[ https://issues.apache.org/jira/browse/ARROW-13993?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17488095#comment-17488095 ]
David Li commented on ARROW-13993: ---------------------------------- Yes, we just want a single tuple. Any tuple will do; the point above is that we can't implement anything else (because the query engine currently lacks support for ordering, beyond sorting outputs at the very end). All hash_ kernels ("hash aggregate kernels") are in [{{hash_aggregate.cc}}|https://github.com/apache/arrow/blob/master/cpp/src/arrow/compute/kernels/hash_aggregate.cc] and it will be very similar to the CountDistinct/Distinct implementation there. > [C++] Hash aggregate function that returns value from first row in group > ------------------------------------------------------------------------ > > Key: ARROW-13993 > URL: https://issues.apache.org/jira/browse/ARROW-13993 > Project: Apache Arrow > Issue Type: Improvement > Components: C++ > Reporter: Ian Cook > Assignee: Dhruv Vats > Priority: Major > Labels: good-second-issue, kernel > Fix For: 8.0.0 > > > It would be nice to have a hash aggregate function that returns the first > value of a column within each hash group. > If row order within groups is non-deterministic, then effectively this would > return one arbitrary value. This is a very computationally cheap operation. > This can be quite useful when querying a non-normalized table. For example if > you have a table with a {{country}} column and also a {{country_abbr}} column > and you want to group by either/both of those columns but return the values > from both columns, you could do > {code:java} > SELECT country, country_abbr FROM table GROUP BY country, country_abbr{code} > but it would be more efficient to do > {code:java} > SELECT country, first(country_abbr) FROM table GROUP BY country{code} > because then the engine does not need to scan all the values of the > {{country_abbr}} column. -- This message was sent by Atlassian Jira (v8.20.1#820001)