Details
-
New Feature
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
Description
Currently, if arrow detects that every value in an int64 column can fit in a 32bit int, it will downcast the column an set the type to integer instead of integer64. Not having a mechanism to disable this optimization makes it tricky if you have many parallel processes (think HPC use case) performing the same calculation but potentially outputting different result values, some being >2^32 and others not. When you go to collect the resulting feather files from the parallel processes, the types across the files may not line up.
Feature request is to provide an option to disable this demotion and maintain the source column type.
Attachments
Issue Links
- is related to
-
ARROW-14509 as_vector() downgrades int64 even when arrow.int64_downcast = TRUE
- Closed
- links to