Re: Linux 2.6.35-rc2

From: Linus Torvalds
Date: Mon Jun 07 2010 - 14:51:45 EST




On Mon, 7 Jun 2010, Konrad Rzeszutek Wilk wrote:
>
> Ah, <sigh> I seem to have a knack for continously missing the merge
> windows. Is it fair to state that:
>
> 1). Once you send an email saying that Linux 2.6.35 is released, the
> merge window has opened.
> 2). Then the the maintainers have two weeks to send you git pull
> request for new features.
> 3). After the two weeks, it is rc1, and only bug-fixes are accepted?

That's the rough outline yes. I say "rough", because:

- You should have had the stuff ready _before_ I even opened the merge
window. So the "two weeks" is literally the time for you to send the
email, not the time for you to do the development work.

(That sounds crazy, and reality is a bit more complicated: there are
people who know they have conflicts from sitting in -next, so strictly
speaking, the two weeks is so that we can sort things like that out,
and also so that I can stagger out the pulls a bit so that we have
nightly snapshots with testing etc that doesn't contain _everything_ in
one big lump etc etc).

- You can (and some people do) anticipate the merge window, and even send
the pull request to me before it opens. HOWEVER, I would suggest you do
this only if you know you're going to be on vacation during the merge
window or something like that, and perhaps ask somebody else to remind
me once the merge window opened, because I get too much email, so while
I try to make sure to process pull requests that happened before, I'm
also not organized enough to at all guarantee it.

- I tend to be lenient. In fact, I tend to be _too_ lenient and end up
pulling, even if I may curse at developers when they send me pull
requests (I really am not very polite even at the best of times). This
release around I'm a bit more hard-nosed about the rules than I usually
am, so now I still curse, but then I actually say "no, I won't pull,
because I'm holding you to the rules".

Normally, I do not believe in black-and-white rules, and many things
are meant to be "those are the rules, but we'll all use our own brains
and judgement". And that's still true, but my judgement this release
around is a lot stricter, just because I'll be off on vacation in the
middle of the release window and thus needed to be stricter.

So the process I generally encourage sub-maintainers to have is roughly

- if you are a git maintainer, use two brances: one called "stable", one
called "devel" (and no, I don't care about the names. Think of them as
just illustrations)

- never _ever_ ask me to pull from the "devel" branch.

- when you see my release announcements (or you just predict it, because
they are pretty predictable: I tend to start saying at around -rc6/7
whether I think that's the last -rc or not etc), you make sure I've
pulled your old "stable" branch (ie that all the bug-fixes from it are
in the mainline tree), and then you delete it and rename your "devel"
tree (if you think it's ready to be merged) to be "stable".

- You are now ready to ask me to pull your new "stable" branch.
Obviously, it should be in a good enough shape that you really think
the name "stable" is ok. If that's not the case, you shouldn't have
done the renaming, and you shouldn't plan on getting it merged.

(Yes, the post-merge-window -rc phase is for "fix up problems", but it
should be to fix up problems that _other_ people notice, not problems
that you already knew about when you asked me to pull. See? If you know
about problems, you're not ready for that merge window)

- Once you've renamed your "devel" tree as "stable", you can start a
_new_ "devel" branch. So you can still continue to develop during the
merge window, but you do it in the branch that you know you won't ask
me to pull (until the next merge window, when the next big renaming
comes along)

- fixes obviously go into the "stable" branch, and you can ask me to pull
those at any time (although I generally frown on daily pull requests).

Now, the above is just a suggestion. Use it as a template for whatever
workflow works for you. Non-git users can also use it as a template for
how they should work, even if the details of exactly how they handle the
stable/devel branches would be different.

And git users may well have different flows. Some people (and I really
appreciate it, btw) have _way_ more than one "devel" branch. They name it
by the actual _feature_ they are developing, and then when they ask me to
pull they will have merged all their feature branches together (or, if
they are big features and you're not sure about them all, you can
certainly ask me to pull them one by one too!).

So really, take all of the above as a "this is a guideline". It's not set
in stone, it's not how you _must_ work, it's not a requirement. But it is
an example of how things can be done, and it's an example of how you can
do development all the time (so you don't have to think of the "merge
window" as some big break in development at all!).

So you absolutely can make whatever changes to the flow that suit the way
you work, or that just make more sense for your subsystem. Part of the
Linux kernel development process is very much a certain amount of
flexibility - with a pretty rough framework to make it possible at all to
make sane releases.

Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/