On 21/08/12 22:00, Ryan Harper wrote:
* Livnat Peer <lpeer@redhat.com> [2012-08-21 13:23]:
On 21/08/12 17:01, Ryan Harper wrote:
* Dave Neary <dneary@redhat.com> [2012-08-21 08:42]:
I am not sure we must have new features in each release, a release of bug fixes seems also reasonable to me. Why not keep it only time-based release regardless of commitments for new features for the release.
I like giving people good reasons to upgrade, but also good reasons to install the current version - and in terms of communication, if we say that 3.2 will be "3.1, with lots of bug fixes", and that it will be along in 3 months, why would anyone install 3.1? We've just said it's a buggy release that will soon be obsoleted anyway.
IMHO, it's better to say "here's what 3.1 does well, here's what 3.2 will be able to do that 3.1 doesn't". I'm not suggesting a revolution with every release, but one thing which is identifiable as "new in 3.2" doesn't seem like a lot to ask.
Definitely agree with this approach. We always want something new for the next release.
I agree we want something new, the question is what is the release criteria. If I understand the above suggestion correctly If there are not enough new features we won't release in 3 months? I think this is a mistake because there are hundreds of bug fixes pushed into the repository and releasing a more stable version IMO has great value.
This is what a stable release stream is for. QEMU maintains a stable release until the next version is available. We could produce a 3.1.1 release and keep that going until 3.2 comes out.
And if we don't have new features in 3 months, then it's probably too short of a release cycle. The last thing we want to do is introduce *more* churn into engine deployments. If we have a stable release, this would make a longer cycle, like 6 months, more reasonable.
I'm perfectly fine with releasing in November a 3.1.1 as a stable release. As long as we don't postpone releasing 'something' further than November. The only issue that comes to mind when we start discussing z-stream is the churn we'll have (as developers): - We'll need to cherry pick instead of rebasing, which means sending patches twice, testing it twice, rewriting parts if needed etc. It would have been ok if we didn't have to cherry pick a few dozens patches in networking alone. - We'll need to take care of upgrades more carefully, which always makes our work more interesting but also more complicated (do we require upgrading to 3.1.z before upgrading to 3.2? testing the upgrade process etc.) I think that going forward we'll do better if we can create the next z-stream branch as soon as we release so we can cherry pick as we go, now we have a gap of about 8 weeks (? - not sure actually) since last rebase, that's a lot of work. I suggest that the release in November will still be 3.2 and hopefully we can get enough features to make it attractive, and then we'll create the z-stream branch, so we can release 3.2.z as needed (instead of doing 3.3 with not enough meaningful content).
For example in Networking we fixed one feature and will probably add one small feature by November, but I know that networking in 3.1 release is very buggy while if you take latest from upstream it is dramatically better, regardless of the features there is a value for releasing in November.
definitely material for a stable release.