[
https://issues.apache.org/jira/browse/HADOOP-12685?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15082068#comment-15082068
]
Kai Zheng commented on HADOOP-12685:
------------------------------------
Now I thought it wouldn't be bad to enforce the refined behavior in the
abstract layer as you did in the patch, even concrete coders may do that
already.
Ref. the following change:
{noformat}
+ // dataLen bytes consumed
for (int i = 0; i < inputs.length; ++i) {
- buffer = inputs[i];
- if (buffer != null) {
- // dataLen bytes consumed
- buffer.position(buffer.position() + dataLen);
+ // move the position forward only if it's not moved during decoding
+ if (inputs[i] != null && inputs[i].position() == inputPositions[i]) {
+ inputs[i].position(inputs[i].position() + dataLen);
}
}
{noformat}
Can be better:
{code}
// dataLen bytes consumed
for (int i = 0; i < inputs.length; i++) {
buffer = inputs[i];
if (buffer != null) {
// dataLen bytes consumed
buffer.position(inputPositions[i] + dataLen);
}
}
{code}
> Input buffer position after encode/decode not consistent between different
> kinds of buffers
> -------------------------------------------------------------------------------------------
>
> Key: HADOOP-12685
> URL: https://issues.apache.org/jira/browse/HADOOP-12685
> Project: Hadoop Common
> Issue Type: Sub-task
> Reporter: Rui Li
> Assignee: Rui Li
> Priority: Minor
> Attachments: HADOOP-12685.1.patch
>
>
> As reported by [~jack_liuquan] in this
> [comment|https://issues.apache.org/jira/browse/HADOOP-11828?focusedCommentId=15071796&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15071796],
> the input buffer position change behavior is not consistent between direct
> and non-direct buffers.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
