[
https://issues.apache.org/jira/browse/HIVE-5976?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14057974#comment-14057974
]
David Chen commented on HIVE-5976:
----------------------------------
[~brocknoland] I realized that the reason why the alter table tests are now
passing and a number of the create table tests are now failing is that the
create table and alter table codepaths do things differently in terms of
setting the SerDe for text and sequencefile. The create table codepath does not
set the SerDe for these two storage formats (see BaseSemanticAnalyzer.
StorageFormat.fillStorageFormat()).
However, the alter table codepath, in fact, does (see
DDLSemanticAnalyze.analyzeAlterTableFileFormat()) and sets them to
LazySimpleSerDe. However, now that both code paths go through the new
StorageFormat class, not setting the SerDe to LazySimpleSerDe for text and
sequencefile causes alter table fileformat to fail because the SerDe remains
unchanged -- which is clearly incorrect -- but setting the SerDe to
LazySimpleSerDe causes the create table tests to fail because now we have one
extra line in the output because the SerDe is now being set.
It seems to me that the create table codepath should also set the SerDe to
LazySimpleSerDe for text and sequencefile. Is there a reason why it is
currently not doing so?
> Decouple input formats from STORED as keywords
> ----------------------------------------------
>
> Key: HIVE-5976
> URL: https://issues.apache.org/jira/browse/HIVE-5976
> Project: Hive
> Issue Type: Task
> Reporter: Brock Noland
> Assignee: Brock Noland
> Attachments: HIVE-5976.2.patch, HIVE-5976.3.patch, HIVE-5976.3.patch,
> HIVE-5976.4.patch, HIVE-5976.5.patch, HIVE-5976.6.patch, HIVE-5976.patch,
> HIVE-5976.patch, HIVE-5976.patch, HIVE-5976.patch
>
>
> As noted in HIVE-5783, we hard code the input formats mapped to keywords.
> It'd be nice if there was a registration system so we didn't need to do that.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
