[2149] in Release_7.7_team
Minutes of 2000-02-23 meeting
daemon@ATHENA.MIT.EDU (Greg Hudson)
Wed Feb 23 16:14:28 2000
Date: Wed, 23 Feb 2000 16:14:17 -0500 (EST)
Message-Id: <200002232114.QAA28295@small-gods.mit.edu>
From: Greg Hudson <ghudson@MIT.EDU>
To: release-team@MIT.EDU
Attending: ghudson tb rbasch miki ajfox aurora zacheiss jweiss wdc
1. lastlog
Red Hat 6.1 has a /usr/bin/lastlog which is different from the consult
locker version. This is a bit different from a naming conflict like
gv or cvs (where it's the same program coming from two different
places, maybe with local improvements in the locker version).
tb suggested making add -f the default, since putting lockers at the
end of the path often violates user expectations. The group wasn't
comfortable making this change.
Resolutions:
* For Linux 8.3.28, add an attach-and-run script
/usr/athena/bin/lastlog pointing at the consult locker.
* For Athena 8.4, rewrite lastlog and bring it into the
release.
* To help address the general problem, strongly consider
adding a front-end to attachandrun called something like
"run" so you could do "run consult lastlog <username>".
Picking a name for "run" is a bit tricky because want it to
be easily memorable but also not likely to itself become the
subject of an OS name conflict.
2. mtools and Solaris
Abby want to know if Solaris 7 has native support for zip disks. miki
will find out.
3. IRIX
SGI delayed the 6.5.7 overlay release. This puts us behind. Bob will
test some things under 6.5.6 for now.
4. Solaris 7
miki has most of the pieces and needs to put them together. We'll
handle that in a smaller meeting.
5. Cluster info and updates
Greg has been looking into moving some logic out of getcluster and
into updates. He proposed a somewhat radical scheme for Linux which
would make the cluster info look like:
sysprefix /afs/dev.mit.edu/system/rhlinux
sysversion 8.3
where the control file would always be the same (control/versions,
say) and would contain all versions. getcluster would then no longer
have to do any kind of version selection from the clusterinfo records,
and update_ws would contain all the logic about which version to pick.
This scheme was rejected by the group, at least for now, partially
because it doesn't make any sense in the traditional release format
(where you can only have one set of system packs attached at a time,
and version information about a given full release is contained in
that release's system packs).
We agreed upon the following clarifications of the Linux update model:
* Control files will only contain versions for one full
release (e.g. control/versions-8.4 will only contain 8.4
versions). getcluster will still do version selection.
* On private workstations, update_ws uses "the old list" to
make some decisions about what packages changes to make. We
will store this old list on local disk rather than looking
up the old version in the control file. (That way we don't
have to worry about looking up an 8.3.x version when
upgrading to 8.4.)
* For manual upgrades to new full releases, update_ws will run
getcluster with the new full release number to find out
where the new list is.
The last point can also be implemented on Solaris and IRIX, as
discussed in athena-ws [2025].
6. Linux status
8.3.27 will go out soon. It will have most of the planned fixes
except for the "named messages in console" fix. A reactivate bug
prevents auto updates from happening.
Bill asked when Andrew is happening for Linux. The solution here will
probably be for Bill to unpack a binary release into the andrew
locker.
Greg's fancy new update program will wait until 8.4.
We need to think about layered release users who want to "get all
packages," since we now have multiple versions of some packages in the
packs. We will probably wind up keeping a list of the athena packages
somewhere. We also need to worry about people installed athena-ws
without knowing what it gets them into.
