15 Sep
2016
15 Sep
'16
1:39 p.m.
On 15 Sep 2016, at 13:23, Ecaterina Moraru (Valica)
<valicac(a)gmail.com> wrote:
On Thu, Sep 15, 2016 at 1:15 PM, Vincent Massol <vincent(a)massol.net> wrote:
By butchering I meant having 8.4, 8.5, 8.6, 8.7, etc. and having to explain
the users that for the 8.x cycle the versioning strategy changed meanwhile
and 8.7 is equivalent to 7.4 final. But the thing is that I forgot that
8.4+ were shorter releases, so I guess you are right: it doesn't change a
thing. By the end of the year we should have just 8.4 and 8.5.
Actually we’ve decided a while back to no longer have N.5 releases because we don’t have
enough time at the end and we usually have to carry the remaining stabilization/polishing
work in the next year.
So the idea is to have 8.4 till around mid November (i.e. 1 month) and then to do bug fix
releases of 8.4 every 2 weeks, i.e. 8.4.1, 8.4.2, etc.
And then on Jan 2nd, start 9.0.
Thanks
-Vincent
On 15 Sep 2016, at 11:40, Ecaterina Moraru
(Valica) <valicac(a)gmail.com>
wrote:
+0 I'm not thrilled since:
- it will put pressure on the RMs (there are several steps additional
done
for final releases + the ci needs to be shiny
perfect +
Could you be more specific? I don’t see any change for the RMs.
increases the
number of jobs and branches for versions + branches for partial
features);
Actually, as I’ve mentioned in my mail, it should decrease slightly the
number of branches/jobs since we’ll have less bugfix releases for stable
versions. Or keep it similar. But I don’t see an increase.
- in theory all the releases should be tested
fully (before we had a
buffer
to catch bugs between Ms + more time for the
community to test the
releases) and
Yes that’s part of the rational I put: increase the testers and use more
the community to test.
Our scope has been increasing over the years (more features, more
extensions). It’s important to use more and more the community to help us.
- the roadmap items planning might need more
issues/bureaucracy.
Possibly a bit. But the way I view this is that we’ll continue to do
roadmaps every 3 months or so for the coming 3 releases and do the steering
every month in case we want to change it slightly.
Also users might be confused about the
'quality'/frequency of releases
and
they might adopt strategies like installing
even/odd release version or
just bugfix releases. So I'm not sure this will fix use case in terms of
users feedback.
It does. If you’re an existing user and you’re upgrading, say every 6
months, it won’t change anything ofc.
But if you’re a new user, you’ll be able to use the latest released
version. And report feedback, bugs, ideas, etc. You won’t start on a 3
months old release.
It's just a feeling I have and I'm a bit
worried since our community is
small, but maybe I'm just resistant to change. I'm willing to try though.
You do look a bit resistant to change :)
But that’s fine. It’s good to have someone playing the devil’s advocate.
However, I haven’t seen any real cause for being worried so far.
Honestly it's almost is a non event and doesn’t change much IMO. It just
means removing the suffix “milestone” from our releases ;)
I would prefer 10 releases N.0 to N.9 (just for
'esthetic' reasons).
Also I
would have prefer to start this new
version/release scheme for the 9.x
cycle, instead of butchering the 8.4+ stabilization releases (for
'consistency' reasons).
Could you explain what you mean by butchering? Could you also explain what
is not clear in:
“[….]Since that version is already supposed to be a stabilization release
(and thus a short 1 month release) it’s not going to change anything.[…]"
Thanks for the explanations,
Caty
>
> So to remind you, the last stabilization release is already meant to be 1
> month (we’ve been doing this for some time already, even doing it for the
> last 2 releases in the past when we had a N.5 release), so as I said, it
> doesn’t change anything.
>
> So in practice I could have omitted this part since it’s the same as
> saying we start only in 9.0 but I wanted to put us in the spirit of the new
> strategy ASAP.
>
> Thanks
> -Vincent
>
>> Thanks,
>> Caty
>>
>>
>> On Thu, Sep 15, 2016 at 12:18 PM, Vincent Massol <vincent(a)massol.net>
> wrote:
>>
>>> Hi devs,
>>>
>>> Executive summary:
>>> * Replace our 3 month release cycle by a 1 month release cycle.
>>>
>>> Needs and advantages:
>>> * Be able to get our changes faster to our users. Importantly, bugfixes
>>> are released faster to users. Note: it’s not because we release faster
> that
>>> our users have to upgrade as fast. They can skip some versions if they
>>> want/need.
>>> * Be able to get more feedback more quickly from users. Right now, most
>>> (if not all) of our users are testing only final versions. They’re not
>>> testing milestones or RCs and thus we usually only know about problems
>>> after the final has been released and we incorporate the change in the
> next
>>> one (to be released 3 months later) or we have to do a bugfix release.
>>> * Get closer to cloud needs. Nowadays, could offerings happen more and
>>> more and operating a cloud means bringing improvements and fixes as
> fast as
>>> possible. Some software in the cloud are even updated/patched several
> times
>>> per day. We’re not there yet but we’re trying to get closer by reducing
>>> from 3 months to 1 month
>>> * This also means more marketing for the xwiki project since other site
>>> relay the new whenever a final version is released
>>>
>>> Proposal Details:
>>> * 1 month split in: 3 weeks to release a RC + 1 week to release the
> final
>>> version. It’s very important to keep the RC as a meeting point and
> ensuring
>>> all is going to be ready on time for the final release (jiras are
> closed or
>>> moved to the next release, CI is passing, documentation and RN are
> ready,
>>> etc).
>>> * Split large features into smaller chunks. It doesn’t matter if some
> code
>>> is released but unused for example (provided the build is passing). For
>>> larger refactoring that absolutely cannot be split into 3 weeks chunks
> (I’m
>>> not sure that really exists) then we can use branches (and create a CI
> job
>>> to ensure integration).
>>> * Less need for bugfix releases of stable versions. For LTS it’s still
>>> required though.
>>> * Note that this 1 month release strategy will not generate more
> releases
>>> (and thus more work) over the year since we’re already releasing every
> 2-3
>>> weeks ATM (milestones, RCs, et)
>>> * Version Naming: from N.0 to N.10 where N is the cycle name. The reason
>>> for 11 releases and not 12 is to account for slippages + potential
> bugfix
>>> release at the end of the cycle for stabilization + holidays. We might
> even
>>> be able to do only 10 releases and not 11 but I’m suggesting we try
> with 11
>>> for now.
>>>
>>> When:
>>> * I’m proposing to start this process with the 8.4 release (starting on
>>> the 10th of October). Since that version is already supposed to be a
>>> stabilization release (and thus a short 1 month release) it’s not going
> to
>>> change anything. We should also do bugfix releases after 8.4 is
> releases:
>>> 8.4.1, 8.4.2, etc till the end of the year
>>> * Then starting 9.0 we really start doing 1 month releases.
>>>
>>> WDYT?
>>>
>>> Here’s my +1 to try this.
>>>
>>> Thanks
>>> -Vincent