Check out the /topic on #centos :
12:53 -!- Topic for #centos: DO NOT ASK ABOUT CentOS 5.4 | Yum problems?
try:
yum clean all | DO NOT PASTE IN HERE (unless asked; 1 line MAX),
use
http://pastebin.centos.org/ | More on /topic
http://centos.org/irc |
Working on getting our wiki view into shape tonight. And I realized
something.
We use locate_message_by_uid() to map a UUID/EUID to a message number ...
this function originally came from our GroupDAV implementation, which
needs similar functionality, because we use EUID's as the
Note to developers: the Wiki view isn't actually disabled but until we
finish it, we're simply disabling (and hiding) the ability to create Wiki
rooms. If your system *has* Wiki rooms, they will work.
So if you want to experiment with the Wiki view:
1. Edit roomops.c and find the
Why does wiki have to refer to byzantine markup syntax ?? I think
the nature of a wiki isn't that it requires its maintainers to learn wiki
markup, but that it can be maintained and edited from within itself.
I suppose we can solicit a few more opinions on this, but so far I'm not
Anyone else having trouble getting the mailbox view to render (at all) in
the current svn head? I had to backport my OpenID patches to an older rev
because the current one wouldn't run properly on Uncensored.
will that produce an expected behaviour?
Evidently not. Calendar and wiki views are both broken now. Wiki shows
plain text with no links; calendar shows nothing at all.
Uncensored was updated to svn head last night and everything seems to be
working. I needed to make distclean to get all of the mime type
registrations into the build. Can you make a note in the commit log when
you check in something that requires rebootstrapping please? Thanks..
davew!! where you been!!!
At the moment I'm happy just to see you lurking about.
Hmm. I wonder if now is the time to take that Big Bloated Step (tm) we've
been contemplating for a few years, and bind a scripting engine into the
system?
Note: wiki rooms are *enabled* in the current build. If we do any new
releases before this module is complete we will have to remember to
disable them again.
What were you thinking of using?
Perl, PHP, something else, something home grown?
JavaScript. It would definitely be JavaScript.
Mozilla's JS engine (I think it's called SpiderMonkey) is designed to be
embeddable, and could be bolted onto Citadel with a relatively
Right, I'll look into it.
Oh ... ok, I wasn't expecting that, but if so, here's where to start:
http://www.mozilla.org/js/spidermonkey/
Read their embedder's guide. It appears to be quite clean. Most of
our work would involve writing all of the Citadel API bindings.
Well, there's also the V8 JavaScript engine, which is what Chrome uses,
and is faster:
http://code.google.com/apis/v8/embed.html
You guys are doing a lot of sweeping changes here. I hope you're testing
extensively. Not just does it compile but really putting the system
through its paces.
Whilst I've been poking about I have noticed some messages (generally in
gdb)
that seem to indicate that some library that citserver currently relies
on
also requires libnspr. Is there a command that can be run on an
executable
and will print out all the library dependencies and
It works!! It works!!!
It uses it for everything, but it's optional for a single-threaded parser?
How's that?
Are there any other JS parsers out there that we should be looking at?
Perhaps the KDE or GNOME projects have their own?
I really don't want to add too many external dependencies, and I
Oh, here's a good starting point:
http://en.wikipedia.org/wiki/List_of_JavaScript_engines
Anyway, the only reason I was erring on the side of SpiderMonkey is
because
most people seem to have it in their distro already whereas V8 doesn't
seem
to be so prevelent yet.
The other issue with V8 is that it's a JIT engine, which restricts the OS
and CPU choices of the
My test server now spits out Server strangled due to machine load
average too high once per second.
Isn't that how greylisting is supposed to work?
If I fill in the subject and then hit the RETURN key it posts the
message
which is usally empty.
Aha! I was wondering why there have been so many blank posts lately.
Yes, we have to fix that.
Re. contexts -- I'm not sure what exactly that would involve, but I
definitely would like to get a new release of Citadel finished before we
make any dangerous changes.
Let's start thinking about Citadel 7.70 and what it will take to get it
finished. I already hit my target (wiki rooms)
The full text index is of course active for Wiki rooms. We just have to
write some code to display the search results in a sensible way.
