Konstantin Ryabitsev <icon(a)fedoraproject.org>:
On 12/24/06, Michael Tiemann <tiemann(a)redhat.com> wrote:
>Konstantin--how many packages do you maintain? I think that rather than
>sniping at a would-be contributor, I'd like to see somebody who is
>maintaining at least 30, and perhaps 50 packages explain how *they* do
>it. Maybe they have a better way. Maybe it drives them almost as crazy
>Eric. How *does* a maintainer of 36 packages would with the Fedora
>process? How *should* one do it? This is the question and the problem
>to be solved.
I maintain a very small number -- only 15. From my own experience, I
can tell you that work spent actually doing something with spec files
is negligeable compared to how much effort is spent tracking what is
going on with a project, doing test builds and verifying that they
work, running rpmlint, responding to bugzilla requests, opening
upstream bugzilla requests for bugs filed with a package, monitoring
cvs commits to see if someone was "messing" with your projects, etc.
I already do all of these things except running rpmlint. They're
part of being a project lead. I'd be happy to add running rpmlint
to my process.
You can't script that. If project "a" releases a
"point update" to
"1.1.2" from "1.1.1," it's not enough to run "make
new-sources". You
have to read the Changelog to see why they have issued this update
(security? rush it through. fixes for solaris builds? whatever, ignore
it). You then have to spend a few days just monitoring the list to see
if there is an "oh shit, that release breaks something" moment --
those tend to happen frequently.
Yes, this sounds very familiar.
Only then, after a few days, you get
to actually run "make new-sources", run a test mock build, run rpmlint
against the packages, then copy sources, .cvsignore, foo.spec into
FC-5 and devel. CVS commit, make tag, make build.
Can it be scripted? I have no doubt. Will that actually be solving any
problems? I don't think so.
With respect, sir, I decline to substitute your speculation for my
experience. At 36 projects I find that the mechanical overhead of
manual point releases drives me batshit. I've written a tool called
'shipper' to avoid it; you can download it from my site. What I want
to be able to do is add a channel to shipper that lets me script-ship
to Fedora the way it presently can to Freshmeat and my own website.
The integral part of caring for fedora
packages is the human element -- making sure that you don't commit
something broken.
I don't disagree. But I also don't understand why you think this
is relevant to my complaint. When I've done all my QA, passed all
ny regression tests, test-built my RPM and done everything I know
how to do to ensure I don't ship a broken release, I want to be
able to push a button and *go*. There's no virtue in my having
to remember details or perform rituals by hand at that point.
Sure, I don't have "36 projects" -- I don't even
have catchy 3 letters
to go by, but I assure you that I keep just as busy.
Ouch. Please don't blame me for "ESR" -- people hung that on me by
analogy with "RMS" and I've never been entirely comfortable with it.
As I note on my Slashdot profile, I didn't ask to be triletterized,
I merely found that my role left me unable to escape it.
--
<a
href="http://www.catb.org/~esr/">Eric S. Raymond</a>