[
https://issues.apache.org/jira/browse/BEAM-1249?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15805287#comment-15805287
]
Thomas Groh commented on BEAM-1249:
-----------------------------------
Resolution for runners is currently up to the runner.
There's no reason for the model to forbid disjoint coders - the contents of the
output {{PCollection}} should be encoded with the coder of that
{{PCollection}}, and the contents of the input {{PCollections}} can be decoded
with their coders, as in any other {{PTransform}}. Depending on how the Runner
handles Flatten, this can be anywhere from a no-op (if the inputs to the
flatten do not need to be decoded nor outputs immediately encoded) to a
decode-and-encode if the inputs are and output should be materialized.
> Flatten with heterongeneous coders does not have a RunnableOnService test
> -------------------------------------------------------------------------
>
> Key: BEAM-1249
> URL: https://issues.apache.org/jira/browse/BEAM-1249
> Project: Beam
> Issue Type: Bug
> Components: sdk-java-core
> Reporter: Kenneth Knowles
> Assignee: Thomas Groh
> Priority: Minor
> Fix For: 0.5.0
>
>
> Because there is no such test, we do not know which runners support this. It
> is perfectly acceptable in the model as long as the types line up (including
> nullability) though in general might incur a needless serde so also
> reasonable to just not support it.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
