Michael Scherer a écrit :
Le samedi 18 juin 2011 à 03:38 -0400, andre999 a écrit :
Michael Scherer a écrit :
Proposal 1:
6 months release cycle -> 12 months life cycle
( Fedora, Ubuntu, Mandriva< 2010.1&& Mandriva != 2006.0 )
Proposal 2:
9 months release cycle -> 18 months life cycle
( ~ opensuse and the one we used for Mageia 1 )
Proposal 3:
12 months release cycle -> 24 months life cycle
( Mandriva> 2010.1 )
First, suggest an amended freeze process (idea from recent report of another
project)
you can say the name of the project, even if I suspect it to be Fedora.
I suspected that it might have been Fedora, if it wasn't a summary of the new mozilla process, but
I couldn't remember. Just the concept intrigued me. Which I reflected on for a few weeks.
Instead of a freeze on cauldron until everything is ready for the release, we do
1) short freeze on cauldron
2) copy cauldron to pre-release branch, which remains frozen until release
3) immediately unfreeze cauldron.
- we avoid blocking cauldron, while leaving pre-release frozen for bug fixes.
- updates can continue on cauldron. Bugfixes can be applied to newer versions,
if present in
cauldron, at the same time as corresponding bugfixes in pre-release.
- activities like translation can continue in cauldron, meaning less rush for
such updates.
- because cauldron is open to changes (virtually) all the time, they don't have
to be put off and
perhaps forgotten.
- the cauldron cycle is extented by the time of the pre-release freeze. e.g.
In a release cycle of
6 months and a pre-release freeze of 1 month, the cauldron cycle would be 7
months.
This allows more time to iron out the pre-release bugs and more time for
cauldron.
- with the longer pre-release freeze, it may be appropriate to modify somewhat
the policy on what
is accepted during freeze. (Certain more recent packages or translations, for
example.)
- note that we would still have to monitor cauldron to avoid freezing partially
implemented complex
changes, such as a major update of kde or gnome or perl, etc. But we have to
do that now, anyway.
So you suggest that in order to help packagers focusing on bug fixing,
that we have them take care of cauldron and the bugfixes for the stable
release ( ie, twice more the load ).
I wouldn't quite put it that way ...
Proposal 1 :
---------------
My personal preference
Pros:
- better hardware support
- up to date versions / upstream projects (must have for developers)
- coincides with kde/gnome releases
- amended freeze process (outlined above) would lengthen both pre-release
freeze time and cauldron
development time.
A 1-month pre-release freeze would add 1 month to cauldron development time.
This would tend to alleviate the rush of the 6-month release cycle.
Let's do some math, shall we ?
great :)
If people work the same amount of time, with work divided on 2 products,
they must share their time, and usually work less than if they focused
only on one product, unless there is twice the ressources. But I doubt
this will happen for us, so let's assume that ressources are fixed.
That was my assumption : resources fixed in terms of time spent.
And why would that divide a contributor's focus more than now ? They would
just have a choice.
Now during the freeze, someone that wants to contribute to cauldron, but can't or chooses not to
contribute to pre-release bugfix, is not contributing.
So in practice, we risk to have more time contributed during the freeze.
Let say :
- the freeze period is Y weeks,
- the time between 2 release is X weeks,
- people divide their time evenly on both products.
That wasn't assumed. Rather that as much time would be spent on bug fixes, etc. in pre-release.
But having a longer freeze period would likely result in better quality, and certainly less rush.
That's a simplification, but I will come back on that later. Let's also
count the time spent as the metrics for the work, even if man/month is a
wrong unit in software development ( but that's a good enough
approximation for our case, given the highly distributed and
decentralized nature of the work of releasing a distribution ).
So when there is the freeze ( at release(n) time - Y weeks ), we will
have Y weeks of work done on both products ( next release, and cauldron
), so Y/2 weeks on each. We have X -Y weeks once the release(n) is out
( before the next freeze for release(n+1) ), and then again Y/2 weeks.
