[06:22:28] --- sxw has become available
[07:57:50] --- sxw has left
[08:42:00] --- Russ has become available
[09:39:05] --- geekosaur has left
[09:39:11] --- geekosaur has become available
[10:15:08] --- shadow@gmail.com/owlCDFD9865 has left
[11:37:20] --- deason has become available
[15:23:09] <kaduk@mit.edu/barnowl> I've been looking more at the (in)ability to stop and restart the FBSD
client.  Increasing the number of tries in the loop (to increase the
timeout) seems to make things more reliable for me (though I'm having a
bit of trouble reproducing the failure again).
However, the SS_ISDISCONNECTED check always fails, and in fact, the
so_state flags field has been zero when I've printed it (just during
shutdown).
Robert Watson tells me that in-kernel socket consumers generally must do
their own synchronization (which is commonly done by having separate
so_snd and so_rcv locks, and both are needed to close), and in fact there
are a couple of race conditions in the socket state code that can be
tickled if one does things that are not the documented usage.
Looking at some code, it seems that the flags are mostly (?) only used
for stateful protocols (i.e. TCP, SCTP) and not necessarily for
datagrams, though I'm less sure about that.
I see that Solaris' osi_FreeSocket loops while(rxk_ListenerPid) doing
(presumed bogus) calls to osi_NetSent until the Pid is gone; is that a
reasonable tack to take for FBSD?  Otherwise, I may try wrapping the send
and receives in SOCKBUF_{UN}LOCK(), but that seems like it might impact
performance.
[17:18:11] --- shadow@gmail.com/owlD253A924 has become available
[17:36:06] --- mdionne has become available
[17:52:21] <mdionne> current master fails to build here, for aklog executables I get: /usr/bin/ld: /s/openafs/git/lib/libafsauthent.a(cellconfig.o): undefined reference to symbol '__res_search@@GLIBC_2.2.5'
[17:54:35] <deason> try -lresolv ?
[17:54:41] <mdionne> adding XLIBS to the link line fixes it, but I'm not sure if that's the right fix
[17:54:44] <shadow@gmail.com/owlD253A924> hm. so how did we elide -lresolv.
[17:54:49] <mdionne> yeah resolv is missing
[17:54:57] <deason> XLIBS got replaced with MT_LIBS for the lwp->pthread switch
[17:56:10] <deason> the i386 linux buildbot slave for thatchange reports it as building, though
[17:58:02] <mdionne> I'm on amd64
[17:59:58] <shadow@gmail.com/owlD253A924> the thing is, MT_LIBS versus XLIBS is basically orthogonal.
[18:00:18] <deason> I would've expect both to require -lresolv, but I dunno
[18:00:59] <mdionne> from the buildbot log, it linked OK without -lresolv
[18:01:08] <shadow@gmail.com/owlD253A924> i'd actually argue that solaris MT_LIBS probably shouldn't have
-lsocket, and places which otherwise want XLIBS want XLIBS
[18:02:56] <shadow@gmail.com/owlD253A924> sadly i think my external disk with my solaris image must not be here. 
[18:04:45] <mdionne> also, change 3366 has the interesting effect that we now get more warnings with a regular configure than with --enable-checking, because CFLAGS_NOUNUSED is used in a few places
[19:26:44] <kaduk@mit.edu/barnowl> Derrick, could I persuade you to look at the logs for my wall of text
from 15:23?
[19:28:28] <kaduk@mit.edu/barnowl> FreeBSD 8.2 and 7.4 are both in the prerelease stage; there are changes
we care about between 8.1 and 8.2, but I don't think there were any
between 7.3 and 7.4.  Do we usually skip making a param.h for the 7.4
release in a case like this?
[19:29:00] <shadow@gmail.com/owlD253A924> in the past we've made a new one which matched the old one, basically
[19:29:13] <shadow@gmail.com/owlD253A924> added one more (set of) ifdefs for e.g. 7.4
[19:29:23] <shadow@gmail.com/owlD253A924> added a new value to the sysnames file for it
[19:29:28] <shadow@gmail.com/owlD253A924> and that's basically it
[19:29:32] <kaduk@mit.edu/barnowl> Okay.
[19:30:25] <kaduk@mit.edu/barnowl> (The guy who was getting the afsd segfault updated to RELENG_8 and as
such should have had an 8.2 param.h, but we don't have one there yet, so
I should commit those.)
[19:30:28] <shadow@gmail.com/owlD253A924> what else did you send? because i apparently screwed my router and, yeah
[19:30:47] <shadow@gmail.com/owlD253A924> well, it's fixed now. i suppose i could go find the log
[19:30:56] <kaduk@mit.edu/barnowl> About the sometimes inability to restart the FBSD client.
[19:31:03] <kaduk@mit.edu/barnowl> http://jabber.openafs.org/openafs@conference.openafs.org/2010-11-26.txt
[19:33:11] <shadow@gmail.com/owlD253A924> yeah, i found it. uh. the solaris way would not be unreasonable.
something else we could do is refcount in and out of e.g.
netsend/netreceive

but the solaris way is an easier start and probably lower impact
[20:10:32] --- mdionne has left
[21:45:40] --- deason has left
[22:40:40] <kaduk@mit.edu/barnowl> In osi_StopListener, we psignal rxk_Listener with SIGUSR1.  Do we install
a special handler for that signal, or just rely on the default handler
interrupting the syscall (or something else)?
[23:20:05] --- Russ has left: Disconnected