[00:54:35] --- Simon Wilkinson has become available
[01:39:06] --- Simon Wilkinson has left
[01:52:54] --- haba has become available
[02:50:08] --- kaj has become available
[02:55:27] --- kaj has left
[02:55:48] --- kaj has become available
[04:25:55] --- haba has left
[04:27:44] --- haba has become available
[05:35:31] --- reuteras has left
[06:04:51] --- steven.jenkins has left
[06:07:04] --- steven.jenkins has become available
[06:08:55] --- steven.jenkins has left
[06:10:17] --- steven.jenkins has become available
[06:13:30] <haba> andersk: Yes I watch here now and then. Thanks for the URLs, will try.
[06:46:01] --- meffie has become available
[06:58:27] --- jaltman has left: Disconnected
[06:59:40] --- steven.jenkins has left
[07:11:12] --- deason has become available
[07:18:06] --- steven.jenkins has become available
[07:55:11] <kaduk@mit.edu/barnowl> Hm, user running my freebsd packaging that pulls git head seems to be
reporting that the osi_vcache refactoring doesn't compile.  It's
unclear whether I'll get a chance to look at it until the weekend.
[07:57:45] --- jaltman has become available
[07:58:22] <deason> just the error messages would probably be helpful; it could just be typos that are obvious if we know what line
[08:28:39] <kaduk@mit.edu/barnowl> Derrick got the mail, but 
/usr/ports/net/openafs-devel/work/openafs/src/afs/FBSD/osi_vcache.c
/usr/ports/net/openafs-devel/work/openafs/src/afs/FBSD/osi_vcache.c: In function
'osi_TryEvictVCache':
/usr/ports/net/openafs-devel/work/openafs/src/afs/FBSD/osi_vcache.c:33: error: 'tvc'
undeclared (first use in this function)
[...]
/usr/ports/net/openafs-devel/work/openafs/src/afs/FBSD/osi_vcache.c: In function
'osi_AttachVnode':
/usr/ports/net/openafs-devel/work/openafs/src/afs/FBSD/osi_vcache.c:100: warning:
'return' with a value, in function returning void
[08:30:36] <shadow@gmail.com/owlB71D23BD> that's as of current i assume
[08:37:10] <deason> yeah, first thing is just s/tvc/avc/... second one though is a little less simple, since osi_AttachVnode doesn't have a way of returning errors yet
[08:38:53] <shadow@gmail.com/owlB71D23BD> i understand the code in question from playing with darwin and actually
was considering rewriting the FBSD osi_AttachVnode to eject that code
[08:40:50] <shadow@gmail.com/owlB71D23BD> in gerrit as 1904
[08:44:32] --- jaltman has left: Disconnected
[08:46:00] <deason> are you intending the osi_AttachVnode change in 1904 to just be temporary, then?
[08:46:29] <kaduk@mit.edu/barnowl> Derrick, do you want to reply to Jan or shall I?

I was probably going to just recommend putting a 'git checkout 1.5.74' in the port's
post-patch, since that should have all of the fbsd fixes before this breakage.
[08:46:36] <shadow@gmail.com/owlB71D23BD> well, it has the effect of working how the code already worked
[08:46:50] <shadow@gmail.com/owlB71D23BD> you can reply to jan. i have been sort of sad all morning.
[08:47:42] <shadow@gmail.com/owlB71D23BD> the thing is, we shouldn't ever lose the race, but the embryonic flag
exists: CVInit. eventually i would like a src/afs/BSD which *BSD and
DARWIN share code from for common code things.
[08:47:52] <shadow@gmail.com/owlB71D23BD> but for now, this will work exactly as it already worked
[08:48:07] <deason> er, it will? don't we now set the refcount to 1 when we 'lose the race'?
