[
https://issues.apache.org/jira/browse/HIVE-2773?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13214979#comment-13214979
]
Ashutosh Chauhan commented on HIVE-2773:
----------------------------------------
bq. If I add new methods, classes implementing the interface will have to
implement the new methods anyway,
Adding or removing methods have different impacts from point of view of
backward compatibility.
If you add new methods in an interface then implementations using the older
interface will continue to work even after they upgrade to new version of hive.
That has no cost on people who have implemented the old interface. Only if they
recompile it, then they have to implement new methods.
On the other hand, if you remove methods, then their old implementation won't
work anymore after upgrade. They have to recompile and this implies work on
their part.
This is the difference which gets factored in backward compatibility. So, as
long as you are adding new methods in interface, I consider it to be backward
compatible. But, if you are removing methods, then its not. Makes sense?
> HiveStorageHandler.configureTableJobProperites() should let the handler know
> wether it is configuration for input or output
> ---------------------------------------------------------------------------------------------------------------------------
>
> Key: HIVE-2773
> URL: https://issues.apache.org/jira/browse/HIVE-2773
> Project: Hive
> Issue Type: Improvement
> Reporter: Francis Liu
> Assignee: Francis Liu
> Labels: hcatalog, storage_handler
> Attachments: HIVE-2773.D1815.1.patch, HIVE-2773.patch
>
>
> HiveStorageHandler.configureTableJobProperties() is called to allow the
> storage handler to setup any properties that the underlying
> inputformat/outputformat/serde may need. But the handler implementation does
> not know whether it is being called for configuring input or output. This
> makes it a problem for handlers which sets an external state. In the case of
> HCatalog's HBase storageHandler, whenever a write needs to be configured we
> create a write transaction which needs to be committed or aborted later on.
> In this case configuring for both input and output each time
> configureTableJobProperties() is called would not be desirable. This has
> become an issue since HCatalog is dropping storageDrivers for SerDe and
> StorageHandler (see HCATALOG-237).
> My proposal is to replace configureTableJobProperties() with two methods:
> configureInputJobProperties()
> configureOutputJobProperties()
> Each method will have the same signature. I cursory look at the code and I
> believe changes should be straighforward also given that we are not really
> changing anything just splitting responsibility. If the community is fine
> with this approach I will go ahead and create a aptch.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
