[jira] [Commented] (DRILL-2185) flatten(*) should return a more appropriate error

["Jason Altekruse (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/DRILL-2185?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14312443#comment-14312443 ] 

Jason Altekruse commented on DRILL-2185:
----------------------------------------

While flatten is materialized into an operator, it is treated as a function for most of the planning process. We can certainly document that * is specifically not supported by flatten, but I have no idea what user would expect this to do something meaningful. It seems in the same vain as aggregating over *, or trying to call a scalar function on * which are both nonsensical in traditional databases. There is an argument that we should support this eventually as input to a complex function, when we merge the ideas of the top level schema map, and a map stored in a particular field. However in regards to flatten, I don't think we would ever support a repeated scalar or repeated map as the top level schema element without a name, because it wouldn't be able to be referenced in the query.

> flatten(*) should return a more appropriate error
> -------------------------------------------------
>
>                 Key: DRILL-2185
>                 URL: https://issues.apache.org/jira/browse/DRILL-2185
>             Project: Apache Drill
>          Issue Type: Bug
>          Components: Functions - Drill
>            Reporter: Rahul Challapalli
>            Assignee: Daniel Barclay (Drill/MapR)
>            Priority: Minor
>
> git.commit.id.abbrev=c54bd6a
> The below query returns an obscure error
> {code}
> 0: jdbc:drill:schema=dfs> select flatten(*) from `data1.json`;
> Query failed: AssertionError: Internal error: invariant violated: conversion result not null
> {code}
> We should mention that it is an in-appropriate use of the flatten operator



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)