {
- sort_string_dictionary::<UInt64Type>(values, v, n,
&options, limit)
- }
- t => {
- return Err(ArrowError::ComputeError(format!(
- "Sort not supported for dictionary key type {:?}",
- t
- )));
- }
- }
+ DataType::Dictionary(_, _) => {
+ downcast_dictionary_array!(
+ values => match values.values().data_type() {
+ DataType::Int8 => {
+ let dict_values = values.values();
+ let sorted_value_indices =
sort_to_indices(dict_values, Some(SortOptions::default()), None)?;
Review Comment:
We need to have fixed sort options (default) when sorting the values of
dictionary. As we will compare keys based on the sorted indices. The sorted
indices represent the order of values which is fixed.
For example, a dictionary (keys = [1, 2, 0], values = [0, 1, 2]). No matter
we sort in ascending order or in descending order, when we compare key = 1 and
key = 2, their relation is always (key = 1) < (key = 2) according to the values
behind the keys.
--
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]
