I would not want to make moves that require people to be on a version later than 0.20.2 at this point. "Most" users have access to 0.20.2 or a work-alike (roughly, CDH3, Amazon EMR); few have access to 1.0 or 0.22+. Mahout should probably be a little behind the cutting edge in this regard.
(That said I don't know of a reason it wouldn't all work with 1.0, unless it's some really tiny detail or API change.) I think branching is unnecessarily complex right now just for the sake of one dependency, albeit a major one. If anyone's itching to do some work here -- please look at replacing the deprecated .mapred. code. This has been sitting for nearly 2 years and so anything that nobody including the original author cares to update, I'd honestly say we consider dropping soon, having put it out in 0.6. On Fri, Jan 13, 2012 at 7:31 PM, Dmitriy Lyubimov <[email protected]> wrote: > Alternatively, we can release 2 branches. happens to other projects > all the time. the benefit here is to put a tag on 1.0-dependent code > that we otherwise think is stable instead of keep sending people to > trunk which may destabilize over the time. > > -d > > On Fri, Jan 13, 2012 at 11:28 AM, Dmitriy Lyubimov <[email protected]> wrote: >> I think main problem is to be able to retest _everything_ with >> distributed MR and 1.0 by release time. While we can get local tests >> running probably, hadoop implementation of local mode vs. distributed >> mode is vastly different, and run and revalidate results in fully >> distributed mode for everything by release time is probably going to >> be hard to ensure. >> >> On Fri, Jan 13, 2012 at 11:23 AM, Grant Ingersoll <[email protected]> >> wrote: >>> >>> On Jan 13, 2012, at 12:00 PM, Josh Patterson wrote: >>> >>>> I agree with Ted here, 1.0 is more of a marketing thing than anything >>>> imho. not a huge technical change. >>> >>> That's why I thought we might want it in 0.6. Mahout 0.6, built on Hadoop >>> 1.0 sounds better than Mahout 0.6 built on Hadoop 0.20.2903.57.19.10 or >>> whatever it was ;-) >>> >>> But yeah, I'm fine with the approach outlined. >>> >>> -G >>> >>>> >>>> On Thu, Jan 12, 2012 at 9:17 PM, Ted Dunning <[email protected]> wrote: >>>>> D has a good idea. >>>>> >>>>> 1.0 should be very nearly identical to what we have now so this isn't a >>>>> big >>>>> change, but it would be very nice to provide just a bit of consistency >>>>> nevertheless. >>>>> >>>>> On Thu, Jan 12, 2012 at 8:27 PM, Dmitriy Lyubimov <[email protected]> >>>>> wrote: >>>>> >>>>>> No, haven't tried yet. >>>>>> >>>>>> i would release it with current dependencies and switch trunk to 1.0. >>>>>> This way we will have time to work out new kinks, and people who want >>>>>> previous release would use 0.6 and people who would like to try 1.0 >>>>>> can use trunk. >>>>>> >>>>>> -d >>>>>> >>>>>> On Thu, Jan 12, 2012 at 5:11 PM, Grant Ingersoll <[email protected]> >>>>>> wrote: >>>>>>> Should we be switching over to 1.0 for this release or wait? Has anyone >>>>>> tried it out? >>>>>>> >>>>>>> -Grant >>>>>> >>>> >>>> >>>> >>>> -- >>>> Twitter: @jpatanooga >>>> Solution Architect @ Cloudera >>>> hadoop: http://www.cloudera.com >>> >>> -------------------------------------------- >>> Grant Ingersoll >>> http://www.lucidimagination.com >>> >>> >>>
