-Re: why hadoop choose maven as its build tool?
Konstantin Boudnik 2012-05-17, 23:10
+1 on that: let's not mix snapshot dependencies and release dependencies.
Latter are 100% deterministic
Former aren't by design.
Case is closed ;) You can apologize now, Sundar ;)
On Thu, May 17, 2012 at 06:15PM, Ted Dunning wrote:
> If you depend on released artifacts that you havent overridden then maven is deterministic.
> If you depend on snapshot artifacts then you get the latest snapshot which
> could change between builds. But you asked for that.
> Sent from my iPhone
> On May 17, 2012, at 5:23 PM, Jagane Sundar <[EMAIL PROTECTED]> wrote:
> > Hallelujah.
> > I thought I was the only one who felt that maven's non-deterministic result was just plain batty.
> > In this mavenized world, is it guaranteed that if I build twice using
> > maven on two different machines in two different locations, I will get the
> > same exact bits?
> > If this guarantee exists, then I apologize for polluting the Internet with
> > this posting.
> > If this is not true, then this is not a build system - it is a right royal cr** shoot system masquerading as a build system.
> > Jagane
> > -----Original Message-----
> > From: Steve Loughran [mailto:[EMAIL PROTECTED]]
> > Sent: Thursday, May 17, 2012 2:10 PM
> > To: [EMAIL PROTECTED]
> > Subject: Re: why hadoop choose maven as its build tool?
> > another reason: some people felt that the ant builds were excessively
> > deterministic. by including random downloads of -SNAPSHOT artifacts the
> > build has been made more enteratining
> > On 16 May 2012 21:40, Konstantin Boudnik <[EMAIL PROTECTED]> wrote:
> >> I would also mention a benefit of testing and inter-component integration
> >> standardization. This is, essentially, what BigTop has done by introducing
> >> and
> >> supporting open-source solution for creating and validating Hadoop stacks.
> >> Funnily enough, this simple concept was too hard to grok for some of the
> >> commercial vendors. The management there were literally insisting on doing
> >> _manual_ testing for their packages (at least at the time that when I was
> >> working for one of them and observing it:) But I guess this is a different
> >> story...
> > I seem to recall showing someone (todd?) at an apachecon a few years back
> > how my ant build would scp the just-built RPMs to a test VM, install them,
> > walk them through their lifecycle and then verify that they uninstall.