[ARROW-12087] [C++] Fix sort_indices, array_sort_indices timestamp support discrepancy

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 6.0.0
• Component/s: C++
• Labels: kernel pull-request-available types
• External issue URL: https://github.com/apache/arrow/issues/27912
• Assignee: David Li
• Reporter: Ian Cook
• Created: 25/Mar/21 16:37
• Updated: 11/Jan/23 08:24
• Resolved: 13/Sep/21 17:49

Description

sort_indices supports sorting by timestamp arrays, but array_sort_indices does not. Here's some example R code to demonstrate this (but this example code depends on ARROW-11703 to run):

tbl <- tibble::tibble(
  dttm = lubridate::ymd_hms(c("2021-01-01 00:00:00", "1900-01-01 00:00:00")),
)
rb <- arrow::record_batch(tbl)

# this fails:
arrow:::call_function(
  "array_sort_indices",
  rb$dttm,
  options = list(order = F)
)
## Error: NotImplemented: Function array_sort_indices has no kernel matching input types (array[timestamp[us, tz=UTC]])

# this fails because it internally calls array_sort_indices
arrow:::call_function(
  "sort_indices",
  rb,
  options = list(names = "dttm", orders = 0L)
)
## Error: NotImplemented: Function array_sort_indices has no kernel matching input types (array[timestamp[us, tz=UTC]])

# this succeeds
arrow:::call_function(
  "sort_indices",
  rb,
  options = list(names = c("dttm", "dttm"), orders = 0L)
) 
## Array
## <uint64>
## [
##   1,
##   0
## ]

Issue Links

• is duplicated by: ARROW-13880 [C++] Compute function sort_indices does not support timestamps with time zones
• links to: GitHub Pull Request #11130