[08:48:34] <deason> since we're returning from osi_AttachVnode, not afs_NewVCache_int, as we were before
[08:48:43] <shadow@gmail.com/owlB71D23BD> did we not before? it looks like
[08:48:50] <shadow@gmail.com/owlB71D23BD> oh crap, this is an edited tree. hang on
[08:49:07] --- abo has left
[08:49:11] <deason> we didn't before, because that 'lose the race' branch returned the vc from afs_NewVCache_int, so it didn't do the other initialization stuff
[08:49:21] <shadow@gmail.com/owlB71D23BD> probably we should actually panic if we lost the race
[08:49:56] <shadow@gmail.com/owlB71D23BD> well. i'll push that and look at CVInit sometime in the next few days.
i just had my hands all over it for bulkstat on macos.
[08:50:04] --- abo has become available
[08:51:05] <kaduk@mit.edu/barnowl> Speaking of races, it looks like the two fbsd machines that have
experienced the hangs are both multi-processor, compared to Derrick's
single-processor (right?) test box.  I have been planning to sit down
with a debugger and a coredump for ... too long.
[08:51:17] --- Kevin Sumner has left
[08:51:41] <shadow@gmail.com/owlB71D23BD> yes. well. i'll be at bsdcan next week. perhaps i'll find time to do it
on the plane so i don't get mocked too roundly
[08:51:47] --- Kevin Sumner has become available
[08:51:52] --- haba has left
[08:53:49] --- haba has become available
[08:54:34] <shadow@gmail.com/owlB71D23BD> actually, it's easier than that
[08:55:35] --- kaj has left
[08:55:47] <shadow@gmail.com/owlB71D23BD> #elif !(defined (AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV))
        VREFCOUNT_SET(tvc, 1);  /* us */
[08:56:36] <shadow@gmail.com/owlB71D23BD> we define XBSD. so the refcount shouldn't be being set to 1. and if it
was, we could look at CVInit, which is already global. so that patch is
right
[08:57:39] <shadow@gmail.com/owlB71D23BD> ben, if you want to verify i will push
[09:00:54] <deason> ah, okay, yeah; BSD doesn't do anything in that post-init section, you were right
[09:01:23] <shadow@gmail.com/owlB71D23BD> well, i didn't think i edited my tree there in any way to matter. i
need to manage sandboxes better
[09:01:29] <deason> er, no, wait, there is some stuff outside of all the ifdefs, but I don't know if it matters
[09:01:31] <shadow@gmail.com/owlB71D23BD> though i seem to have fewer than simon
[09:01:45] <shadow@gmail.com/owlB71D23BD> it doesn't appear to set the version (the postprocess stuff)
[09:01:49] <deason> 'tvc->dchint = NULL' and below in the original NewVCache_int
[09:01:55] <deason> yeah, it doesn't
[09:02:18] <shadow@gmail.com/owlB71D23BD> we can certainly bail on that easily.
[09:02:29] <shadow@gmail.com/owlB71D23BD> blowing the dchint is ~harmless
[09:03:07] <deason> we used to do this:
    tvc->dchint = NULL;
    osi_dnlc_purgedp(tvc);      /* this may be overkill */
    memset(&(tvc->callsort), 0, sizeof(struct afs_q));
    tvc->slocks = NULL;
    tvc->f.states &=~ CVInit;
    if (seq) {
        tvc->f.states |= CBulkFetching;
        tvc->f.m.Length = seq;
    }
    afs_osi_Wakeup(&tvc->f.states);
[09:03:36] <deason> we appear to not do all of that post-attachvnode, but, well, I'm only skimming the code
[09:03:45] <shadow@gmail.com/owlB71D23BD> slocks might matter. vSetType(VREG) in the FBSD specific pieve will
matter.
