incubator-ooo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jens-Heiner Rechtien <jhrecht...@web.de>
Subject Re: Subversion & Git (was: Proposed short term goals)
Date Tue, 14 Jun 2011 18:01:00 GMT
On 06/14/2011 05:22 PM, Greg Stein wrote:
> On Tue, Jun 14, 2011 at 11:08, Simon Phipps<simon@webmink.com>  wrote:
>> On Tue, Jun 14, 2011 at 3:59 PM, Sam Ruby<rubys@intertwingly.net>  wrote:
>>
>>> On Tue, Jun 14, 2011 at 10:48 AM, Christian Grobmeier
>>> <grobmeier@gmail.com>  wrote:
>>>>
>>>> Do you see any serious problems if the code will go into svn in the
>>>> first step (which will last for a while)?
>>>
>>> I'd like to rephrase that question: does anybody here have any
>>> credible alternate proposal?  To be credible, requires not just an
>>> outline of a plan, but actual volunteers with a demonstrated ability
>>> to follow through.
>>>
>>
>> If Git was going to be available at Apache in the future I am pretty sure
>> the LibreOffice developers would consider temporarily hosting a Git
>> repository for developers here to use.
>
> That is unworkable right now.
>
> Git has been "in the works" for several *years* now. We cannot rely on
> it appearing at any given point in time. Certainly not by the time
> that we want to release an Apache OOo.
>
> You also have authentication problems between Apache and the LO
> installation. You also have code guarantee problems (the ASF
> guarantees that third parties have not tampered with its development).
>
> Seriously. If it was "that easy", then Infra would have just set up a
> git repository and be done with it. But it isn't "that easy".
>
> There are thousands of open source projects using Subversion. And
> there are downstream users of those projects that do not use
> Subversion. I fail to see how this one is different. I also fail to
> see any credible reason that Subversion cannot handle the development
> here at Apache.


Before I comment on the SCM aspect, please let me introduce myself, as 
I've just subscribed to this list. My name is Jens-Heiner Rechtien 
(hr@openoffice.org). I've been with SO/OOo for more than 14 years, most 
of the time as OOo release engineer and as technical lead of SO release 
engineering.

I was quite involved in the evolution of the OOo SCM tooling from 
PVCS->CVS->SVN->Mercurial over the years so maybe I can add my 2 cent here.

We migrated in 2008 from CVS to SVN and it was a horrible disaster right 
from the start. Mostly my fault, as I relied on the SVN 1.5.x merge 
tracking mechanism to replace our homegrown merge tracking on top of 
CVS. Merge tracking was brand new in SVN at that time and just wasn't 
polished enough to cope with the OOo development model (feature branches 
with frequent merges from the main code line into the branches to keep 
things current) and size.

I got it working more or less eventually, but the experience left 
probably some deep scars in the OOo development community. I'm pretty 
sure that SVN merge tracking has improved a lot in SVN 1.6.x but I'm not 
convinced that some of the more intricate architectural problems with 
SVN merge tracking are really solvable.

I guess it boils down which development model Apache OOo will follow. If 
it's similar to the one we had before, with its heavy reliance on 
feature branches I would definitely advise against SVN. If committers 
simply want to commit directly from the workspaces into the main 
development code line, thus branches are restricted to the occasional 
release code line, than SVN is plenty good enough.

As for DSCMs, both Git and Mercurial are suitable choices for OOo. We 
decided in 2009 to use Mercurial mostly due to the perceived less steep 
learning curve. IMHO this is still true but other aspects might be more 
important nowadays and Git would certainly be a great choice.

Heiner

--
Jens-Heiner Rechtien

Mime
View raw message