[OpenAFS-port-darwin] Recent changes to MacOS AFS client in 1.5 series

[Derrick Brashear]

As we've not necessarily trumpeted these changes, I felt it would be
prudent to share the current status and issues with the MacOS client.

Beginning in 1.5.69 (as well as in 1.4.12), we now play better with
Application Level Firewall (the firewall you can enable from the
Security panel in Preferences)

Since the same time, dmg installer bundles now include a dSYM version
of the kerbel module. This is to enable debugging after the fact for
crashes, either by you or if you've reported an issue, by us. The
Panic Decoder we include is now also installed in
/Library/OpenAFS/Tools/tools for your use. It can be used with a
kernel debug kit or with your running kernel.

New in 1.5.72 (and will be in 1.4.12), opening /afs when dynroot and
afsdb are both enabled should be much faster, nearly instant. Before,
we erroneously did ._cell lookups in DNS. No more...

Starting with the next (1.5.73) release, bulkstat will finally be
enabled. This means that opening a directory with lots of objects,
especially in Finder, will be faster.

In gerrit awaiting testing, you'll find support for allowing
cross-volume drag to move in Finder to work when it previously did
not. An experimental build of this code is available in
/afs/andrew.cmu.edu/usr/shadow/OpenAFS-1.5.72-Snowleopard+bulkstat+umove.dmg
Problem reports are welcome; RT ticket 34255 contains details. The
implementation uses a new userspace helper mechanism, so bugs may
still be present. Code can be reviewed in gerrit, notably changes 1371
and 1384.
If you'd like Apple to fix this correctly, open a problem report and
describe the error you get from finder; You can note it's a duplicate
of 5590394 (which has been pending, alas, since well before 10.6).

At this time there is no immediate hope of PAG support. RT 123563
explains the issues in play. A Directory Services plugin is the only
likely possibility at this point.

Finder's issues with what it notices as far as callback-generated or
authentication-state-change-caused changes (see RT 23781) cannot be
fixed with proper use of the KPI, as none of KNOTEs, vnode_notify or
fsevents are properly usable by us. (It may be possible to open
/dev/fsevents and write events to it and have them be redelivered if
they parse correctly, however, this is completely undocumented and
most likely not supported; an alternate mechanism, probably similarly
abusive but using only available and documented interfaces, may be
possible)

Also, versions 1.5.69 through 1.5.72 had an issue which tickled a bug
in coreservicesd and could result in Finder hanging or offering error
-10810. RT 124740 details the issue and mentions the rdar Apple is
tracking it with. 1.4.12 will not ship with this issue, and it's fixed
in 1.5.73.

That's the news for now.
-- 
Derrick