Re: [Pan-devel] Please test pan if you are able to prepare for the next release.

[Duncan]

Heinrich Müller posted on Sun, 08 Jan 2012 13:51:29 +0000 as excerpted:

> Thanks for any people willing to participate. Leave bugs at bugzilla or
> here.

It's working here, but I've not had time to try secure connections to 
gmane again due to tracking bugs elsewhere and a just completed full 
system rebuild (of course bringing out more bugs to track down) after 
upgrading to gcc-4.6.2 (from 4.5.x.  But it's building, including the 
security stuff (and without notifications enabled, now that you fixed 
/that/ bug) and running at least in unencrypted mode.


I really hate to bring it up again as it's a long-term problem, since the 
Charles era, that I've been unable to trace down, but somehow, some way, 
either gmane or pan has a problem with certain cross-posted articles if 
pan has visited/subscribed both affected groups, where pan is simply 
unable to see new messages on one of the groups.

The problem went away for a period, I'd guess due to the down time that 
gmane had a few months ago, then came back, but only on SOME of the 
previously affected groups.

In particular, I'm subscribed to two gmane.comp.freedesktop.xorg.* 
groups, announce and drivers.ati (which is for the free drivers, not the 
proprietary ones).  Obviously driver version announcements go to both.  
For some time, that caused issues, but they went away and and haven't 
come back.

In contrast, as a gentooer I'm subscribed to several gmane.linux.gentoo.* 
lists, including devel, devel.announce, and project.  I was also 
subscribed to council for awhile, but that list/group folded into project 
and is no longer active.  devel is the most active and the oldest, with 
me being subscribed there since 2004 I think.  devel.announce, council 
and project are newer, and at least devel.announce and project had this 
bug at one point due to crossposts probably with devel.  Then the problem 
went away, and stayed away with devel.announce, but project won't update, 
again.

The bug has to have something to do with the seen messages and/or high-
water-mark tracking in relation to crossposting, but that's pretty much 
as far as I've gotten, despite occasional failed attempts to bisect the 
problem down to a specific user data file in $PAN_HOME -- it seems to be 
an interrelationship of more than one, such that the problem can't be 
isolated to a single file.

What seems to be happening is that either gmane is giving pan the message 
sequence numbers for the wrong group, or pan is using the sequence 
numbers from the wrong group when a message is cross-posted.  Either way, 
once pan sees the higher sequence numbers for the more active or longer 
history group, it believes it has seen everything the server has to offer 
and won't even attempt to fetch headers for that group any longer, as it 
believes it's ahead of what the server is saying it has.

I've been going to try a test with a clean test config, then subscribe to 
JUST the devel and project groups (and download all headers), with devel 
first, then wipe the config and try project first, and see if that makes 
a difference, but I've not done so yet.

Anyway, this can't be release-critical as it has existed for years and 
years, thru multiple releases.  But it sure is frustrating!  If you could 
either give me specific instructions for running gdb to try to catch 
this, or supply a patch that would spit out diagnostics information about 
it to stderr, I'd love to track this down, but I know the description is 
rather vague for that.  The problem is, I've been frustrated in all 
attempts so far to narrow it down further.

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman