[ https://issues.apache.org/jira/browse/ARROW-835?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16210452#comment-16210452 ]
Jeff Reback commented on ARROW-835: ----------------------------------- computationally using ns precesion causes overflow a fair amount, range is only 106k days (though it’s easily detected); might consider using an alternative storage (eg store days in an int32 and then ns in a int64), but this might be at odds with other systems > [Format] Add Timedelta type to describe time intervals > ------------------------------------------------------ > > Key: ARROW-835 > URL: https://issues.apache.org/jira/browse/ARROW-835 > Project: Apache Arrow > Issue Type: Improvement > Reporter: Jeff Reback > Assignee: Wes McKinney > Priority: Minor > Fix For: 0.8.0 > > > xref https://github.com/apache/arrow/pull/551 and > https://github.com/apache/arrow/pull/551#issuecomment-294325969 > this will allow round-tripping of pandas ``Timedelta`` and numpy > ``timedelt64[ns]`` types. The will have a similar TimeUnit to TimestampType > (s, us, ms, ns). Possible impl include making this pure 64-bit. -- This message was sent by Atlassian JIRA (v6.4.14#64029)