[jira] [Comment Edited] (ARROW-6536) [C++] CSV reader accept schema

["Francois Saint-Jacques (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ARROW-6536?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16933452#comment-16933452 ] 

Francois Saint-Jacques edited comment on ARROW-6536 at 9/19/19 2:44 PM:
------------------------------------------------------------------------

If we don't pass a schema, the minimum we should provide is common options for other readers (Parquet, Json) for selecting columns type hinting when required (JSON, CSV). This way, the dataset facility can write  a single function to do this.

I wouldn't tackle extra columns in the reader, only subset + type.


was (Author: fsaintjacques):
If we don't pass a schema, the minimum we should provide is common options for other readers (Parquet, Json) for selecting columns type hinting when required (JSON, CSV).

> [C++] CSV reader accept schema
> ------------------------------
>
>                 Key: ARROW-6536
>                 URL: https://issues.apache.org/jira/browse/ARROW-6536
>             Project: Apache Arrow
>          Issue Type: Improvement
>          Components: C++
>            Reporter: Neal Richardson
>            Priority: Major
>              Labels: csv, dataset
>
> The CSV reader lets you specify {{column_types}}, but this is an {{unordered_map}} of column name and type. Why not accept a Schema instead? Isn't that essentially an ordered map? Seems that if you took a Schema, some of the validation of what's being passed in would already have been handled. Plus, I suspect that the Datasets project will want to do even more with passing a Schema (e.g. selecting a subset of columns). 
> Thoughts [~pitrou] [~fsaintjacques] [~bkietz]?



--
This message was sent by Atlassian Jira
(v8.3.4#803005)