Re: [Ksummit-discuss] Allowing something Change-Id (or something like it) in kernel commits

[Doug Anderson <dianders@chromium.org>]

Hi,

On Fri, Aug 23, 2019 at 9:35 AM Joel Fernandes via Ksummit-discuss
<ksummit-discuss@lists.linuxfoundation.org> wrote:
>
> On Fri, Aug 23, 2019 at 12:19 PM Dmitry Torokhov
> <dmitry.torokhov@gmail.com> wrote:
> >
> > Hi Thomas,
> >
> > On Fri, Aug 23, 2019 at 05:48:55PM +0200, Thomas Gleixner wrote:
> > >
> > > Yes, it's work for the submitter, but it's always work if the submitter
> > > wants to have a proper trace.
> >
> > Here is where I disagree with you. As a patch submitter, I frankly could
> > not care less about proper trace, history, etc. I might be putting cover
>
> But that is exactly what the problem statement is. Doug does care
> about tracing/history and wants that to be more robust etc.

Well, I care because I'm a system integrator, not because I'm a patch
submitter.  I try to make my patch submissions useful for system
integrators because I am also a system integrator.  ...but most patch
submitters don't try so hard.  We need to make this easy for them.


> > letter and outline the version changes, but I am doing that to reduce
> > friction and help committer to land my change. Thant's it. And
> > committers seem to have enough context from the provided version history
> > and memory of the previous iterations.
> >
> > Who this new ID benefits most is a system integrator that is tracking
> > all bits and pieces that are needed for their board to boot and work
> > properly. Hopefully the system integrator is a good community citizen
> > and not only wants to apply patches locally, but also make sure
> > stragglers end up in mainline after all. But they need additional
> > history to know whether the series was just forgotten, a new solution
> > was adopted instead, and so on.
> >
> > It also can help maintainers who need (maybe years later) to research
> > why particular change was made and whether there was additional
> > discussion around certain point of the change.
> >
> > The process you outlined above (collecting msg ids from previous
> > submissions, putting them into cover letter, etc) adds too much friction
> > at submission time so that submitters will not actually do any of that.
> > However having a git hook adding an ID to any new commit is much more
> > workable.
>
> I don't think a git-hook will solve what Doug is asking. There is
> always some need for manual labor to add the metadata pointing to
> previous patch posting. How does anyone but the patch submitter know
> what the link/id/whatever of the previous patch posting or series is?
> Unless git has some telepathy abilities that I did not know about.
>
> Since my day job involves gerrit usage, I have used gerrit and
> whenever I have a brand new patch for something; I have had to
> manually add the old Change-Id to the new change so that gerrit
> recognizes that it is linked to the old patch. That process is not
> done by a hook. Sure the initial generation is through a hook, but
> still.

You must have a different flow than I do.  For me the Change-Id just
naturally stays there as I make changes to the patch.  When I post
patches to an upstream list I keep them around as a local branch.
When changes are requested, I pick the old patches and amend to them.
...or I use "git rebase -i" and do a "FIXUP" to make changes to a
previous patch.  With this workflow the Change-Id is super low
friction and automatic.

...so for me the git-hook works perfectly well and the Change-Id just
naturally stays there in most cases.

Sure, if a given patch in the series is totally wrong and I re-write
it (but still want the re-write to be considered a new version of the
previous patch) then I have to manually copy the Change-Id for a
gerrit upload.  ...but I'd also be OK if people just get a new
Change-Id in this case.


-Doug