[
https://issues.apache.org/jira/browse/ARROW-1734?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16220434#comment-16220434
]
Wes McKinney commented on ARROW-1734:
-------------------------------------
I would say option 2 is the simplest thing. It would be great to be able to do
option 3 eventually, but we'd need to do a little work to enable the cast
kernels to write into sliced outputs. We will likely want to have the option to
either map the chunks independently (which will always be required for string /
variable length output) or write output into a single large array. I'll open a
JIRA for that (since if the offset is non-zero, the implementations are
incorrect)
> C++/Python: Add cast function on Column-level
> ---------------------------------------------
>
> Key: ARROW-1734
> URL: https://issues.apache.org/jira/browse/ARROW-1734
> Project: Apache Arrow
> Issue Type: New Feature
> Components: C++, Python
> Reporter: Uwe L. Korn
> Assignee: Uwe L. Korn
>
> Add a {{Cast}} function on a column level that casts all underlying arrays.
> There are three possible options we could implement:
> * Make a special cast implementation that takes a ChunkedArray and returns a
> simple Array
> * Make a universal implementation that takes a ChunkedArray and applies a
> Kernel on each Array, returning a ChunkedArray with the same chunking
> * Make a universal implementation that takes a ChunkedArray and applies a
> Kernel on each Array, returning a plain Array with no chunking
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
