[
https://issues.apache.org/jira/browse/HADOOP-310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13053344#comment-13053344
]
Aaron T. Myers commented on HADOOP-310:
---------------------------------------
Hey Brock, thanks a lot for reviving this issue (after 5 years!) I have a few
small concerns about this patch:
# The constructor which only takes a {{byte[]}} can now be implemented in terms
of this new constructor you've introduced.
# We now have {{set(byte[] bytes, int length)}} and {{set(byte[] bytes, int
offset, int length)}}, which differ not only in signature, but in their
semantics with respect to copying of the input data. This seems like it might
lead to confusion. Perhaps we should name the new method something like
{{setDirect(...)}}? Or perhaps we should amend the existing {{set(...)}} method
to not do a copy? The latter is obviously a trickier and backward-incompatible
change, but probably clearer.
> Additional constructor requested in BytesWritable
> -------------------------------------------------
>
> Key: HADOOP-310
> URL: https://issues.apache.org/jira/browse/HADOOP-310
> Project: Hadoop Common
> Issue Type: Improvement
> Components: io
> Reporter: p sutter
> Assignee: Owen O'Malley
> Priority: Minor
> Attachments: bytes-writable-zero-copy-interface-0.patch,
> bytes-writable-zero-copy-interface-1.patch
>
>
> It would be grand if BytesWritable.java had an additional constructor as
> below. This allows me to use the BytesWritable class without doing a buffer
> copy, since we have a less-than-fully-utilized byte array holding our key.
> Thanks!
>
> /**
> * Create a BytesWritable using the byte array as the initial value.
> * @param bytes This array becomes the backing storage for the object.
> */
> public BytesWritable(byte[] bytes, int size) {
> this.bytes = bytes;
> this.size = size;
> }
>
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
