You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Fabian Hueske (JIRA)" <ji...@apache.org> on 2016/01/26 09:58:39 UTC
[jira] [Commented] (FLINK-3227) Translate physical Table API plan
into DataSet program
[ https://issues.apache.org/jira/browse/FLINK-3227?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15116949#comment-15116949 ]
Fabian Hueske commented on FLINK-3227:
--------------------------------------
As per discussion in [PR #1544|https://github.com/apache/flink/pull/1544], we decided to embed the dataSet translation code directly in the Flink RelNodes. Hence, this issue becomes obsolete.
Any objections to close it?
I will start a discussion on FLINK-3226 to split the work into smaller parts.
> Translate physical Table API plan into DataSet program
> ------------------------------------------------------
>
> Key: FLINK-3227
> URL: https://issues.apache.org/jira/browse/FLINK-3227
> Project: Flink
> Issue Type: Sub-task
> Components: Table API
> Reporter: Fabian Hueske
> Assignee: Timo Walther
>
> The RelNode tree generated by FLINK-3226 is a 1-to-1 representation of the final DataSet API program. Hence, the translation step only needs to create corresponding operators with correct keys and user-functions. User-functions are implemented as RichFunctions that wrap generated code. In a first version, the wrapping functions are parameterized with a source code string which is generated during the program construction. As a first step, we can reuse much of the code generation logic of the current Table API.
> The source code string is compiled in the open() method of the wrapping function and later invoked by the wrapper.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)