I'm +1 as long as the goal is still to eventually transition to a
single, supported API over upcoming releases. We can go over 1:1
guarantees over a JIRA perhaps, to evaluate if any library work still
needs to be done.

On Wed, Nov 23, 2011 at 3:30 AM, Arun C Murthy <[email protected]> wrote:
>
> On Nov 22, 2011, at 2:34 AM, Steve Loughran wrote:
>
>> On 22/11/11 02:18, Harsh J wrote:
>>> Are gonna be maintaining both APIs forever?
>>
>>
>> That's a key point. Either there are two apis forever or, some time after 
>> the new APIs are both a proper superset of the old APIs (are we there yet?) 
>> and the majority of downstream project have migrated to the new APIs, then 
>> the old ones can be removed.
>>
>> And how do you persuade developers to move off the old APIs except through 
>> deprecation warnings?
>>
>
> It isn't clear, yet, if the new apis are a proper superset of the old APIs 
> i.e. only in 0.21 did they come close to completion (added new Cluster apis 
> etc.).
>
> As a result, with very little exposure, it's hard to mark them as 'stable' 
> yet. Hence, my proposal to keep the old apis as stable and new one as 
> evolving for hadoop-0.23.
>
> Furthermore, my personal preference would be to further tweak the new apis 
> along the lines of https://issues.apache.org/jira/browse/MAPREDUCE-1183 
> before we declare them stable (that has real benefits for Pig/Hive etc.), but 
> that is some way away.
>
> Arun



-- 
Harsh J

Reply via email to