Fleeb came up with a brilliant idea: we should use Wiki rooms for our
internal bug tracking. That way we can annotate stuff as much as we want.
And
Hmm, I was thinking more about a list of every page in the wiki so I can just
browse it instead of having to search on some random keyword that may or may
not be included in every page.
Sure, why not? For search, we could use the very same renderer with a
search-reduced list of pages.
As
* identify the room we're in by REST; first draft.
Ummm ... isn't DAV already REST by its very nature?
/groupdav/roomname/...
similar casts that my 64bit box complains about, don't want to mess with
your
work in progress :-)
It's more or less finished. Tune it if you need to. But of course test
test test afterwards.
kinetix:
Forgive me if I'm late to the game and you've already tried this, but ...
Have you looked at the Easy Install script to analyze what it's doing?
It contains all of the necessary flags for building a private copy of
Citadel with private copies of its libraries.
kinetix: if the method used by Easy Install doesn't work for you, then I'm
not sure what else you could try. It took a while for me to figure out
that method, and it seems to have worked for countless thousands of Easy
Install users.
I suppose your other option would be to do your test
Hey, nothing ventured nothing gained, right? Enjoy your development
platform.
Ok, all done
Hmm, was there some sort of change made to the RSS module other than the initial Atom support? Some of our feeds reloaded and posted a lot of duplicates.
I most probably could use some hints on the duplicate detection,
since it seems as if it doesn't work with the atom of freshmeat..
It's quite simple: if there is a guid field present, we prepend rss/
to it, and hit the use table to determine whether it's a duplicate. If
Ok, Uncensored just crashed because it hit the abort() at threads.c:781.
Why is there an abort() in our code?
..and it crashed again. What's going on here? What changed?
Can we change it back? And how did you get that nifty view with the rev
numbers and names?
Ok, I'm going to revert r8058 and we'll see if that fixes it.
in general blocking on shut down for blocked tcp/ip connections (for
example) is another nail to the coffin for creating reliable backup
strategies.
Here's a reliable backup strategy: if it doesn't shut down within 10-15
seconds, kill -9 the server. We have a journaling database
If one adds an RSS feed, do we catch that? My impression was, that it
needs a server restart to make it collected?
No server restart required, but you still have to wait until the next
time it runs. The RSS aggregator and the POP3 aggregator are started in
the same way.
Y'all are thinking too much. The original thread manager was deliberately
made simple, almost naive. I don't see any need to ever reduce the size
of the thread pool, and I certainly don't see any need to gracefully shut
down anything other than the database. Quite frankly, I think we should
How unrecognizable is the dav server going to be when I eventually get
around to implementing CalDAV?
Ok, well you can do all the unit tests you want; the real test is whether
Kontact still works.
Our URL parser handles either one, but the '' gets interpreted by Tidy as
an error.
Yes, we definitely need to get something out soon.
I'd say that's a feature, not a bug. And yes, it should do that, as long
as the feed provider is using the same uuid's for both formats. Already
seen is already seen. I think that's great, because it means you can
switch feed formats without seeing duplicates.
If you want to run
CommonJS effort sets _javascript_ on path for world domination
CommonJS is a grassroots campaign to quickly produce standards and a standard library for _javascript_. Our man on the inside reports on JSConf Europe, where the effort to produce interoperable _javascript_ embeddings was evident.
would it be reasonable to output the number of messages in the smtp
outbout queue via the mrtg command?
I don't see why not. MRTG uses a very simple data format.
Hey everyone,
Just wanted to mention that I'm sorry I haven't been immersed in Citadel
development over the last couple of weeks. I'm currently facing the
double whammy of both being overwhelmingly busy at work *and* suddenly
dealing with some personal issues (read elsewhere on
Uncensored is now running on the stable-77x branch. Assuming everything still looks good tomorrow morning, I will push this branch out to Easy Install and build some tarballs too.
I don't want this to be a long-lived branch for anything other than Debian stable. The last time we lived too long
Yes, we should continue to make use of release branches. I just don't want them to live *too* long. We'll keep stable-77x around for backports of bug fixes if that's the release that makes it into Debian stable, of course.
