Re: [Server-devel] ejabberd-2.1.10 issues

2012-02-09 Thread Martin Langhoff
On Thu, Feb 9, 2012 at 3:46 PM, Daniel Drake  wrote:
> As you've seen - getting this old version to build is difficult.
> Getting it to run is even harder.

Yep. Was just a stab in the alternative path. Now that you've
root-cause'd the issue, things look better.

> I think I've found the issue: it is a client bug, now exposed through
> ejabberd getting stricter:
> https://support.process-one.net/browse/EJAB-1533
> https://bugs.freedesktop.org/show_bug.cgi?id=45853
>
> And I have an ejabberd patch to apply (attached) until we've fixed the

Right, so you have a partial revert of the "strict" patch at EJAB-1434
- you plan sounds good.

> Would you prefer for this fork to be named 'ejabberd' (with xs-setup
> modifying the EPEL repo file to exclude ejabberd upgrades from epel),
> or ejabberd-xs as before?

This merits a 'release' identifier, but we're only carrying 1 patch
over CentOS, and done the work to split the init script and config off
to xs-config, so keeping the ejabberd name makes sense.




m
-- 
 mar...@laptop.org -- Software Architect - OLPC
 - ask interesting questions
 - don't get distracted with shiny stuff  - working code first
 - http://wiki.laptop.org/go/User:Martinlanghoff
___
Server-devel mailing list
Server-devel@lists.laptop.org
http://lists.laptop.org/listinfo/server-devel


Re: [Server-devel] ejabberd-2.1.10 issues

2012-02-09 Thread Daniel Drake
On Wed, Feb 8, 2012 at 6:42 PM, Martin Langhoff  wrote:
> Ok. Following the breadcrumbe here, I get to
> https://support.process-one.net/browse/EJAB-919 which has a nice patch
> authored by the Geoff Cant, a fellow Catalyst-er. I can help w a
> rebuild here.

As you've seen - getting this old version to build is difficult.
Getting it to run is even harder.

I think I've found the issue: it is a client bug, now exposed through
ejabberd getting stricter:
https://support.process-one.net/browse/EJAB-1533
https://bugs.freedesktop.org/show_bug.cgi?id=45853

And I have an ejabberd patch to apply (attached) until we've fixed the
client and let the fix propogate for a while.  We need to fork the
ejabberd package again, at least for the time being.
Would you prefer for this fork to be named 'ejabberd' (with xs-setup
modifying the EPEL repo file to exclude ejabberd upgrades from epel),
or ejabberd-xs as before?

Thanks
Daniel


persist-all-pubsub.patch
Description: Binary data
___
Server-devel mailing list
Server-devel@lists.laptop.org
http://lists.laptop.org/listinfo/server-devel


Re: [Server-devel] ejabberd-2.1.10 issues

2012-02-08 Thread rihowa...@gmail.com

On Feb 8, 2012, at 2:40 PM, Daniel Drake wrote:

> Hi,
> 
> With the XS rebase to CentOS6.2, EPEL-6 brought us ejabberd-2.1.8.
> We found an issue that the Online group cannot be marked as @online@
> correctly. Upgrading to ejabberd-2.1.10 (from EPEL6-testing) fixes
> this.
> 
> However, there is a regression with collaboration: presence doesn't work 
> right.
> 
> I tried downgrading to the old ejabberd-xs package, but it needs a
> rebuild, and after some hassle of getting it to build, it crashes on
> launch. Looks like it is not compatible with the new erlang in EL6:
> http://www.ejabberd.im/node/3698
> 
> So, I think we have to look into the ejabberd-2.1.10 issue.
If I remember correctly the XS's  ejabberd.cfg calls mod_ctlextra .
>From ejabberd 2.1 and greater this has been replaced by mod_admin_extra. 
See http://www.ejabberd.im/mod_ctlextra and 
http://www.ejabberd.im/mod_admin_extra

I also remember reading a doc saying the default node names have been changed.

