Re: Who wants to do the 2.12 RC release?

[Elijah Newren <newren gmail com>]

On 8/18/05, Vincent Untz <vuntz gnome org> wrote:
> Le jeudi 18 août 2005 à 12:30 -0600, Elijah Newren a écrit :
> > Right, I'll do my best to help if necessary but other things were
> > planned when you said you could be available as the backup.  But, I'm
> > guessing we're just being overly cautious--I bet Vincent has
> > everything done for the release and ready early Tuesday so that we
> > have a day to spare.
> 
> Ahah :-) Ready early Tuesday. You're joking. Let me think... It will
> take me at least 5 days to do the release ;-)

Actually, it is possible.  If you just say that tarballs are due 23:59
UTC on Monday (like I did in my last call for tarballs) AND happen to
be located in the Western hemisphere (say Utah though Hawaii would be
better), then you can actually start on the release Monday night which
leaves plenty of time to finish by "early Tuesday".  ;-)

To give you more of a reference, though, I _think_ I could do a
release in the time constraints below (I didn't time myself last time
but then again I was doing lots of debugging of my script and
restarting of the process):

a) Checking the smoketesting script stuff in the releng module out of
cvs -- about 1 minute or so.

b) Running my script to generate a new jhbuild moduleset (and download
all the necessary tarballs at the same time) -- about 12-15 minutes
(though I have a really fast connection at school; so downloading all
tarballs including external dependencies only takes about 10 minutes
or so)

c) Backing up ~/.jhbuildrc, copying a temporary one their for the
tarball build -- about a minute

d) building with jhbuild -- close to the same amount of time as
building a cvs release though an extra 5 minutes may be needed for the
little inevitable snags if not avoided beforehand (e.g. fixing autogen
flags since configure requires slightly different flags than autogen
does, removing dependencies that aren't really dependencies (pyorbit),
skipping mozilla as jhbuild can't build mozilla as a tarball (jhbuild
bug was filed), manually downgrading the gstreamer version since my
script picked up too new a version (sorry, haven't fixed that bug
yet); i.e. all the things I wrote as gotchas in the comments near the
top of my script).

e) Running the newly build version of Gnome and testing it--could be
pretty short but I used it for about 20-30 minutes (gotta play
mahjongg and freecell a few times to make sure the release is ready
for the masses...)

f) Editing the versions file to add modules that weren't picked up by
my script (perl and java bindings and pkg-config): probably about 5
minutes

g) Running all the necessary commands on window -- I don't remember as
the NEWS-generating scripts were kind of slow, but maybe 10-15
minutes?

h) Making the release announcement -- a couple minutes to find another
one off gnome-announce-list and a couple more to modify it to be up to
date.

(d) and (e) are the only huge time takers (and the potentially
problematic ones as if something doesn't build or doesn't work, it may
mean contacting module authors and getting new tarballs which could
really slow things down).  Most of the process is either automated
given the convert-to-tarballs.py script, the release scripts, and
jhbuild.  Also, everything other than (g) can be tried by someone any
day any time without messing anything up.

> More seriously, I'll look in detail at the MakingARelease page and ask
> you all the questions I'll have in the next two or three days. And I'll
> start doing the release on Tuesday, if all the modules are released.

I think we should just set a deadline for the tarballs (e.g. 23:59
UTC) and stick to it unless build issues arise.  Not doing so caused
problems with understanding of freezes (as witnessed on the recent
thread) and gives us less time to get the release out as people have
pushed to making releases later and later.  Does anyone see a problem
with doing this?

Cheers,
Elijah