Hi Alec,

On github you should be using a different branch other than master (say,
riscv-fs). Then, you can just rebase that branch on top of master (or
origin/master if you've got other commits on master). I often use "git
rebase -i master" when doing this. Then, during the rebase you can
squash/fixup and reword commits so they fit the gem5 style guides. After
rebasing, you can simply push to gem5.googlesource.com to create the review
requests :).

With rebasing you do lose the "local" history, but if you want to keep that
you can simply rebase to a branch with a different name (say,
upstream/riscv-fs in this example). I would argue that it's OK to lose the
history in most of these cases as it's really the final cleaned-up commits
that matter.

I hope this is what you were asking :). I should do a write up on what my
git+gem5 workflow is... I'll put that on the (increasingly long) list of
blog posts I need to write.

Cheers,
Jason

On Sat, Apr 21, 2018 at 5:29 PM Alec Roelke <[email protected]> wrote:

> That's essentially what I was getting at, but I think you put it better.
> If I do that, then what is the best way to create a patch from the changes?
>
> On Fri, Apr 20, 2018 at 6:15 PM, Jason Lowe-Power <[email protected]>
> wrote:
>
> > Hi Alec,
> >
> > What you're describing is *hard* and requires a lot of coordination
> between
> > people. I think the best way to do it today is to use something
> out-of-band
> > like a github mirror and then have one person rebase, etc. and submit to
> > the mainline. That's basically what AMD and ARM do internally. It's
> harder
> > to do with a geographically diverse set of people.
> >
> > If you need hosting and want to use gerrit for code review during
> > development, we can set you up with a repository on
> gem5.googlesource.com.
> > However, I think the path of least resistance is to set up a github repo.
> > E.g., I have a github repo for my research group and that's what we used
> > when I was at Wisconsin, too.
> >
> > Cheers,
> > Jason
> >
> > On Thu, Apr 19, 2018 at 12:02 PM Alec Roelke <[email protected]> wrote:
> >
> > > I'm not trying to suggest that there should be an "official" way to
> > > collaborate on patches, since that's probably not a common enough
> > > occurrence to warrant the effort of maintaining such a feature; I was
> > > mostly wondering if there was a way to leverage what's already
> available
> > to
> > > do it.  Downloading and uploading revisions to others' patches is
> clunky
> > > and, as Gabe mentioned, can lead to loss of work.  I was thinking more
> of
> > > something like what Jason mentioned with creating a "riscv-fs" clone of
> > > gem5 in GitHub or similar and creating patches out of commits to it
> > > somehow.  Is that possible?
> > >
> > > I think overall what I'm trying to figure out is proper version control
> > for
> > > patches before they're uploaded to Gerrit so that two or more people
> can
> > > work on one thing without worrying about losing work to conflicts.
> > >
> > > On Wed, Apr 18, 2018 at 8:49 PM, Gabe Black <[email protected]>
> > wrote:
> > >
> > > > In my past experience with gerrit, anybody can overwrite anybody
> else's
> > > > patch. That may not be true for our instance here, but I suspect it
> is.
> > > All
> > > > you have to do is make sure the Change-Id: field is the same since
> > that's
> > > > how gerrit identifies changes. It's generally something to avoid
> since
> > it
> > > > can be inconvenient for the original author to download your new
> > version
> > > to
> > > > replace their own locally, to make your edits locally, etc., to avoid
> > > > blowing away your changes when they upload their own new version.
> > > >
> > > > I think the linux kernel uses both multiple git repositories and
> > multiple
> > > > git branches within those repositories to separate independent
> > projects.
> > > A
> > > > big problem with that approach is that it takes concerted effort to
> > keep
> > > > all those different threads in sync and make sure they reconverge at
> > all,
> > > > and ideally frequently. They have vastly more developers and projects
> > to
> > > > coordinate than we do, and I don't think the overhead of that
> approach
> > > > makes a lot of sense for us at this point and with our current
> capacity
> > > for
> > > > administrative overhead.
