Am Dienstag, 14. Februar 2012, 15:18:32 schrieb Ruediger Meier:

> On Tuesday 14 February 2012, Adrian Schröter wrote:

> > Am Dienstag, 14. Februar 2012, 14:14:21 schrieb Ruediger Meier:

> > > On Tuesday 14 February 2012, Adam Spiers wrote:

> > > > Adrian Schröter (adrian@suse.de) wrote:

> > > > > Am Montag, 13. Februar 2012, 22:13:54 schrieb Adam Spiers:

...

> > > - not losing commit history when sr'ing/merging or linking packages

> >

> > Actually, this would be also a not acceptable regression, if you do

> > not see anymore when you have merged requests. So, there would be a

> > concept needed how to tag and handle reverts.

>

> Of course always "git merge --no-ff" into factory/master.

 

that is not a solution.

 

> > > - locally doing atomic (readable) commits, pushing them squashed to

> > > OBS to avoid re-building every commit.

> >

> > Oh, well, priorisation system should take care of that.

>

> My major point here is that changes across different topics should never

> be mixed into the same commit. But svn-like VCS leads to do it the bad

> way. Submit requests could be reviewed much easier if you see all the

> atomic commits behind a huge submit-request.

 

Frankly, I see this really as a minor thing during packaging work.

 

If you are unsure about your work, you branch anyway before submitting.

 

> > > - offline work is possible when OBS is down (which happened very

> > > often in the last time)

> >

> > actually, offline work (offline build) is supported meanwhile in osc.

>

> Is it possible to commit local only?

 

No. But again, I think this is a minor thing. It is way more important to have offline builds to be able to do packaging work.

 

> > And dependency calculation is still required server side in any case,

> > independend of the used src revisions system.

> >

> > > - fetching several package clones (regardless their names) into the

> > > same working copy to speed up testing/comparing

> >

> > we would need that anyway in seperate build environements or we can

> > handle the parallel build on server.

> >

> > > - rebasing commit history to have nice commit series to be pulled

> > > from the major projects (devel, factory etc.).

> >

> > ...

> >

> > Do not get me wrong, I am not against an additional git backend. But

> > there are _MANY_ of unresolved problems with that and there do not

> > even exists concepts at all how to solve all the regressions and

> > major problems a git backend would have.

> >

> > But everybody can take of course the existing code and work on it.

> > But the resulting code needs not to work on a single small package,

> > but scale also for projects like openSUSE:Factory.

> >

> > The api test suite may be helpfull to see _some_ of the problems when

> > you replace the backend.

>

> For now I just do "git init" within ocs working copies. It does already

> much of what I need allthough it can be confusing. An intermediate next

> step could be to improve the client side only by providing a git-osc

> similar to git-svn.

> Finally having git backend on OBS side someday would be really nice.

 

Well, if that works for you that is fine. But using git as repo is a complete

different level of problems. Just because git offers a feature for you it does

not mean it offers all the features (revision counting, on-the-fly-merge, revision tracking, shadows sources and so on) the obs needs.

 

Anyway, since people always only say that git would solve everything but no

one works on it, I am stopping now here ;)

 

Feel free to work on it and I promise we will help you run into problems...

 

bye

adrian

 

--

Adrian Schroeter

SUSE Linux Products GmbH

email: adrian@suse.de