[ https://issues.apache.org/jira/browse/ARROW-1950?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16304052#comment-16304052 ]
Phillip Cloud edited comment on ARROW-1950 at 12/26/17 10:59 PM: ----------------------------------------------------------------- There are a couple issues here: 1. We're mapping arrow null types to float64 because of nans, but I don't think that's correct since null is both tye type of typeless missing values _and_ the type of empty things, including the value type for empty lists and maps. Really the {{pandas_type}} should be {{'empty'}}. 2. The C++ {{table_to_blocks}} function doesn't actually support null types was (Author: cpcloud): There are a couple issues here: 1. We're mapping arrow null types to float64 because of nans, but I don't think that's correct since null is both tye type of typeless missing values _and_ the type of empty things, including the value type for empty lists and maps. Really the {{pandas_type}} should be {{'empty'}}. 2. The C++ table_to_blocks function doesn't actually support null types > [Python] pandas_type in pandas metadata incorrect for List types > ---------------------------------------------------------------- > > Key: ARROW-1950 > URL: https://issues.apache.org/jira/browse/ARROW-1950 > Project: Apache Arrow > Issue Type: Bug > Components: Python > Reporter: Wes McKinney > Fix For: 0.9.0 > > > see https://github.com/pandas-dev/pandas/pull/18201#issuecomment-353042438 -- This message was sent by Atlassian JIRA (v6.4.14#64029)