I think what I'm trying to say here is that we should never let svn
Sounds very cool, but is this going to be one of those things that only
davew understands and therefore becomes difficult to maintain?
Easy Install has been updated to stable-77x. Tarballs to follow, after
we let a few people act as guinea pigs :)
I wonder if I should strip the new bbsview code out of stable-77x, since
it's not going to get used anyway?
ok, done; I hope no more old messages make it over? ;-)
Yeah, I don't know what happened there.
Should we put this code in stable too?
OK, now that I've been out of this for so long, where are the svn repos
at?
svn+ssh://er...@uncensored.citadel.org/appl/svn/
I guess you can email me the details if there's a security problem. If
it's really something that can't be fixed in a compatible way, then we'll
change the command name. Old clients will fall back to unencrypted
connections.
Well folks, I think I finally figured out the disappearing netconfigs
issue.
It happens when you run out of disk space.
That happened to me a couple of days ago. All gone. :(
at least it should send an aide message in that case?
If the disk is full, the database isn't going to get written either.
Actually, this shows how good Berkeley DB is. Perhaps some records
didn't get written, but the database did not get corrupted. If the disk
wasn't full,
r8154 fixes the Slashdot feed (which is really a disgusting mess of xml
that should be taken out back and shot)
Is that the whole change?
I guess private locales are legal, but I don't see the point.
Cool. Thank you.
Pushing an update out to Easy Install now. (I will continue to do this at
any time during the stable-77x series, under the assumption that
everything committed to that branch is indeed stable.)
Give it a try. I think it's pretty nice!
davew: any reason for this? If we have libgc support at all it's probably more sensible to manually activate it...
Fix race condition that caused segfaults in imap and xmpp as seen on
uncensored.
Ok, I'm gonna put that into production.
The Add, change, delete user accounts link in WebCit is broken. It displays
an
error page didn't find Template [edituser_select] 15 15.
Can this be fixed easily ... and how many more of these errors might be
lurking?
ups, that was moved to aide/edituser/select
i'll fix it.
Ok, the screen appears now, but the user list in it is empty.
r8201 *completely* broke the networker.
Huzzah indeed. Considering that it would fail 100% of the time, I'm certain
that
it wasn't tested.
And y'all know how I feel about checking in code without TESTING IT!!!
Ok, so that explains why the mailing list digests were broken. I've gone through serv_network.c and fixed all the places where the return value from fread() was handled as if it was the number of bytes read instead of the number of blocks read.
Was this type of change made anywhere other than
For what it's worth, this issue is getting exposed by a test server that is
constantly getting hammered with an extremely large volume of spam, so there is
a
large incoming SMTP load.
I'm showing it on a 32 bit machine. Do you want a core dump and log sample?
Last coupla lines of logs...
2010/01/21 2:18:20.198476 New client socket 33
2010/01/21 2:18:20.228583 [36312] SMTP server: QUIT
2010/01/21 2:18:20.229020 Purging session 36312
2010/01/21 2:18:20.229089 RemoveContext() session 36312
2010/01/21 2:18:20.229116 [36312] xmpp_queue_event(1, )
Now would be the time to start playing, yes ... but do keep in mind that
Uncensored is running on HEAD, so don't check in anything that's known to be
outright broken.
I have the ability to slam a server with connections, through the use of an
abandoned domain name that gets a TON of nonstop spam. I'll give it a try.
Sorry dude:
2010/01/30 12:48:31.967497 Closing socket 60
2010/01/30 12:48:31.967520 Done with RemoveContext()
2010/01/30 12:48:32.002170 Interrupted CtdlThreadSelect.
2010/01/30 12:48:32.002234 Thread Worker Thread caught signal 1.
2010/01/30 12:48:32.002284 Thread Worker Thread (0xa8ec4b90)
davew: if you want, I can point this mega-spam domain's MX record at the server
of your choice so you can work on the thread issue. It usually only takes a
little while to make the problem come up.
