dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, Jun 17, 2011 at 10:49:47AM +0100, Garth N. Wells wrote:
>
>
> On 17/06/11 10:43, Anders Logg wrote:
> > On Fri, Jun 17, 2011 at 10:33:23AM +0100, Garth N. Wells wrote:
> >>
> >>
> >> On 17/06/11 10:28, Anders Logg wrote:
> >>> On Fri, Jun 17, 2011 at 11:24:21AM +0200, Marie E. Rognes wrote:
> >>>>
> >>>> On 17. juni 2011, at 11:05, "Garth N. Wells" <gnw20@xxxxxxxxx> wrote:
> >>>>
> >>>>>
> >>>>>
> >>>>> On 17/06/11 09:34, Anders Logg wrote:
> >>>>>> On Fri, Jun 17, 2011 at 08:46:46AM +0100, Garth N. Wells wrote:
> >>>>>>>
> >>>>>>>
> >>>>>>> On 17/06/11 08:35, Garth N. Wells wrote:
> >>>>>>>>
> >>>>>>>>
> >>>>>>>> On 12/06/11 21:36, Anders Logg wrote:
> >>>>>>>>> On Fri, Jun 10, 2011 at 10:33:41PM +0100, Florian Rathgeber wrote:
> >>>>>>>>> On 11/05/11 15:43, Anders Logg wrote:
> >>>>>>>>>>>> On Wed, May 11, 2011 at 03:57:52PM +0200, Anders Logg wrote:
> >>>>>>>>>>>>> On Tue, May 10, 2011 at 10:57:07PM +0200, Martin Sandve Alnæs wrote:
> >>>>>>>>>>>>>> On 10 May 2011 17:52, Anders Logg <logg@xxxxxxxxx> wrote:
> >>>>>>>>>>>>>>> On Tue, May 10, 2011 at 03:49:18PM +0200, Martin Sandve Alnæs wrote:
> >>>>>>>>>>>>>>>> On 3 May 2011 18:25, Johannes Ring <johannr@xxxxxxxxx> wrote:
> >>>>>>>>>>>>>>>>> On Tue, May 3, 2011 at 3:52 PM, Anders Logg <logg@xxxxxxxxx> wrote:
> >>>>>>>>>>>>>>>>>> It happens to me a lot. Johannes has tried to explain to me why it
> >>>>>>>>>>>>>>>>>> happens a number of times but I still don't understand why.
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>> Maybe he can try to explain it again to you and then I might also
> >>>>>>>>>>>>>>>>>> understand. :-)
> >>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>> I think I just bring up the following instructions (which I think looks good):
> >>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>> http://wiki.squid-cache.org/BzrInstructions
> >>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>> Thanks Johannes.
> >>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>> Basically the problem is that bazaar numbers commits with contiguous
> >>>>>>>>>>>>>>>> integers, and when Bob and Alice works locally they will get the same
> >>>>>>>>>>>>>>>> commit ids for different commits. When you stand in branch Alice and
> >>>>>>>>>>>>>>>> merge from branch Bob, the commit numbers of branch Alice are
> >>>>>>>>>>>>>>>> conserved and a single new merge commit is recorded on top there. The
> >>>>>>>>>>>>>>>> commit numbers from branch Bob are lost in the merge. Therefore, to
> >>>>>>>>>>>>>>>> conserve the commit ids in the central branch, you have to merge from
> >>>>>>>>>>>>>>>> your own branch into the server branch, not the other way around.
> >>>>>>>>>>>>>>>> Otherwise we can never safely use the commit revisions from the
> >>>>>>>>>>>>>>>> central branch, since they may change every time somebody merges the
> >>>>>>>>>>>>>>>> 'wrong way'.
> >>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>> This problem does not occur with hg or git, because they use a hash
> >>>>>>>>>>>>>>>> value to identify a each commit.
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>> So if I'm Bob and Alice has pushed some changes to the main branch
> >>>>>>>>>>>>>>> before me, which exact commands should I write?
> >>>>>>>>>>>>>>
> >>>>>>>>>>>>>> Depends on how you set up your repositories, where your branches are
> >>>>>>>>>>>>>> located, etc...
> >>>>>>>>>>>>>>
> >>>>>>>>>>>>>> You really have to read up on it and try it out a bit to understand
> >>>>>>>>>>>>>> it, and I doubt I can write it better than what Johannes linked to +
> >>>>>>>>>>>>>> the bazaar docs.
> >>>>>>>>>>>>>>
> >>>>>>>>>>>>>> I plan to keep a local repository with multiple branches like this:
> >>>>>>>>>>>>>>  ~/dev/fenics/ufl/ - local ufl repository
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> Is this a repository? Or is it just a directory named ufl inside which
> >>>>>>>>>>>>> you keep a number of different repositories?
> >>>>>>>>>>>>
> >>>>>>>>>>>> Talked to Martin during lunch. Here's a simple summary of what needs
> >>>>>>>>>>>> to be done to set things up correctly (Cc to dolfin-dev so everyone else
> >>>>>>>>>>>> sees this):
> >>>>>>>>>>>>
> >>>>>>>>>>>>  bzr init-repo foo
> >>>>>>>>>>>>  cd foo
> >>>>>>>>>>>>  bzr checkout lp:foo trunk
> >>>>>>>>>>>>  bzr branch trunk work
> >>>>>>>>>>>>
> >>>>>>>>>>>> We should add this to the developer page in the documentation.
> >>>>>>>>>>>>
