Benjamin Reed wrote:
> Max Horn wrote:
> > I) We could move all our code *except the package tree* right now to a
> > new system. Right? I.e., no need to touch the selfupdate code of fink
> > for that. Personally, I'd be happy to see that happen sooner rather
> > than later
> Yeah, we can.
> > II) For the package tree, the main blockers (I think) are
> > 1) adding selfupdate support for the new system
> > 2) migrating anybody using "selfupdate-cvs" cleanly over to a new
> > system
> > For 1), well, I looked at the Fink code, and I think adding a "fink
> > selfupdate-svn" should be relatively straightforward. Maybe a little
> > less so for git/hg/bzr, but they should also be feasible.
> Again, agreed.
> > III) Moving away from SF.net vs. staying: personally, I see no great
> > need to move away from SF.net, their SVN support has worked rock-
> > stable for us in the past 3 or so years. But I understand if you guys
> > want to move on -- yet I think that this is actually a separate from
> > the other two: We could first switch to SVN, then later migrate away
> > from SF.net. The "svn switch" command (and its likes in the other VCs)
> > make this quite easy to do.
> Honestly, SF has been mostly OK for a while now. They still have the
> occasional hiccough but it hasn't been a game-ender.
> > IV) A better bug/issue/patch tracker system: SF.net now offers trac
> > and Mantis, both are pretty good tools and way better than the SF.net
> > tracker. Problem in both cases (and also when hosting an external
> > tracker): Data import. SF.net recently signaled to me that they are
> > working on (but not promising anything!) a solution to migrate data
> > from their trackers to trac.
> This is the most important thing we could do. SF's "trackers" suck.
Amplifying a little bit on this:
Bug reports made on the extant bug tracker only get assigned to their
respective personnel if somebody has had time to do so. Monic
Polynomial has been very helpful in this regard of late, but having to
read through the reports and take action is a thankless job.
And even when this is done, we have maintainers who aren't on the list
of people to whom items are assigned, and so they have to be contacted
Unless there's an easy way to farm the reports out from another system
to appropriate maintainers, I'd say to dispense with using a bug tracker
for packages other than fink itself--email messages can find _all_ of
our maintainers, and it seems like an inefficient use of person-hours to
have somebody have to go through and do the magic to contact the
Submitting new packages is a different matter: since somebody has to
test-build the things anyway, the manual intervention aspect isn't
hugely inefficient. Ideally, I'd rather see something customized for
_our_ needs than using a more general-purpose tool, e.g. a checklist
that has to be filled out with mandatory and optional validation steps
before the package description gets uploaded; but I'm certainly not
prepared to write one. ;-)
> > So, assuming we can agree on a VC system, we could tackle the above
> > points in the order I mentioned. Starting with converting our repos
> > from CVS to a different format. Quite a task, but maybe we can even
> > find a volunteer to do this. With ScummVM, we did this migration a
> > couple years ago, importing all our CVS history, and we would never
> > consider going back. (Though we might switch from SVN to git/hg/bzr
> > one day).
> We are in the easy position that we don't have a lot of branches, so the
> conversion should go very smoothly.
> > In closing, let me say this about hosting externally: As long as we
> > have full control, and somebody to maintain and update all the bits
> > and pieces, fine by me. OTOH, I have been quite satisfied with the
> > many new features SF.net introduced recently, and how it alloss me to
> > use trac, MediaWiki and phpBB without having to worry about installing
> > security updates or separate user management. I know that we have had
> > terrible experiences with their CVS offering, and the trackers are
> > very rudimentary. But the same can't be said about the SVN and Trac
> > offerings, for example.
> Honestly, I don't see a pressing need to move off SF. I think it would
> be useful to use a distributed system like git for maintainers' sanity
> (experimental branches get a lot simpler) but SVN would not be the end
> of the world.
> SVN would also give us the ability to have a 'fink selfupdate-http' that
> can do raw pulls from the SVN repo without any extra tools.
This latter point is very important, to my mind. Even though the
built-in cvs on Leopard supposedly has some sort of HTTP proxy support
built in, I've not been able to confirm whether it works or not (I don't
need to go through one anywhere; and I haven't really had time to set
one up on my home box). Nor do I know if our cvs-proxy even works on 10.4.
A decent number of our users seem to have to deal with them, though; and
it would be nice if we had a robust option for them to do selfupdates.
Fink User Liaison