Hi Martin, On Fri, 2008-12-19 at 10:37 -0800, Martin Buchholz wrote: > On Fri, Dec 19, 2008 at 09:53, Mark Wielaard <[email protected]> wrote: > > Still, if they were around I would indeed take a look and compare stuff > > a bit with my local builds if I had any strange test failures for > > example. > > > > Fedora 9 binaries would be fine for me, I am on Fedora 10 already :) > > Although my servers are a mix of Debian stable and CentOS 5. > > > >> > But how would be tag these builds as "official"? > >> > >> I'm not sure I understand the question. > >> Here "official" means there's been some release engineering love > >> and an expectation of a minimal level of quality. > > Let me expand on that a little. > Every "build" coming from Sun has already had > a significant amount of testing applied to it. > For comparative testing, I would like to try the very same binaries > that Sun must have already created.
Sure. But you need something that is completely automated and completely reproducible by the rest of the community. For example in icedtea we integrated all the tests in such a way that a simple make && make check runs them. Producing binary artifacts only makes sense really if you can do it methodically, otherwise you risk publishing things that depend on some individual's setup. That also means having enough capacity to do it on an ongoing basis. I'll see if we can finally expand builder.classpath.org to provide something like this over the next weeks, or that we would need to throw more hardware at it (which I think we will need seeing that we are already using the servers mostly at their capacity and having a build-bot for openjdk/icedtea will be pushing it a bit I am afraid). Another issue is that we aren't currently at zero-fail (ignoring the non-automated tests). It would be nice make this a requirement. Cheers, Mark
