On Sep 25, 2009, at 10:36 AM, Philip Zeyliger wrote:
I did a quick grep, and--perhaps I'm missing something--no
interfaces are
currently marked as InterfaceStability.Stable. FileContext is the
only
interface that uses the org.apache.hadoop.classification package.
(Well, QueueManagerTestUtils has it commented out...)
Do you have a proposed list of classes to be marked
@InterfaceStability.Stable?
Yes, please see
https://issues.apache.org/jira/browse/HADOOP-5073?focusedCommentId=12664679&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel
#action_12664679
There is a proposal at the bottom of that comment.
(there have been a few changes proposed in that jira but that proposed
tagging is mostly intact. I will update it shortly.)
I think your question is whether we are committed to keeping the
APIs marked
as stable in the course of 0.22 development to remain so through
1.0. (You
wrote 0.21.0 below, but you probably meant 0.22.0.) Yes, +1.
What's our commitment to methods that are @Deprecated within 0.22's
@InterfaceStability.Stable classes? (I don't know that there are
any, but I
bet there are or will be.)
-- Philip
On Fri, Sep 25, 2009 at 10:16 AM, Owen O'Malley <omal...@apache.org>
wrote:
> We are getting closer to being able to release a Common/HDFS/
MapReduce 1.0.
> I'd hope that we'll get the last set of things in to 0.22 that
mean that it
> would be labelled 1.0. Toward that end, I'd like to start locking
down the
> APIs that we've marked as public stable. What that would mean is
that any
> interface that is tagged with the @InterfaceStability.Stable and
> @InterfaceAudience.Public in the 0.21.0 release should not have
any changes
> committed that require a recompilation of client code. This will
provide a
> stable basis for our users' applications and reduce the costs of
upgrades.
>
> Clearly, I'm +1.
>
> -- Owen
>
