[ https://issues.apache.org/jira/browse/ARROW-12042?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17306420#comment-17306420 ]
Antoine Pitrou commented on ARROW-12042: ---------------------------------------- I definitely don't see this in Python. This looks like a bug on the R side? {code:python} >>> arr = pa.chunked_array([[2, 1], [4, 3]]) >>> pc.sort_indices(arr) <pyarrow.lib.UInt64Array object at 0x7f4e4a121c90> [ 1, 0, 3, 2 ] >>> arr.take(pc.sort_indices(arr)) <pyarrow.lib.ChunkedArray object at 0x7f4e4a11e170> [ [ 1, 2, 3, 4 ] ] {code} > [C++] Change or rationalize output of array_sort_indices on ChunkedArray > ------------------------------------------------------------------------ > > Key: ARROW-12042 > URL: https://issues.apache.org/jira/browse/ARROW-12042 > Project: Apache Arrow > Issue Type: Task > Components: C++ > Affects Versions: 3.0.0 > Reporter: Ian Cook > Priority: Major > > Currently when the {{array_sort_indices}} compute function is called on a > ChunkedArray of two or more Arrays, it returns a ChunkedArray of Arrays of > _local_ sort indices for each Array. Demonstrating this with the R bindings > (but note that these R examples will not run until ARROW-11703 is merged): > {code:java} > > x <- ChunkedArray$create(c(2L, 1L), c(4L, 3L)) > > arrow:::call_function("array_sort_indices", x, options = list(order = > > FALSE)) > ChunkedArray > [ > [ > 1, > 0 > ], > [ > 1, > 0 > ] > ] > {code} > Compare to the {{sort_indices}} compute function which returns an Array of > _global_ sort indices in this case: > {code:java} > > arrow:::call_function("sort_indices", x, options = list(names = "", orders > > = 0L)) > Array > <uint64> > [ > 1, > 0, > 3, > 2 > ]{code} > Is this behavior deliberate? If so, we should document it clearly. If not, we > should change it. > Note that the docs currently states that {{array_sort_indices}} only works on > Arrays [https://arrow.apache.org/docs/cpp/compute.html#sorts-and-partitions] > (see note (4)) but evidently that is not exactly correct. -- This message was sent by Atlassian Jira (v8.3.4#803005)