[
https://issues.apache.org/jira/browse/ARROW-13617?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17415501#comment-17415501
]
David Li commented on ARROW-13617:
----------------------------------
Another reason why this would be good: nominally Decimal is the same as
FixedSizeBinary, however, in compute kernels, currently they need separate
handling because their scalar representations are different (Decimal vs
Buffer). You could imagine trying to work with all these types via
util::string_view or const uint8_t*, however, the representation for
BasicDecimal128 makes this hard; there is no native-endian representation that
we can just create a pointer to. (Well, the fields in BasicDecimal128 are in
native-endian order so you can reinterpret_cast a pointer to the right field,
but it would be a little cleaner to have both implementations of
native_endian_array() return a std::array& instead of a std::array).
> [C++] Make Decimal representations consistent
> ---------------------------------------------
>
> Key: ARROW-13617
> URL: https://issues.apache.org/jira/browse/ARROW-13617
> Project: Apache Arrow
> Issue Type: Improvement
> Components: C++
> Reporter: David Li
> Priority: Major
>
> Right now Decimal128 is represented as two 64-bit integers (in native endian
> order) and Decimal256 is represented as a std::array (in native endian
> order). We could make these consistent (presumably there should be no
> performance impact). It may also make it easier to extract a common base
> class or other niceties.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