Robert H

> 
> The issue in question: When an XO connects to the network, it does not
> "see" any buddies that are already on the network. However, when new
> buddies join, it will see them.
> 
> With 2 XOs, A and B:
> 
> A sees B
> B does not see A
> - restart B
> A sees B
> B does not see A
> - restart A
> A does not see B
> B sees A
> 
> In the cases when an XO connects to the network with existing clients,
> the XO *is* told about the existence of the other clients, but it
> receives an empty dictionary from the OLPC BuddyInfo telepathy
> interface:
> http://cgit.freedesktop.org/telepathy/telepathy-gabble/tree/extensions/OLPC_Buddy_Info.xml
> 
> The Sugar code specifically waits for the 'key' to arrive before
> displaying the icon for the buddy (I guess the key is used for
> collaboration), hence the buddy does not appear.
> 
> I'll keep looking into this; debugging tips welcome.
> 
> Thanks,
> Daniel
> ___
> Server-devel mailing list
> Server-devel@lists.laptop.org
> http://lists.laptop.org/listinfo/server-devel

___
Server-devel mailing list
Server-devel@lists.laptop.org
http://lists.laptop.org/listinfo/server-devel


Re: [Server-devel] ejabberd-2.1.10 issues

2012-02-08 Thread Martin Langhoff
On Wed, Feb 8, 2012 at 5:40 PM, Daniel Drake  wrote:
> With the XS rebase to CentOS6.2, EPEL-6 brought us ejabberd-2.1.8.
> We found an issue that the Online group cannot be marked as @online@
> correctly. Upgrading to ejabberd-2.1.10 (from EPEL6-testing) fixes
> this.

That's good news. As I mentioned on irc, debugging string handling
buglets in Erlang code wasn't going to be easy :-)

> However, there is a regression with collaboration: presence doesn't work 
> right.

Argh. The behaviour you describe has been seen before... in reverse!
The classical problem was pubside event propagation that let to the
opposite behaviour.

> I tried downgrading to the old ejabberd-xs package, but it needs a
> rebuild, and after some hassle of getting it to build, it crashes on
> launch. Looks like it is not compatible with the new erlang in EL6:
> http://www.ejabberd.im/node/3698

Ok. Following the breadcrumbe here, I get to
https://support.process-one.net/browse/EJAB-919 which has a nice patch
authored by the Geoff Cant, a fellow Catalyst-er. I can help w a
rebuild here.

I would also suggest getting the Sugar factor out of the way, testing
briefly w 3 laptops with OS 10.1.3. OS 10.1.3 is known to DTRT. The OS
11.x.y series has a number of collaboration regressions, and while I
don't think it hits this problem interoperating with OS 0.6, it's
worth a check.

cheers,



m
-- 
 mar...@laptop.org -- Software Architect - OLPC
 - ask interesting questions
 - don't get distracted with shiny stuff  - working code first
 - http://wiki.laptop.org/go/User:Martinlanghoff
___
Server-devel mailing list
Server-devel@lists.laptop.org
http://lists.laptop.org/listinfo/server-devel


[Server-devel] ejabberd-2.1.10 issues

2012-02-08 Thread Daniel Drake
Hi,

With the XS rebase to CentOS6.2, EPEL-6 brought us ejabberd-2.1.8.
We found an issue that the Online group cannot be marked as @online@
correctly. Upgrading to ejabberd-2.1.10 (from EPEL6-testing) fixes
this.

However, there is a regression with collaboration: presence doesn't work right.

I tried downgrading to the old ejabberd-xs package, but it needs a
rebuild, and after some hassle of getting it to build, it crashes on
launch. Looks like it is not compatible with the new erlang in EL6:
http://www.ejabberd.im/node/3698

So, I think we have to look into the ejabberd-2.1.10 issue.

The issue in question: When an XO connects to the network, it does not
"see" any buddies that are already on the network. However, when new
buddies join, it will see them.

With 2 XOs, A and B:

A sees B
B does not see A
- restart B
A sees B
B does not see A
- restart A
A does not see B
B sees A

In the cases when an XO connects to the network with existing clients,
the XO *is* told about the existence of the other clients, but it
receives an empty dictionary from the OLPC BuddyInfo telepathy
interface:
http://cgit.freedesktop.org/telepathy/telepathy-gabble/tree/extensions/OLPC_Buddy_Info.xml

The Sugar code specifically waits for the 'key' to arrive before
displaying the icon for the buddy (I guess the key is used for
collaboration), hence the buddy does not appear.

I'll keep looking into this; debugging tips welcome.

Thanks,
Daniel
___
Server-devel mailing list
Server-devel@lists.laptop.org
http://lists.laptop.org/listinfo/server-devel