Re: RFC: using supersedes: trailer to indicate patch/series revision flow

From: "Rafael J. Wysocki" <rjw@rjwysocki.net>
To: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Han-Wen Nienhuys <hanwen@google.com>,
	Konstantin Ryabitsev <konstantin@linuxfoundation.org>,
	workflows@vger.kernel.org
Subject: Re: RFC: using supersedes: trailer to indicate patch/series revision flow
Date: Fri, 08 Nov 2019 12:00:16 +0100	[thread overview]
Message-ID: <8377829.SsT3G3NtT2@kreacher> (raw)
In-Reply-To: <CAMuHMdXrPKUf5sJhw8s-NYCGpwDoLGOptN=poaCJb5dWDNkVaw@mail.gmail.com>

On Friday, November 8, 2019 10:59:55 AM CET Geert Uytterhoeven wrote:
> Hi Han-Wen,
> 
> On Fri, Nov 8, 2019 at 9:31 AM Han-Wen Nienhuys <hanwen@google.com> wrote:
> > On Fri, Nov 8, 2019 at 12:45 AM Rafael J. Wysocki <rjw@rjwysocki.net> wrote:
> > > > 2. Should supersedes: link to the previous version of the patch, or the
> > > >     first ever version of the patch? I am leaning towards the latter,
> > >
> > > And then how do you know that version 2 was superseded by version 3?
> >
> > You throw the message ID into a search engine, and see what it returns.
> >
> > The advantage of keeping the patch series ID stable is that you can
> > consider a patchseries as a document and then easily index it inside a
> > service (say, patchwork) using Lucene, ElasticSearch or some other
> > common technology.
> >
> > If you make the "supersedes" refer to specific versions, a workflow
> > service will be more susceptible to errors if messages were lost, and
> > the service has to work harder to aggregate the different versions of
> > a patchseries together.
> >
> > Is it common for different authors to superseed each other's patch
> > series? If yes, "superseeds: precise version" is more precise, if not,
> > you get the same information from the timestamp of the cover letter.
> 
> All/most of the above applies only to versioning of patch series that
> implement a fixed feature, with a fixed scope.
> So Vn+1 is really an improved version of Vn, and nothing more or less.
> 
> But this does not apply to many patch series in Linux kernel development.
> Patch series may
>   - grow (more patches added),
>   - shrink (some patches rejected, or already applied independently),
>   - be split in multiple series,
>   - dropped but some parts reused in other series (possibly by someone
>     else),
>   - ...
> That's the hard work in patch tracking.
> 
> So series IDs do not cope well with this, and "superseded" really should
> apply to individual patches, not series, too.

Right.