[09:03:59] <shadow@gmail.com/owlB71D23BD> we do do all of it
[09:04:44] <deason> oh, yeah, I'm confusing afs_PostPopulateVCache with osi_PostPopulateVCache, agh
[09:05:04] <deason> afs_PostPopulateVCache does do most of that, and osi_* does the other stuff
[09:05:15] --- Kevin Sumner has left
[09:05:33] --- Kevin Sumner has become available
[09:05:53] <shadow@gmail.com/owlB71D23BD> easy to fix tho. don't PostPopulate if not CVInit
[09:06:35] <shadow@gmail.com/owlB71D23BD> it means we won't set a bulkstating flag. that's fine, if this is
changing under us, we probably shouldn't
[09:14:41] <kaduk@mit.edu/barnowl> I'll probably not get to it until late this afternoon.
[09:18:26] --- deason has left
[09:19:52] --- deason has become available
[09:29:14] --- phalenor has become available
[09:55:54] --- haba has left
[09:57:29] --- jaltman has become available
[09:58:54] --- jaltman has left: Replaced by new connection
[09:58:55] --- jaltman has become available
[10:23:57] --- meffie has left
[10:45:03] --- jaltman has left: Disconnected
[10:45:12] --- Russ has become available
[12:00:56] --- jaltman has become available
[12:59:09] --- meffie has become available
[13:35:31] --- mdionne has become available
[14:09:00] --- Simon Wilkinson has become available
[14:13:35] <Simon Wilkinson> Is CVInit only FSBD?
[14:14:11] <shadow@gmail.com/owlB71D23BD> no
[14:14:13] <shadow@gmail.com/owlB71D23BD> global
[14:14:43] <Simon Wilkinson> Just wondering if that change (to not postprocess) should only be on FBSD, or it it can be global.
[14:14:56] <shadow@gmail.com/owlB71D23BD> it can be global.
[14:14:56] <Simon Wilkinson> Trying to avoid introducing new #ifdefs into the NewVCache code path.
[14:15:37] <Simon Wilkinson> Is it currently FBSD only just to reduce the risk of the patch?
[14:17:09] <shadow@gmail.com/owlB71D23BD> basically.
[14:40:09] --- meffie has left
[14:47:01] --- Kevin Sumner has left
[15:04:58] --- Simon Wilkinson has left
[15:34:45] <kaduk@mit.edu/barnowl> "FBSD only kind of works anyway, so it can be a guinea pig"
[16:01:09] --- deason has left
[16:11:20] <Russ> Just uploaded 1.4.12+dfsg-4 to Debian fixing various problems if one mounts AFS somewhere other than /afs and incorporating all the Linux-relevant patches on the stable branch since the previous upload.
[16:11:33] <Russ> working on 1.5.74.1 packages now.
[16:18:35] --- deason has become available
[16:27:07] <Russ> Bleh, and there was a bug in that, so 1.4.12+dfsg-5 is coming in just a second.
[16:27:59] <andersk@mit.edu/dr-wily> Oh, what kind of a bug?  Should I cancel the PPA build?
[16:49:18] <Russ> If you enable kill_all_afs but you stop AFS when it wasn't started, it would kill every process on the system.  Doh.
[16:49:30] <Russ> Not a problem that's enabled by default, though.
[16:49:40] <Russ> 1.4.12+dfsg-5 uploaded.
[16:50:39] <Russ> And tagged and pushed to Git.
[16:55:32] <Russ> This is all so much easier and nicer with 1.5 now.
[16:55:39] <Russ> I can hardly wait until it's the stable release.
[18:15:20] --- nelhage@mit.edu/lunatique has left
[18:16:42] --- nelhage@mit.edu/lunatique has become available
[19:05:43] --- mdionne has left
[19:36:05] --- abo has left
[19:36:45] --- abo has become available
[20:10:43] --- Born Fool has become available
[20:46:11] --- jaltman has left: Replaced by new connection
[20:46:12] --- jaltman has become available
[21:16:40] --- Born Fool has left
[22:11:21] --- reuteras has become available
[22:34:06] <Russ> 1.5.74.1-1 packages uploaded to Debian experimental, although they'll have to go through NEW since I broke out the FUSE package.  Will tag in just a second.
[22:35:54] --- deason has left
[22:35:59] --- kaj has become available
[23:26:08] --- kaj has left