[
https://issues.apache.org/jira/browse/BEAM-2676?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16107771#comment-16107771
]
Robert Bradshaw commented on BEAM-2676:
---------------------------------------
The motivation for this came up because (1) we're currently lacking a Generic
tuple-like class (which would be useful for joins and a Java port for
https://github.com/tensorflow/transform , to name a few) and (2) if we have
such a class, SQL should use it rather than do its own thing if at all
possible.
The BeamRow object itself is not at all SQL centric, however BeamRowType is.
The original proposal was to have BeamRow (I would call it BeamRecord, but
don't feel that strongly) be a generic type in core, a sibling to KV, along
with an abstract BeamRowType (which, among other things, vends the Coder from a
list of component coders). Then one would have a concrete BeamSqlRowType
subclass of BeamRowType which is instantiated via java.sql.Types. This is the
most forward compatible state without tying the tuple to SQL.
(I added an item to mark the entire API as Experimental for now, which I think
makes a lot of sense.)
> move BeamSqlRow and BeamSqlRowType to sdk/java/core
> ---------------------------------------------------
>
> Key: BEAM-2676
> URL: https://issues.apache.org/jira/browse/BEAM-2676
> Project: Beam
> Issue Type: Test
> Components: dsl-sql
> Reporter: Xu Mingmin
> Assignee: Xu Mingmin
> Labels: dsl_sql_merge
>
> BeamSqlRow/BeamSqlRowType is the fundamental of structured data processing in
> Beam, like joins, simple projections/expansions. It's more visible to move
> them to sdk-java-core.
> It contains two parts:
> 1). remove SQL word in the name,
> BeamSqlRow --> BeamRow
> BeamSqlRowType --> BeamRowType
> 2). move from package {{org.apache.beam.dsls.sql.schema}} to
> {{org.apache.beam.sdk.sd}} (sd stands for structure data), in module
> {{beam-sdks-java-core}}
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
