[2159] in Release_7.7_team
Minutes for 2000-03-08 meeting
daemon@ATHENA.MIT.EDU (Greg Hudson)
Wed Mar 8 16:45:52 2000
Date: Wed, 8 Mar 2000 16:45:43 -0500 (EST)
Message-Id: <200003082145.QAA10170@small-gods.mit.edu>
From: Greg Hudson <ghudson@MIT.EDU>
To: release-team@MIT.EDU
Attending: ajfox, othomas, zacheiss, aurora, rbasch, tb, ghudson,
jweiss, miki, wdc
1. SGI update
Right now the update mostly works except it goes off into la-la land
during the RQS step. We can skip this step, and it's just a library
optimization, so Bob should be able to hand over a functional SGI
update tomorrow.
2. Solaris update
The Solaris install/update is taking a while--an hour on a sun4u, ~100
minutes on a sun4m. The bulk of the time is split between making
symlinks and applying patches.
We think we can speed up making symlinks by eliminating some packages
from the install or by mounting the local filesystems asynchronously
(although mounting async would place another limitation on the ability
to recover from interrupted updates).
We could speed up the application of patches on cluster machines by
writing our own patchadd which doesn't do any verification, but that
would be of limited value since update time is more of an issue on
private workstations. (On the other hand, it's not clear that we have
any way of recovering gracefully if the verification fails, so perhaps
we don't want to do it on private workstations anyway.)
Unfortunately, the Solaris patchadd is an inscrutable 5000-line shell
script, so modifying it is untenable.
If we can speed up the application of symlinks, 30-45 minutes to
install patches is probably not a killer.
3. Linux update and other Linux issues
Hopefully the Linux update should be easy (as soon as we can actually
build the mainline Athena tree, which should happen RSN).
A beta of Red Hat 6 is apparently downloadable now. Because the Linux
update is fairly simple (or we hope so), we can probably use Red Hat
6.2 for Athena 8.4 if it comes out soonish, as long as AFS doesn't get
in the way.
Thomas is working on patch releases 29 and 30 (rc.conf re-merging is
waiting for 30). He notes that athena-ws will need to contain some
commands to make sound work on cluster machines and that this fragment
will need to stick around forever; on other ports, such code would
have gone into an update script and only get run once. This is okay
as long as these things live at the end of the script and are
well-commented.
owls wanted input on "when are we no longer a pilot?" specifically
for the question of when we can safely choose to buy machines for the
summer deployment and for quickstations. The release team believes
the pilot is a success in that it identified no unsolvable problems
other than the lack of Framemaker. (We may consider trying out the
beta release.)
4. mwm num lock issue
A bunch of people have recently reported the problem that mwm fails to
respond to keyboard or mouse events when num lock is on. Greg is
working on a source-level fix.
5. Features for new release
Our tex install is fairly broken and hard to fix. Greg will have a go
at replacing it with tetex within the next few days.
gnome libraries would be good.
6. More less?
Users are noting that less is native on Linux. Should we have it in
the release for other platforms? Greg said no (we should be
conservative about putting random useful utilities in the release) and
no one disagreed.
7. IRIX fam vulnerability
There is apparently a vulnerability in fam which allows remote
attackers to get names of files and directories but not contents. fam
is used by mediad, so we can't just disable it. We'll wait for a fix.
8. IRIX panic on core dump
IRIX kernel stack size limits can yield panics when programs run from
AFS dump core into a local filesystem directory. jweiss will send
mail to the infoagents locker maintainers asking them to limit
coredumpsize 0 in the IRIX netscape wrapper script to limit exposure.
9. attachandrun setuid problems
attachandrun is setuid (since it modifies the attachtab), which means
the program it runs loses a bunch of environmental state
(LD_LIBRARY_PATH, core dump size limits, etc.). To fix this problem
we will probably need to make attachandrun invoke attach in a
subprocess instead of being part of the attach suite. Greg will get
to this eventually.