> >>>>>>>>>>>> Everyone should adopt this and we should pick on anyone that pushes
> >>>>>>>>>>>> removed changesets.
> >>>>>>>>>
> >>>>>>>>> There's an effective way to make these pushes impossible and disable the
> >>>>>>>>> bzr "feature" of renumbering revisions: set the option
> >>>>>>>>> append_revisions_only=True in <yourbranch>/.bzr/branch/branch.conf
> >>>>>>>>>
> >>>>>>>>> For branches on launchpad this works as follows:
> >>>>>>>>> 1) sftp bazaar.launchpad.net
> >>>>>>>>>     cd ~user/project/branch/.bzr/branch
> >>>>>>>>>     get branch.conf
> >>>>>>>>> 2) edit the downloaded file, adding append_revisions_only = True
> >>>>>>>>> 3)   put branch.conf
> >>>>>>>>>
> >>>>>>>>> I suggest doing this for all branches on launchpad to enforce consistent
> >>>>>>>>> revision numbers.
> >>>>>>>>>
> >>>>>>>>> More background: http://stackoverflow.com/questions/5413602
> >>>>>>>>>
> >>>>>>>>>> Thanks. I've fixed this now for DOLFIN, FFC, UFL, UFC, FIAT.
> >>>>>>>>>
> >>>>>>>>
> >>>>>>>> Could you please undo this. I can't push changes from my personal branch
> >>>>>>>> to DOLFIN. I don't see that this change has any use. (If we want cvs,
> >>>>>>>> then we should use cvs.)
> >>>>>>>>
> >>>>>>>> I've tried to follow the instructions to undo the change, but can't get
> >>>>>>>> it to work.
> >>>>>>>>
> >>>>>>>
> >>>>>>> I've undone this for DOLFIN so I could push my changes.
> >>>>>>
> >>>>>> You should have figured out how to do the merge properly instead. We
> >>>>>> should add it back to force everyone to learn how to use bzr. ;-)
> >>>>>>
> >>>>>
> >>>>> Merge is 'bzr merge xxx'. That's a proper merge.
> >>>>>
> >>>>>> The point is to not rewrite history for the common repo. This is not
> >>>>>> the same as cvs. It's still distributed but it means merges have to be
> >>>>>> done more carefully.
> >>>>>>
> >>>>>
> >>>>> There is no history re-writing. It's just adding changesets. Unique
> >>>>> changeset numbering that bzr does will always be problematic with
> >>>>> distributed version control. If you want a unique identifier, use the
> >>>>> revision id.
> >>>>>
> >>>>>> Just do this next time and it should work:
> >>>>>>
> >>>>>> 1. Make sure you have a local bound dolfin branch:
> >>>>>>
> >>>>>>  bzr checkout lp:dolfin trunk
> >>>>>>
> >>>>>> 2. Merge *from* that branch, not push to it:
> >>>>>>
> >>>>>>  cd trunk
> >>>>>>  bzr merge <path to your local repo>
> >>>>>>  bzr commit -m merge
> >>>>>>
> >>>>>
> >>>>> It just worked before. It was simpler, and I could work against any
> >>>>> branch, like by personal branch under dolfin-core.
> >>>>>
> >>>>
> >>>> After trying the no-revisions-removed approach for a while, I also find it significantly more cumbersome, especially with the main vs personal branches.
> >>>>
> >>>> Although I see the point, I never encountered a problem with the changing revision numbers before.
> >>>
> >>> If Garth can't be bothered, maybe you could describe a specific
> >>> example that doesn't work?
> >>>
> >>
> >> Why would I bother with something that I think is pointless and
> >> cumbersome? Like Marie, I have never had a problem with the present
> >> approach.
> >
> > I disagree. I think there is a point to it and that it's not
> > cumbersome.
> >
> > I'm just asking for a simple example that I can try. Otherwise it's
> > just handwaving.
> >
>
> You made the change, so the onus is on you to make a case, not the other
> way around. I'm changing it back because I don't have any inclination to
> change unless someone can make a case why. The status quo rules until
> there is a consensus.
>
> What I don't want is to work in a CVS way. See:
>
>   http://wiki.bazaar.canonical.com/BzrForCVSUsers
>
> It advocates checkout for those who want to keep their CVS work flow,
> and says of the approach:
>
> "This section explains how to perform common CVS behaviours in a Bazaar
> world. Unfortunately, this means that I won't be able to teach you many
> of the things that are unique to decentralized revision control systems.
>
> This section covers how to use Bazaar in checkout mode. Reading section
> 3, which covers standard Bazaar methods, is highly encouraged."
>
> Section 3 describes what we've been doing all along.

I think you still need to make the case that it doesn't work. I claim
it does and if you say that it fails so badly, it should be easy to
come up with a single example of where it doesn't work.

I've already made the case for the change: to not change history of the
common branch (which append_revisions_only prevents).

--
Anders