[
https://issues.apache.org/jira/browse/ARROW-608?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wes McKinney updated ARROW-608:
-------------------------------
Description:
While we've decided to make the primary IPC date type be int64 milliseconds
since the UNIX epoch, in many libraries dates are represented as integer
(int32, usually) days since some epoch. In the Python standard library, the
epoch is the year 0:
{code}
>>> d = datetime.date(2017, 1, 17)
>>> d.toordinal()
736346
>>> d.toordinal() / 365
2017
{code}
At least in Cplusplus-land, in working on ARROW-452 I ran into the problem of
how to do zero-copy reads of such data, while preserving the metadata to know
that the values are dates. I added a cpp-only "date32" type to support this use
case https://github.com/apache/arrow/pull/365
I'm not sure whether we should add a new logical type, but thought it would be
worth bringing up in any case
was:
While we've decided to make the primary IPC date type be int64 milliseconds
since the UNIX epoch, in many libraries dates are represented as integer
(int32, usually) days since some epoch. In the Python standard library, the
epoch is the year 0:
{code}
>>> d = datetime.date(2017, 1, 17)
>>> d.toordinal()
736346
>>> d.toordinal() / 365
2017
{code}
At least in C++-land, in working on ARROW-452 I ran into the problem of how to
do zero-copy reads of such data, while preserving the metadata to know that the
values are dates. I added a C++-only "date32" type to support this use case
https://github.com/apache/arrow/pull/365
I'm not sure whether we should add a new logical type, but thought it would be
worth bringing up in any case
> [Format] Days since epoch date type
> -----------------------------------
>
> Key: ARROW-608
> URL: https://issues.apache.org/jira/browse/ARROW-608
> Project: Apache Arrow
> Issue Type: New Feature
> Components: Format
> Reporter: Wes McKinney
>
> While we've decided to make the primary IPC date type be int64 milliseconds
> since the UNIX epoch, in many libraries dates are represented as integer
> (int32, usually) days since some epoch. In the Python standard library, the
> epoch is the year 0:
> {code}
> >>> d = datetime.date(2017, 1, 17)
> >>> d.toordinal()
> 736346
> >>> d.toordinal() / 365
> 2017
> {code}
> At least in Cplusplus-land, in working on ARROW-452 I ran into the problem of
> how to do zero-copy reads of such data, while preserving the metadata to know
> that the values are dates. I added a cpp-only "date32" type to support this
> use case https://github.com/apache/arrow/pull/365
> I'm not sure whether we should add a new logical type, but thought it would
> be worth bringing up in any case
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
