[Pan-users] Re: New version of pan ?

[Duncan]

K. Haley posted on Tue, 25 Jan 2011 14:21:49 -0700 as excerpted:

> While there's no ui, the ability to score on any header has been in for
> some time now.  As for scoring on the body....

Learned something new!  Thanks! =:^)

>> There is one very frustrating but equally long-standing bug having to
>> do with pan not seeing new posts in groups just subscribed (or freshly
>> visited without subscribing) if there are cross-posts between it and a
>> long-term subscribed group, that I'd love to have fixed before 1.0, but
>> I've not been able to sufficiently narrow it down or describe it well
>> enough to really form a good bug report, and as the bug has persisted
>> for years, I don't see that it's practical to hold up 1.0 for it,
>> either.

> What could be happening is that marking a cross-posted article as read
> updates the read articles in all relevant groups.  This could make pan
> think it has already retrieved all headers up to that article.  If so
> then a work around would be to request the last x or x days headers
> instead of new.

I think it must be related to that, yes, but the bug is worse than that 
and tougher to work around.  I've tried getting the last N day's articles, 
the last N headers... nothing seems to work.  If I hadn't verified that a 
clean start worked, I'd be sure the groups were empty.  I've also fiddled 
with the files, resetting the newsrcs, and that alone doesn't seem to fix 
it either, tho of course it resets the unread count on my long-standing 
groups.  It seems to be some weird interaction between the newsrc and the 
group info storage mechanism (IIRC I verified that the message cache 
itself is NOT involved), such that resetting only the newsrc or the other 
group data doesn't cure the problem, only resetting both does.  Which is 
what makes it so frustrating and so difficult to figure out.

I've decided there's only two possibilities for tracking it down, either 
compiling with debugging symbols and running a gdb session on it (I've 
only run a gdb session once, and that was rote following of someone's 
specific step-by-step, so this would take some time and be a challenge), 
or specific and repeated trial and error editing of both the newsrc and 
other files until I find /something/ isolatable there, which will again 
take a sizable chunk of time and quite some patience.

The problem is that it can't be just any time.  As I've gotten a bit older 
I've discovered I can't just pull all-nighters (or all-dayers) and expect 
to work in "new" territory.  If I'm reasonably familiar with the domain 
and it's mostly rehash, not a problem, but for learning something entirely 
new or for this sort of detailed troubleshooting, looking for I don't 
quite know what, I gotta be well rested and ready to go.  Simply going on 
caffeine or trying to do it when I'm tired won't cut it -- I've tried and 
end up dead-ended.  Getting such blocks that are free from other 
scheduling issues, not already feverishly tracing down some other issue 
(like bisecting a kernel bug, I regularly run pre-release kernels and bug-
report when necessary -- 2.6.37 has a reported-by/tested-by line by 
Duncan, IIRC the first time they've actually put that in the signoffs 
=:^) ) **AND** where I'm in the appropriate mood is extremely difficult 
and quite rare.  I believe I've been working on this issue for three years 
or longer, now, and I **STILL** haven't been able to get it properly 
traced or isolated, frustrating indeed for someone who's used to reporting 
on kernel bugs and getting them fixed, etc!

It's also possible that it's unique to gmane's way of handling things and 
may in fact be a gmane bug, since that's the only place I see it, but it's 
not like I have a /whole/ lot of other servers I've tried, either (and 
gmane's my only configured server ATM, as the Cox bundled news service 
terminated mid-year and the news provider itself extended it until the end 
of the year, 2010, but of course that's done with now, and I've not 
bothered getting anything else, as about the only non-gmane groups I was 
following were Cox groups anyway). Regardless of gmane or pan, it'd be 
nice to get it pinned down and fixed.

I've written about it in posts here a time or two, and I /think/ I've seen 
someone else report the problem at least once (with groups on a different 
server), one of the times I wrote of it here, but obviously as neither one 
of us has isolated the issue, it might be something different, entirely.  
But mostly I've been quite about it as I've been frustrated with my own 
inability to properly trace it down as I think I /should/ be able to.

-- 
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