So for the release (n+1), we spend :
Y/2 + X - Y + Y/2
= 2 * Y/2 - Y + X
= Y - Y + X
= X
So that give X weeks of work. Fascinating, isn't it ?
Not really. Being my basic assumption :)
Now, of course, we can say "what if people do not divide their work in
2 ?"
So let's call :
- F the time spent on bugfix during the freeze
- C the time spent on cauldron during the freeze
We can assume that :
C + F = Y
So the equation become :
C + ( X - Y ) + F
= C + F - Y + X
= X
So no matter how you divide the time, you still have the same amount of
time spent overall.
As I assumed :)
Now, the real important question is "can we really exchange work done as
part of C for work done as part of F".
And so "if I do regular packages updates on cauldron at the begining of
the cycle, does it count as bugfixing for the release in the end of the
cycle" ?
To me, the answer is clearly no. If it was somethig we could exchange,
we would not have to make a freeze in the first place to make sure that
only bugfixes are uploaded in cauldron.
So the only way to maximize the time spent on bugfixes is to have F = Y,
and so C = 0. Ie, do like we do now.
I really don't follow this line of reasoning.
The focus on bug fixes starts with the freeze. So a longer freeze would give more time to focus on
bug fixes.
Sure, there are updates and bug fixes in cauldron before the freeze, as a normal part of any
development process. (Even in the non-libre world.)
And unless you show that letting people work on cauldron will bring more
ressources , and more than the one we will lose du to people who do not
want to work on bugfixes and the release, I doubt this will change.
Ok. Obviously I need to clarify my point of view.
Firstly, my assumption was that at least as much time would be spent on bug fixing during the
longer freeze, but being less rushed, would tend to produce better quality results. (And less
aggravation for ennael) (That is certainly how it works in the non-libre world.)
I don't see how having the choice between contributing to pre-release or cauldron during the freeze
will lead to us loosing _any_ contributors.
As well, since cauldron would be out of freeze virtually all the time, there would be (virtually)
no period where contributions to cauldron are blocked.
Packager time is not an ubiquitous resource. Some packagers are perl experts, other python, etc.
Each packager is more familiar with some packages than others. Some packagers are excellent
developers; others are challenged by basic scripts. There is a wide range of skills and interests.
If during freeze, a packager has a choice between attempting to help with a bugfix in pre-release
for a package with which s/he is not familiar, or contributing to cauldron for something with which
s/he is familiar, it would be evidently more efficient to contribute to cauldron.
Similarly, if a packager contributes a bug fix to pre-release, and a newer package already exists
in cauldron for which the same bug fix must be applied, it is more efficient to apply the same
patch right away, than to wait until freeze is over. (Personnally I've encountered this sort of
situation with similar but different software many times. Any experienced programmer should
understand this point.)
So there are a lot of (admittedly small) synergies which should lead to packager time being more
efficiently used.
Not counting the likelyhood that some packagers would contribute somewhat more time, being able to
contribute to cauldron during freeze.
The major benefit in my mind is the longer freeze period.
How much this would help, I don't know. But I think it is worth a try.
(Even if we end up going for a 9-month release cycle, instead of my preferred 6
months.)
- short life cycle
would be alleviated by having periodic long term support releases (lasting at
least 2 years).
As said before, the support is decided in another discussion, and depend
more on the ressources we will have than anything else.
ok :)
Proposal 2
----------------
Cons:
- not synchronized with gnome or others that use a 6 month cycle
- potentially release when there isn't much activity (like during Holidays)
- release would not be the same month every year
e.g. 2011 june ; 2012 mar ; 2012 dec ; 2013 sep ; 2014 june ...
so users won't know when to expect a release
I do not expect our users to be farm animals, so they can perfectly cope
with lack of seasonal hints regarding release cycle.
We may not be farm animals, but I suspect that we are still creatures of habit
:)
In any case, it seems to me that the bigger liability would be being out of sync with the 6-month
release cycle of kde, gnome, as well as many other distros.
another 2 cents :)
--
André
