RFR: GIT Package VCS

Dave Jones davej at redhat.com
Sun Jun 10 18:20:23 UTC 2007

On Fri, Jun 08, 2007 at 02:11:30AM +0200, Karel Zak wrote:
 > On Thu, Jun 07, 2007 at 11:04:56PM +0200, Nicolas Mailhot wrote:
 > > Not surprisingly, someone like Andrew Morton that needs to rebase
 > > regularly from Linus *and* trace transient patches uses quilt instead of
 > > an exploded vcs
 >  Ask Linus what he thinks about Andrew's quilt :-) I think that's
 >  personal choice only.

It's not. They're doing two different jobs, and the tools to do
each efficiently are different.

Linus' role is to pull from many disparate sources and end up
with a consistent tree. For this git wins, hands down.
However, stuff that gets into Linus' tree usually stays there.
If something is merged that turns out to be broken, it's usually
patched with a follow-up patch, or very rarely.. reverted.

Andrew on the other juggles thousands of patches, and regularly
just drops a bunch of them if they turn out to not work out.
The ability to do this easily is why quilt wins over git.
Doing a dozen git reverts on patches you merged earlier
_really_ sucks if you've got subsequent patches that sit
on top of those.

 > I'm almost sure that someone other (for example
 >  Linus) will use GIT instead quilt for same job.

I tried it during F7 after FUDCon.  I thought it'd work out too.
It didn't.  As soon as rebasing broke Xen, it became a nightmare,
as I couldn't easily drop it.  It became easier to just
regenerate the tree from scratch without xen ever having been
included.  (This however kills your history).

Compare this to the method we use today, where I just comment
out some %patch's, and maybe rediff 1-2 of the follow-on patches.
This approach isn't too unlike quilt.



More information about the fedora-devel-list mailing list