Not a problem, let me know when you're ready. And don't worry, it won't
saturate
a 2M link because most of the spam is for accounts that don't exist (actually,
all of it, since I'm not actually running that company, but I did activate a
couple of addresses such as info@ so a little of the mail
I would say that it should be decoded by libcitadel when the vcard is
deserialized into our in-memory data structure.
It's about time. So we do have to figure out what code we have in our framework
that exists solely to support obsolete browsers. That'll be the challenge.
(Matt? Thierry? Anyone know for sure?)
davew: yes, it's been running very nicely. I think you got it this time. We
will continue to observe its operation for a while.
* use mmap to read the download file for output; this way we don't need to
copy
it into memory first and can let the kernel do this job
Does mmap() read the whole file into memory at the kernel level? Or does it
merely provide userspace with the illusion that this has happened?
Fair enough. Is it portable?
it seems as if the new bbsview fails to mark everything read?Â
behaves a little funny to me over here.
Same as before. Messages are marked as read when you Goto out of the room.
New buffered I/O code crash:
#0 0x080594f2 in client_write (buf=0xb6dc4df0 250 Message accepted.\r\n,
nbytes=23) at sysdep.c:488
488 FD_SET(Ctx-client_socket, wset);
(gdb) bt
#0 0x080594f2 in client_write (buf=0xb6dc4df0 250 Message accepted.\r\n,
nbytes=23) at
#0 0x08058bbb in worker_thread (arg=0x0) at sysdep.c:1083
1083if ( (FD_ISSET(ptr-client_socket, readfds))
(gdb) bt
#0 0x08058bbb in worker_thread (arg=0x0) at sysdep.c:1083
#1 0x08077cbf in ctdl_internal_thread_func (arg=0x9fd08a0) at threads.c:839
#2 0xb7e5f50f in
Using SVN, or doing anything else, on Windows is SO VERY WRONG.
BIG PROBLEM!
Sometime recently, something broke in the main server context loop.
Asynchronous events are not firing. This is a big problem for the XMPP service,
which makes heavy use of its async loop in order to determine whether there are
any events in the queue which require action
I made several assumptions on the types of contexts that could be selected on
and which ones to check the fd of after select. Maybe these assumptions are
flawed.
What exactly happens to the context when an async message is sent?
Check out modules/xmpp/xmpp_queue.c arolines
Fix async messages (hopefully).
Nope.
If you send a message from Cit client to pidgin the message doesn't arrive.
Got that bit.
Question. Does the message from Cit client arrive when pidgin sends a new
message to Cit client?
It should, at least it did when I tested it.
Pidgin should also receive
As soon as my pidgin sends to the citserver the citserver dives off into an
endless loop in dothebarts new buffered IO code.
Now thats a different problem that causes me issues instead.
I didn't see that, but perhaps I'll try it again.
By the way, Citadel clients *poll*
Ok, it seems that the buffered I/O *is* causing a problem.
sysdep.c : 1147
HaveMoreLinesWaiting() *always* returns 1, until the session ends. As a
result,
we never fall through the loop until the session is ending, which explains why
we
never get around to processing our async
When pidgin sends to webcit the message is slightly corrupted. Seems the last
char is dropped and probably the terminating NUL too.
I have identified this as a bug in memfmout(). When I use cprintf() it goes
away.
I don't know if this is related to characters missing in messages, but I've
been
noticing that room names being deleted by the DAP are having the last
character
truncated as well in the Aide room posts.
This could quite possibly be a memfmout() or fmout() issue as well!
client connections are timed out based on lastcmd which means they never time
out because every client sends NOOP
Is this correct / intentional?
That is correct. It's not intended to log off idle humans; it's intended to
kill off stale connections.
let me take an educated guess... It fails to output a \0 if there is
a string without newline?
so aoeu\0aoeu wouldtrigger the issue?
That sounds about right. We are definitely giving it strings without a
trailing
newline.
hmm ... that didn't fix it. The problem is somewhere else. :(
which fixes the saving of messages via imap.
messages copied in so far are b0ken, their unparsed headers are glued
together without newlines :-(
I think the reason I thought there was more to it than that was because it had
gone unnoticed for a long time.
By the way, it
801 - 900 of 1678 matches
Mail list logo