On Sat, Feb 6, 2010 at 21:13, Daniel Friesen <li...@nadir-seen-fire.com> wrote:
> I believe we've had two discussions in the past on switching to git.
>
> I talked to Tim about various other advantages of .git, like the lack of
> autoprops annoyance, and corrected the notion that there isn't a Windows
> client and his reponse was "maybe in a year or two".
>
> Generally the limitation is the fact that we're currently abusing svn's
> ability to only check out specific directories rather than an entire
> repo to make it easy to check out all the extensions or individual ones
> without any trouble.
> We've had ideas like using git submodules to mark stable versions of
> extensions so extension repos can be flexibly checked out.
>
> Oh something interesting. With a bit of trickery I recently managed to
> splice the entire history of one git repo into a branch of another git
> repo creating a git repo that has two separate initial commits in two
> separate branches. And from the looks of it it's perfectly possible to
> fetch history from the original repo into the proper branch. So it
> should be interestingly possible to create a script that fetches history
> updates for every extension at once by embedding them all into separate
> branches of a single git repo, and then locally (with no network
> latency) pulling the history from those branches into the real repos.

It's interesting that the #1 con against Git in that document is "Lots
of annoying Git/Linux fanboys".

I guess this is as good a time as any to plug the git-svn notes I
scribbled down yesterday: http://www.mediawiki.org/wiki/Git

In order to convert to Git it would help to collect a list of things
that should be split into separate repositories:

 * /USERINFO, /civicrm and /wikimedia-web
 * Everything in /trunk/*
 * Additionally /trunk/extensions/* and maybe some /trunk/tools/*

That should yield around 500 repositories. That might sound crazy but
best practice for any distributed version control system is that
repositories should be split at the boundaries at which code doesn't
pass over, and when's the last time /trunk/FOO shared some code with
/trunk/extensions/BAR for instance?

And if someone really wants to check out all 430 extensions that's
easy enough with an "extensions" project with 430 submodules, but the
most common case should be someone checking out MediaWiki.git + 3-5
extensions.

I'm doing some experiments with splitting up MediaWiki's Git mirror[1]
using git-filter-branch[2]. It takes a *long* time with this huge
repository but a working conversion is the fastest way to get people
on board.

Of course this is a great chance to clean up some aspects of the
repository, such as:

 * Rewrite the commits to give everyone real names / emails, like Tim
Starling / tstarl...@wikimedia.org instead of tstarling. This can be
done automatically by parsing the USERINFO files & adding to them
where appropriate.
 * Combine users like magnusmanske and magnus_manske into one
 * Rename/drop branches/tags if someone wants that

1. http://gitorious.org/mediawiki-svn-mirror
2. 
http://stackoverflow.com/questions/359424/detach-subdirectory-into-separate-git-repository

_______________________________________________
Wikitech-l mailing list
Wikitech-l@lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wikitech-l

Reply via email to