[5709] in testers
Re: patch release breaks Sun dev environment for non-updated
daemon@ATHENA.MIT.EDU (Greg Hudson)
Mon Mar 8 14:23:43 2004
From: Greg Hudson <ghudson@MIT.EDU>
To: Robert A Basch <rbasch@mit.edu>
Cc: testers@mit.edu, rel-eng@mit.edu
In-Reply-To: <200403081813.i28IDQtC002111@anhedonia.mit.edu>
Content-Type: text/plain
Content-Transfer-Encoding: 7bit
Message-Id: <1078773549.25713.32.camel@error-messages.mit.edu>
Mime-Version: 1.0
Date: Mon, 08 Mar 2004 14:19:10 -0500
On Mon, 2004-03-08 at 13:13, Robert A Basch wrote:
> The release of 9.2.28 to the dev cell broke the development environment
> on my 9.2.27 Sun, due to the absence of /usr/include/sys/rwstlock.h;
> this header is added by the 112233-11 jumbo kernel patch we applied,
> and is now included by /os/usr/include/sys/vnode.h (to which we have a
> symlink on local disk).
Here are the options a few of us were able to come up with, starting
with the favored one:
* Add a hack to update_ws on the 9.2 branch to create a symlink for
rwstlock.h if the update doesn't run. (Downside: machines which never
reactivate don't get the fix unless they reboot. We think the
population of server machines like this is minimal, but people might
also notice a problem if they simply haven't logged out since the patch
release went to field.)
* Back out the kernel jumbo patch release. At this point, that means
convincing a bunch of dev cell machines to back it out, which is doable
but complicated. This option is particularly unpleasant for ops, since
it would force them to manually apply the jumbo kernel patch on new
hardware they install with 9.2.
* Hack up /os somehow so that <sys/vnode.h> works without a
<sys/rwstlock.h> symlink. (There are various options.)
We like the first option because it's simple (it doesn't create any
permanent complexity), but it isn't perfect. Perhaps the third option
can be made to be simple, but I'm not sure how.
