master process handling patch

[Henrique de Moraes Holschuh]

On Wed, 21 Jul 2010, Patrick Goetz wrote:
> This isn't entirely true.  This patch, which I mentioned previously,
> looks like it could lead to header corruption in the saved header
> stored in ibuf:

I don't think so.  Look at it very carefully.  Still, it was clearly useless
so it was reverted quite recently (as in the last 48H).

The one patch that I am _really_ not that sure about is the one about using
SIGALRM to break locks.  Might disable that one before uploading to
unstable.  THAT one certainly needs some upstream eyes.

> This is kind of an insane situation that there are bug fix patches
> hanging around that have been in the debian source package since
> 2.3.8 which haven't been pushed upstream.  As a result, I don't

Some of those patches have been in Debian without being pushed upstream
since 2002.  Granted, those were in Cyrus 2.1.  They were forward-ported to
2.2, and then to 2.3.

It is all going to the upstream BTS soon.

> trust any of the patches until I've checked them myself.  And since
> I don't understand what patches 12,13 are doing, I'm hoping you'll
> have time to go over them soon.

Patches 12 and 13 have been cleaned up (13 was broken up into separate
patches as well), and documented.  Now I am going to try fixing the !@#$
master state machine for good, it is subtly broken, and it has been subtly
broken for a *LONG* time.  Needs a rather pathological load to cause any
issues, though.

You should get very worried if you start getting any sort of complains about
unregistered children.  Master will need a restart if it deals with too many
of them.

> Meanwhile, I'm going over the patches the redhat people added to
> cyrus-imapd-2.3.16-5.src  (actually, first comparing the differences
> between this newer version and cyrus-imapd-2.3.16-3.fc13.src) to see
> if there's anything there that needs to be migrated to the debian
> package.

And upstream :-)

> I haven't done any C programming in a loooong time, so part of the
> delay on my part was getting back up to speed in C.  Now that that's
> done, I hope to be through this in a few days (modulo being out of
> town all next week).

Thanks for doing this!

-- 
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh