Hi Martin, On Tuesday 24 April 2007 21:45, Martin Hollmichel wrote:
> I had a lot of interesting discussion about various distributed SCM and > subversion in the meantime. For me it looks the discussion and evalution > of distributed SCM needs some more time than just a month because the > use of such a system also can improve or at least change our development > processes. I also have the impression that both subversion and a DSCM > (git, mercurial) can serve our needs. Unfortunately, SVN does not serve our (= people around ooo-build) needs :-( [CVS even less, of course.] What we need is a tool that allows us to have a modified codebase (ooo-build) while we are still able to contribute the modifications back to up-stream easily. Today we solve it with huge amount of patches (~1000), we have to create CWS'es from that, and it's becoming hard to manage, especially with big features like the VBA interoperability or cairo canvas. DSCM is so far the most suitable way these days (and git in particular seems to be the best). > The situation looks like this: > > We will have next week an infrastructure available where we have the > possibility to do the transition from CVS to subversion. The subversion > tool chain seem to be mature (support for all our main development > platforms), it has been tested on all these platforms also with firewall > restrictions (a test protocol matrix with benchmarks should be available > soon). A transition to subversion would address our biggest pain: branch > creation. Sure, I'm not against evaluating SVN ;-) - so far I had the possibility to see just the OOo SVN tree by Martin Kretzschmar & Kai Backman which did not contain the entire OOo history, just some milestones. From their experience even this had performance problems :-( I also hope you don't want to do just a partial import as I saw somewhere; one where the history would consist just from 'integration commits' & the history in the branches would be abandoned. This would completely screw the ability to blame someone for a change that happend in CVS; and there would be no way back even if we decided to go with git (or any other DSCM) later - the information would be missing then. > A switch to DSCM is the bigger and a promising step: enhancements of our > development model and style look possible, I can't list all aspects now. > But a transition to a DSCM seems to need more preparation and > development: we need to review and maybe redesign our processes, we need > a more mature system with support for all major platforms, need to think > about additional infrastructure and more things. Yes. A prerequisite for seeing the most advantages from the move to DSCM is a split of the sources to smaller parts - URE, OOo w/o URE, copies of the system libraries, translations, ODK. [And ideally even a separate ODF toolkit ;-)] OTOH - to some extent this can be prepared in parallel with the current development; and then a 'big bang' switch ;-) - if we agree that this is what we want. > So I would like to suggest the following: > Evaluate the effort for a transition to subversion first, Heiner already > did some preparations for this. If this efforts seems to be reasonable > low, do the transition from CVS to subversion first in the next few months. There is no OOo SVN tree yet, or do you already have one, please? > As a second step or in parallel keep the evaluation for alternative SCM > open, lets use face-to-face meetings on next OOoCon for a more in depth > discussion about the future developments in our process. I guess some of > you may think that with a transition from CVS to subversion this > discussion might be dead, but I definitely don't think so. The > opportunities we have with a modern SCM are to much to get ignored but > we need some time for such thought which I don't want to let get passed > by by sitting on our old CVS. This generally sounds OK to me if we agree that: - No switch to SVN until a converted tree will be publically available for testing for some time - It will contain all the history so that a move to another SCM later is possible, and 'svn blame' works well Please note that here's no advantage of switching to SVN for ooo-build community, so I'll be still pushing DSCM, particularly git ;-) Of course, we can live even with SVN [we lived with CVS for quite some time], but in fact nothing will change for us, and we will have to think again of how to solve the problems described above if the discussion about DSCM closes with 'no'. Regards, Jan --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
