It feels like we approaching a consensus on that if we include MRUNIT-138, which is backwards incompatible but an improved user API, we should bump the major version. Assuming MRUNIT-138 is included, is there anyone that would -1 a release with the 1.0 designation?
Brock On Thu, Sep 13, 2012 at 11:42 AM, Brock Noland <br...@cloudera.com> wrote: > I agree that the changes in this release are not nearly as substantial > as handling the Tool interface but I do they are major improvements. > For example, we now allow users to specify many input key/values and > have distributed cache support. For quick reference: > http://s.apache.org/NQY > > Brock > > > On Thu, Sep 13, 2012 at 8:16 AM, Jim Donofrio <donofrio...@gmail.com> wrote: >> Yes graduation has nothing to do with the quality or state of the code, >> graduation is all about community and should not influence a release number. >> >> I agree that 1.0 would signal a breaking change but 1.0 should also signal >> major improvements, api changes, new features. I dont think this release >> contains any drastic features. I think we should continue in the 0.* >> versions for awhile until we add major new features such as Tool support. At >> that time you can change the package names to org.apache.mrunit and go to >> 1.0. I would rather not become like Firefox or Chrome and do major release >> number changes on every release. >> >> >> On 09/13/2012 06:31 AM, Jarek Jarcec Cecho wrote: >>> >>> I do have similar reasoning here: >>> >>> 1.0 - in case that we're breaking backward compatibility >>> 0.10 - in case that we're not breaking backward compatibility >>> >>> I personally do not see graduation of the project important enough for the >>> version to jump to next major. We've recently graduated sqoop and flume and >>> remained on the same major version without any issues. >>> >>> But I'll support next reason no matter the final version. >>> >>> Jarcec >>> >>> On Wed, Sep 12, 2012 at 11:41:08PM +0200, Bertrand Dechoux wrote: >>>> >>>> And I would say the same in a reverse way. >>>> If we do a 1.0 release, all required incompatible changes should be done >>>> so >>>> that there would be no need to drag unneeded deprecated stuff from the >>>> 1.0 >>>> up to the 2.0. >>>> >>>> For me, the question is whether we should break compatibility for the >>>> next >>>> release. If yes, then break all which is necessary for a clean future. If >>>> not, then assure full compatibility. If yes, it should be 1.0. If not, it >>>> should be 0.10. >>>> >>>> The following question is then : if we keep compatibility what will the >>>> next release ship with? Is a release worth the new features/bug fixes? On >>>> that point, I am not knowledgeable enough to answer. I would accept the >>>> decisions of the more 'ancient' devs. But it should indeed be discussed. >>>> >>>> Regards, >>>> >>>> Bertrand >>>> >>>> >>>> >>>> >>>> >>>> On Wed, Sep 12, 2012 at 9:05 PM, Wei, Jianbin <jianb...@paypal.com> >>>> wrote: >>>> >>>>> If it is an incompatible change in non-trivial way, I would strong in >>>>> favor a 1.0 release. >>>>> >>>>> Regards, >>>>> >>>>> -- Jianbin >>>>> >>>>> On Sep 12, 2012, at 10:01 AM, Brock Noland wrote: >>>>> >>>>> OK, we have the following viewpoints with supporting reasons: >>>>> >>>>> 0.10 - supported by a number of people (reasons: none given, 1.0 >>>>> should be used for Tool interface support) >>>>> 1.0 - supported by a number of people (reasons: none given, recent >>>>> graduation, due to the incompatible change) >>>>> >>>>> I tilt towards the 1.0 release due to the incompatible changes but I >>>>> am not strongly committed to that viewpoint. I am strongly committed >>>>> to a release whatever the number! :) It would seem easy enough to vote >>>>> on the matter but I think votes can become divisive. I have seen that >>>>> in the Hadoop community when voting is used to resolve issues it ends >>>>> up much like the state of US politics. As such, I'd prefer to settle >>>>> this via discussion. >>>>> >>>>> We have all stated our preferences but not our convictions. Is there >>>>> anyone who strongly in favor of / opposed to a 0.10 release? Is there >>>>> anyone who is strongly in favor of / opposed to a 1.0 release? If so >>>>> please state your reasoning. >>>>> >>>>> Brock >>>>> >>>>> On Fri, Sep 7, 2012 at 11:50 AM, Wei, Jianbin >>>>> <jianb...