[
https://issues.apache.org/jira/browse/HIVE-270?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12672764#action_12672764
]
Joydeep Sen Sarma commented on HIVE-270:
----------------------------------------
looks pretty good!
for the LazyString conversion - can use static function Text.decode. The
text.append will do an unnecessary byte copy into the text's internal byte
array.
LazySimpleStructObjectInspector.java: public List<Object>
getStructFieldsDataAsList(Object data) {
this is probably not used anywhere - but if data==null seems like we should
just return null? (based on looking at other object inspectors)
parse(): comparison for null sequence: the compare method is somewhat generic.
since we only care for equality - can do a simple comparison of lengths first
to find if things are unequal (should be a little faster) and do full
comparison only if lengths are equal.
it may be possible to speed up the serialize considerably as well (go directly
from Primitive types to bytes and append to a bytebuffer) - but would make
sense to punt on that.
> Add a lazy-deserialized SerDe for space and cpu efficient serialization of
> rows with primitive types
> ----------------------------------------------------------------------------------------------------
>
> Key: HIVE-270
> URL: https://issues.apache.org/jira/browse/HIVE-270
> Project: Hadoop Hive
> Issue Type: New Feature
> Components: Serializers/Deserializers
> Reporter: Zheng Shao
> Assignee: Zheng Shao
> Attachments: HIVE-270.1.patch, HIVE-270.3.patch
>
>
> We want to add a lazy-deserialized SerDe for space and cpu efficient
> serialization of rows with primitive types.
> This SerDe will share the same format as
> MetadataTypedColumnsetSerDe/TCTLSeparatedProtocol to be backward compatible.
> This SerDe will be used to replace the default table SerDe, and the SerDe
> used to communicate with user scripts.
> For simplicity, we don't plan to support nested structure with this SerDe.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
