Message-ID: <20230713-irritant-rarity-5f7b424fe43e@spud>
Date:   Thu, 13 Jul 2023 16:39:52 +0100
From:   Conor Dooley <conor@...nel.org>
To:     Greg KH <gregkh@...uxfoundation.org>
Cc:     Thorsten Leemhuis <linux@...mhuis.info>, stable@...r.kernel.org,
        linux-doc@...r.kernel.org, linux-kernel@...r.kernel.org,
        Sasha Levin <sashal@...nel.org>,
        Jonathan Corbet <corbet@....net>
Subject: Re: [RFC PATCH v1 0/3] docs: stable-kernel-rules: add delayed
 backporting option and a few tweaks

On Thu, Jul 13, 2023 at 05:06:22PM +0200, Greg KH wrote:
> On Thu, Jul 13, 2023 at 10:48:14AM +0200, Thorsten Leemhuis wrote:
> > On 12.07.23 21:00, Greg KH wrote:
> > > On Wed, Jul 12, 2023 at 07:02:34PM +0200, Thorsten Leemhuis wrote:
> > >> On 12.07.23 17:16, Greg KH wrote:
> > > [...]
> > >>>>   .. warning::
> > >>>>      The branches in the -stable-rc tree are rebased each time a new -rc
> > >>>>      is released, as they are created by taking the latest release and
> > >>>>      applying the patches from the stable-queue on top.
> > >>>
> > >>> Yes, that is true, but they are also rebased sometimes in intermediate
> > >>> places, before a -rc is released, just to give CI systems a chance to
> > >>> test easier.
> > > [...]
> > >> Nevertheless makes me wonder: is that strategy wise in times when some
> > >> ordinary users and some distributions are building kernels straight from
> > >> git repos instead of tarballs? I'm one of those, as I distribute
> > >> stable-rc packages for Fedora here:
> > >> https://copr.fedorainfracloud.org/groups/g/kernel-vanilla/coprs/
> > > 
> > > As we keep the patches in quilt, not git, it's the best we can do.  The
> > > -rc releases are never a straight-line if we have to do multiple ones,
> > > we remove patches in the middle, add them at the end or beginning, and
> > > sometimes even change existing ones.
> > > 
> > > All of this is stuff that a linear history tool like git can't really
> > > model well, so we keep a quilt series of the patches in git for anyone
> > > that want to generate the tree themselves, and we provide the -rc git
> > > tree for those that don't want to generate it and can live with the
> > > constant rebasing.
> > 
> > /me first didn't want to reply, as this is not really important, but
> > then reconsidered; again, feel free to just ignore this
> > 
> > FWIW, I do not consider that rebasing to be problem at all; it are those
> > rebases "sometimes in intermediate places, before a -rc is released,
> > just to give CI systems a chance to test easier" make things this
> > slightly annoying bit harder when you want to distribute stable-rc
> > releases to users.
> > 
> > But as I said, I can fully understand why you do those as well. I just
> > with there was a way to reliably get a -rc release from git as well.
> > Simply tagging them when you do a -rc release would solve all that. Is
> > that maybe something that could be easily added to your -rc release scripts?
> 
> I can add a tag, but it would have to be a tag that can be rebased, and
> git doesn't like that very well :)

I figure the desired tagging behaviour is that you do it when the email
is sent out for a corresponding version & so the tag "should" not need to
be rebased?

> > /me looks at https://github.com/gregkh/gregkh-linux/tree/master/stable
> > but failed to find the -rc release script :-/
> 
> Hah, no github, it's at:
> 	https://git.kernel.org/pub/scm/linux/kernel/git/stable/stable-queue.git/tree/scripts/quilt-mail
> 
> But I don't think tags will help much.  I'll let anyone who actually
> runs a CI that uses this to speak up to see if it would before adding
> them.

I'm not sure that it is particularly valuable to the usual flow of
testing what is about to come down the tracks, at least in my simple
case where I trigger it based on the -rc emails or whenever something
else interesting happens, like a patch being dropped that breaks the
build.

I suppose it may be useful if an issue presents itself but disappears
when a backport is dropped from the queue & some developers are
interested in figuring out why the backport went awry?

Other than that, I'm not sure what the value is in "I just with [sic]
there was a way to reliably get a -rc release from git as well", in
_my_ CI use case I don't care about the superseded stable -rc versions,
just whatever is about to be released.

Others with more complex CI infrastructure, like Linaro etc, might feel
differently :)

> Also, as proof this works, I just got a report of someone testing the
> queues and finding a problem at the moment, before we sent anything out
> for review.  So this is working well today.
> 
> thanks,
> 
> greg k-h

Download attachment "signature.asc" of type "application/pgp-signature" (229 bytes)

Powered by blists - more mailing lists