@paypal.com<mailto: >>>>> jianb...@paypal.com>> wrote: >>>>> Agree with Dave that when it becomes incompatible, the major version >>>>> number should be increased. Major changes also warrant a major number >>>>> change. >>>>> >>>>> Regards, >>>>> >>>>> -- Jianbin >>>>> >>>>> On Sep 7, 2012, at 8:06 AM, Brock Noland wrote: >>>>> >>>>> As I understand it, if we implement >>>>> https://issues.apache.org/jira/browse/MRUNIT-138 as described in the >>>>> JIRA. That is, all the drivers keep state of the inputs, we can >>>>> undeprecate the methods depcrecated in >>>>> https://issues.apache.org/jira/browse/MRUNIT-64? >>>>> >>>>> Brock >>>>> >>>>> On Fri, Sep 7, 2012 at 7:58 AM, Jim Donofrio <donofrio...@gmail.com >>>>> <mailto:donofrio...@gmail.com>> wrote: >>>>> I think we need to keep those deprecated methods around for awhile, no >>>>> reason to anger users. >>>>> >>>>> >>>>> On 09/07/2012 08:35 AM, Bertrand Dechoux wrote: >>>>> >>>>> Then the question is about when/if the compatibility should be broken. >>>>> >>>>> https://issues.apache.org/jira/browse/MRUNIT-139 would be quite easy >>>>> without the history of MRUnit and the @Deprecated.... >>>>> >>>>> On Fri, Sep 7, 2012 at 2:19 PM, Dave Beech <d...@paraliatech.com<mailto: >>>>> d...@paraliatech.com>> wrote: >>>>> >>>>> I think this depends on what we decide to do about MRUNIT-138. We were >>>>> discussing an incompatible change, and if we do decide to do that I >>>>> think >>>>> the version number should increase to 1.0.0 to reflect this (and also >>>>> the >>>>> fact that this is the first version since graduation). >>>>> >>>>> If we later go ahead with the API rewrite (MRUNIT-69), this could form >>>>> MRUnit 2.0.0! Would line up nicely with Hadoop's own numbering strategy >>>>> ;) >>>>> >>>>> On 7 September 2012 07:54, James Kinley <kin...@cloudera.com<mailto: >>>>> kin...@cloudera.com>> wrote: >>>>> >>>>> +1 for the 1.0.0 release. I think it's a good idea to increase the major >>>>> version number considering the recent graduation and the included >>>>> >>>>> changes. >>>>> >>>>> On 7 Sep 2012, at 07:29, "Wei, Jianbin" <jianb...@paypal.com<mailto: >>>>> jianb...@paypal.com>> wrote: >>>>> >>>>> My bad, 0.9.0 --> 0.10.0 is also version increase. My eyes are not >>>>> >>>>> used >>>>> >>>>> to have a 2 digits minor version yet. However, I still prefer a >>>>> >>>>> one-digit >>>>> >>>>> minor version as most software do that in practice. >>>>> >>>>> Regards, >>>>> >>>>> -- Jianbin >>>>> >>>>> On Sep 6, 2012, at 10:41 PM, Bertrand Dechoux wrote: >>>>> >>>>> I am not sure to understand "It is not good to backtracking version.". >>>>> Does it mean that the version after graduating should show the 'step'? >>>>> Is that a common way to do it? >>>>> >>>>> Not taking into account the graduation, I would also favor the "0.10.0" >>>>> instead of "1.0.0". >>>>> >>>>> Regards >>>>> >>>>> Bertrand >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> -- >>>>> Apache MRUnit - Unit testing MapReduce - >>>>> http://incubator.apache.org/mrunit/ >>>>> >>>>> >>>>> >>>>> >>>>> -- >>>>> Apache MRUnit - Unit testing MapReduce - >>>>> http://incubator.apache.org/mrunit/ >>>>> >>>>> >>>> >>>> -- >>>> Bertrand Dechoux >> >> > > > > -- > Apache MRUnit - Unit testing MapReduce - http://incubator.apache.org/mrunit/ -- Apache MRUnit - Unit testing MapReduce - http://incubator.apache.org/mrunit/
