[Scilab-Dev] Re: Re: Leaving SVN to GIT

[Barry Cardiff]

Hi,
Yes Linus certainly seems to have strong views on the subject, but lets not
loose sight of a couple of things:
- Linux (for some reason) had to change version control system from
bitkeeper anyway - Scilab doesn't
- by Linus' own admission SVN works and works well for many complex
projects.
Don't get me wrong, in a perfect world, I'd say 'go for it' so long as the
workflow and responsibilities are clearly defined. It does appear to be a
technically a better solution but we have to be careful not to do something
that raises the barrier to entry for part-time scilab contributors to a
point where they'll just not bother.
One of Scilab's strengths is that MS visual studio express is a supported
development environment making the addition of new code pretty easy - I fear
that moving to GIT will turn off some windows developers.

With regard to you 'workflow' examples, I don't see any problems using
branching on SVN to achieve the desired result. With a branch one team can
work independently from the others on some new 'rdically different' code
whilst allowing others to see (and use) this 'work in progress', without
effecting the main trunk. Furthermore it is possible to 'cherrypick' changes
(bug fixes / new features) on the trunk to be replicated on the branch as
required.

Despite what Linus claims in his talk - all merging is a painful process -
there will always be conflicts. What I fear is that (unlike Linux, where
Linus is the boss) we'll end up in a situation where lots of people will
have perfectly good trunks but yet at some point we'll need to make a single
binary release and these two situations are at odds with each other, and its
not clear to me how they be resolved.

Barry

2008/9/22 Sylvestre Ledru <sylvestre.ledru at scilab.org>

> Hello Barry,
>
> Nice to have new people coming into the discussion. ;)
> (Jonathan, I am also trying to answer to your questions in this email)
>
> First, if you have time, listening Linus talking about git is a
> pleasure: http://www.youtube.com/watch?v=4XpnKHJAok8
>
> > What we need prior to making a firm decision on this is a document
> > that clearly illustrates the desired work flow, pointing out how GIT
> > enables this work flow and how SVN doesn't. Further to this a
> > discussion on you vision for the future work flow is also necessary.
> Moving to git is not for fun or to be fashion. It has a cost and it is
> harder than SVN to use.
>
> Our main concern here is more about organisation than technical.
>
> The gap between version 4.X & 5.0 was much too long. We don't want to
> reproduce the same "mistake". Therefor, we need to be able to work on
> two versions in the meantime.
>
> For the next year and half coming, we will be working on three versions:
> The 5.0.X, 5.X and 6.0
> Keeping track of changes, we need to push the changes from one version
> to the other but we also need a new structure.
>
> Let me take an example.
> We have a few people who are currently working on rewriting the core of
> Scilab (rewriting gateways, memory management, parser, etc).
> We obviously cannot put these changes into the current trunk of Scilab.
> It is a long work.
> But we do need a few things:
> * this stuff must be synchronized with the actual trunk (bug fix, new
> features)
> * let other people get access to this branch. For example, I am not
> working on this for now but I would like to be able to checkout this and
> use for some experiments (I love getting seg faults ;).
> * we can experiment stuffs quickly and easily together. An other
> example, I am going to work on a new Scilab-MPI version. Yann Collette
> asked me to have a look asap on this. Thanks to git, I will just have to
> publish my branch and that's it. I won't have to worry to sync with the
> main trunk and the v6.X version (I will probably needs improvement from
> both).
>
> Of course, I won't compare Scilab with the Linux Kernel but I like the
> idea of organisation which drives the dev of Linux. Sorry for the URL, I
> had a better example but I cannot find it right now [1]:
> http://www.firstmonday.org/Issues/issue10_5/iannacci/#i3
> (Don't look this graph is term of person but more in term of "version").
> If I am unclear here, let me know :)
>
> About the fact that there is no good click-click-double-clik Windows
> client for Git, I admit that it is an issue... but not a blocker issue
> for us (at least for now).
>
> Sylvestre
>
> [1] Well, I am lying here, I know a better one but it is in a video:
>
> http://www.free-electrons.com/pub/video/2008/ols/ols2008-james-bottomley-git.ogg
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.scilab.org/pipermail/dev/attachments/20080924/f0f7c373/attachment.htm>