> > > >
> > > > If you want to push up your own independent version of somebody's CL
> > for
> > > > some reason, all you have to do is replace the Change-Id field
> (delete
> > it
> > > > and let git recreate it with a new value, for instance) and gerrit
> will
> > > > treat it as a new patch even if everything else is the same. There
> are
> > > > other ways to put changes up on gerrit which are work in progress
> sorts
> > > of
> > > > things. Those used to be draft changes, but I think they (the gerrit
> > > > developers) reworked that into one or two similar but different
> > > features. I
> > > > think one is actually called a work in progress CL?
> > > >
> > > > Gabe
> > > >
> > > > On Wed, Apr 18, 2018 at 3:18 PM, Jason Lowe-Power <
> [email protected]
> > >
> > > > wrote:
> > > >
> > > > > Hi Alec,
> > > > >
> > > > > First, I'm not sure if collaborating on a single changeset is a
> > common
> > > > use
> > > > > case for code review platforms. Though, I have run into a number of
> > > times
> > > > > where it would have been useful.
> > > > >
> > > > > Am I understanding your suggestion correctly? What you're
> envisioning
> > > is
> > > > > there would be a feature branch (say, riscv-fs) and many people
> could
> > > > push
> > > > > changesets onto that branch. Then, when the feature is working you
> > > would
> > > > > merge into master? Or, squash and go through code review? I'm not
> > sure
> > > > how
> > > > > the final part would work.
> > > > >
> > > > > I'm open to this kind of workflow. We can create branches on
> gerrit.
> > > > > Another option is what ARM and AMD are doing which is having their
> > own
> > > > repo
> > > > > hosted on gem5.googlesource.com.
> > > > >
> > > > > What does the Linux kernel do in this case, I wonder. They have
> done
> > a
> > > > good
> > > > > job working out procedures for massive open source development.
> They
> > > are
> > > > > usually a good inspiration.
> > > > >
> > > > > Second, as a maintainer, I believe you have enough access to just
> > > > overwrite
> > > > > someone else's commit. For instance, if you download a changeset,
> > then
> > > > > rebase, then push to refs/for/master it will update the review
> > request.
> > > > So,
> > > > > as a maintainer, you can make a small change instead of asking the
> > > > original
> > > > > author. (I think this is right... I know *I* can modify patches
> and I
> > > > think
> > > > > it's the maintainer flag, not admin. If I'm wrong we can talk about
> > > > > changing this permission.)
> > > > >
> > > > > Cheers,
> > > > > Jason
> > > > >
> > > > > On Wed, Apr 18, 2018 at 1:02 PM Alec Roelke <[email protected]>
> > > wrote:
> > > > >
> > > > > > Hi Everyone,
> > > > > >
> > > > > > Is there a way to allow multiple people to collaborate on a
> patch?
> > > > Right
> > > > > > now it looks to me like the only person who can make changes to a
> > > patch
> > > > > is
> > > > > > the owner, which means that others' contributions have to be
> > manually
> > > > > > added.  Would it be possible for a person to create a personal
> > branch
> > > > of
> > > > > > gem5 on GitHub or other git repository and then somehow push
> > changes
> > > > from
> > > > > > there to Gerrit?
> > > > > >
> > > > > > Thanks,
> > > > > > Alec Roelke
> > > > > > _______________________________________________
> > > > > > gem5-dev mailing list
> > > > > > [email protected]
> > > > > > http://m5sim.org/mailman/listinfo/gem5-dev
> > > > > _______________________________________________
> > > > > gem5-dev mailing list
> > > > > [email protected]
> > > > > http://m5sim.org/mailman/listinfo/gem5-dev
> > > > >
> > > > _______________________________________________
> > > > gem5-dev mailing list
> > > > [email protected]
> > > > http://m5sim.org/mailman/listinfo/gem5-dev
> > > >
> > > _______________________________________________
> > > gem5-dev mailing list
> > > [email protected]
> > > http://m5sim.org/mailman/listinfo/gem5-dev
> > _______________________________________________
> > gem5-dev mailing list
> > [email protected]
> > http://m5sim.org/mailman/listinfo/gem5-dev
> >
> _______________________________________________
> gem5-dev mailing list
> [email protected]
> http://m5sim.org/mailman/listinfo/gem5-dev
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev

Reply via email to