[
https://issues.apache.org/jira/browse/HADOOP-11740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14383161#comment-14383161
]
Kai Zheng commented on HADOOP-11740:
------------------------------------
Thanks [~zhz] for the good thoughts.
bq.We can get rid of ErasureCoder since it has a single subclass now
(AbstractErasureCoder)
Yes it's often a good question to think about either interface or abstract
class in Java. My feeling is that if it's in a framework, subject to be
pluggable and implemented by customers, an interface would be good to have. So
I guess we could keep {{ErasureCoder}} interface, and convert
{{ErasureCodingStep}} interface to a class.
bq.If ECBlockGroup can provide erased indices, we can further combine encoding
and decoding classes
I'm not sure, as erased indices have to be be computed according to input
blocks and output blocks just for decoders, and encoders don't have the related
logics. Currently in RS coder the decoding is rather simple but I believe it
will be much complicated for codes like LRC and Hitchhiker, so separating
encoding class and decoding class is desired.
> Combine erasure encoder and decoder interfaces
> ----------------------------------------------
>
> Key: HADOOP-11740
> URL: https://issues.apache.org/jira/browse/HADOOP-11740
> Project: Hadoop Common
> Issue Type: Sub-task
> Components: io
> Reporter: Zhe Zhang
> Assignee: Zhe Zhang
> Attachments: HADOOP-11740-000.patch
>
>
> Rationale [discussed |
> https://issues.apache.org/jira/browse/HDFS-7337?focusedCommentId=14376540&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14376540]
> under HDFS-7337.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
