[
https://issues.apache.org/jira/browse/HIVE-9555?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14366783#comment-14366783
]
Gopal V commented on HIVE-9555:
-------------------------------
[~prasanth_j]: I saw that, which is why I so missed Object mix-ins for caching.
The main reason is that I suspect the way [~sershe] has done the visibility
change is that it's the most performant way of doing this instead of wrapping
abstraction layers on top of this.
An alternative approach to prevent exposure is to have a
RecordReaderImplFactory and declare all the TreeReaders as protected static
inside that, inherit an EncodedRecordReaderImplFactory and define the encoded
reader classes in that.
Leaving some other comments on RB.
> assorted ORC refactorings for LLAP on trunk
> -------------------------------------------
>
> Key: HIVE-9555
> URL: https://issues.apache.org/jira/browse/HIVE-9555
> Project: Hive
> Issue Type: Bug
> Reporter: Sergey Shelukhin
> Assignee: Sergey Shelukhin
> Attachments: HIVE-9555.01.patch, HIVE-9555.02.patch,
> HIVE-9555.03.patch, HIVE-9555.04.patch, HIVE-9555.05.patch,
> HIVE-9555.06.patch, HIVE-9555.07.patch, HIVE-9555.patch
>
>
> To minimize conflicts and given that ORC is being developed rapidly on trunk,
> I would like to refactor some parts of ORC "in advance" based on the changes
> in LLAP branch. Mostly it concerns making parts of ORC code (esp. SARG, but
> also some internal methods) more modular and easier to use from alternative
> codepaths. There's also significant change to how data reading is handled -
> BufferChunk inherits from DiskRange; the reader receives a list of
> DiskRange-s (as before), but instead of making a list of buffer chunks it
> replaces ranges with buffer chunks in the original (linked) list.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
