Re: [PATCH 2/4] XFS: Use the inode tree for finding dirty inodes

[Russell Cattelan <cattelan@xxxxxxxxxxx>]

Mark Goodwin wrote:
> Russell Cattelan wrote:
> > > Internally, we're attempting to refine our patch acceptance processes,
> > > (e.g. gitify our internal dev tree and mirror it on oss so it's much
> > > easier to push back out to oss).
> > I'm sure you have seen this before:
> > http://oss.sgi.com/cgi-bin/gitweb.cgi?p=cattelan/xfs-import/.git;a=summary' 
> >
> > That is a running mirror of the ptools tree into git. (via the cvs tree)
>
> yes. But it's git -> ptools scripts that we need, preserving history, 
> etc.
> Niv has some scripts for this - they're not production quality yet, but
> we're getting there. Once this transitions, it'll be a *lot* easier for
> us to pull in patches from external developer branches because we'll all
> be using git for checkin.
Personally I don't see a reason to keep a ptools tree in lock step with
with a git tree. I all for not losing history (and I spent a bit of time 
when
the tree was re-organized to keep the rcs history in tack).
At this point the git tree has full xfs history and I would think this would
be sufficient for what ever code archeology  comes up.
> > It would be really nice to move all xfs development to git finally shut
> > down
> > the whole ptools -> cvs update process.
>
> once our internal dev tree is git based and internal git->ptools merging
> is fully automatic, there is no actual need to shutdown the cvs crons.
> It's worked for years and can stay running, no harm.
Ya I'm still amazed those scripts are holding up given nobody is giving
them any TLC.  :-(

> > This would help facilitate creation of more "experimental" trees and/or
> > branches
> > so there would not be such a long delay of getting patches distributed.
>
> I think we'd just end up with a git dev branch on oss, maybe with a
> daily pull from the internal dev tree (Russell, that would render
> your cvs->git mirror obsolete I guess). In any case, patch flow and
> turn-around should be greatly improved.
The one thing about about SCM's that they are entirely a pain in the
ass, but one of the most important tools in software engineering.
So whatever happens it should be simple yet sufficient.

> Anyone have comments on any of the above?
>
> Cheers
> -- Mark