[ 
https://issues.apache.org/jira/browse/HIVE-298?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12675821#action_12675821
 ] 

Zheng Shao commented on HIVE-298:
---------------------------------

The alias for serde (shortNames) turned out to be a pain while we were 
migrating hive1.  Java does not automatically load classes, so in order to 
support serde aliases we would have to load all classes which takes more than 
10 seconds in usual case.

As I remember that is why we have moved away from shortNames (there are no 
shortNames in the code or in the MetaStore right now)

We actually still have serde name mappings in both serde 1.5 and serde 2 code, 
to make sure they are compatible with each other.


I agree we can't keep putting forward compatibility patches. This is just meant 
for safer migration from 0.2 to 0.3. There won't be such a need for 0.2 to 0.4.
Without this patch it is harder to push people to test LazySimpleSerDe (since 
table created by them won't be usable by other people unless they are all on 
branch 0.3).

But I am OK to go without it, as long as we can get enough people to test out 
LazySimpleSerDe.


> Make branch 0.2 forward compatible with tables created with LazySimpleSerDe
> ---------------------------------------------------------------------------
>
>                 Key: HIVE-298
>                 URL: https://issues.apache.org/jira/browse/HIVE-298
>             Project: Hadoop Hive
>          Issue Type: Improvement
>            Reporter: Zheng Shao
>
> We can easily make branch 0.2 forward-compatible with tables created with 
> LazySimpleSerDe (starting branch 0.3), since LazySimpleSerDe has exactly the 
> same serialized format as TCTLSeparatedProtocol and 
> MetadataTypedColumnsetSerDe.
> The plan is to directly map it to TCTLSeparatedProtocal.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to