Re: [Pan-users] problem using 0.135? on Win7pro.

[Bruce Bowler]

On Wed, 05 Oct 2011 08:34:53 +0000, Duncan wrote:

> Steve Davies posted on Tue, 04 Oct 2011 22:49:00 +0100 as excerpted:
> 
>> On 4 October 2011 22:26, Bruce Bowler wrote:
>>> On Tue, 4 Oct 2011 22:04:14 +0100 Steve Davies wrote:
>>>
>>>> It looks from the source code as if Pan is asserting that a news
>>>> article MUST have an article-id. What server are you connecting Pan
>>>> to? NNTP really should not be able to distribute a group unless all
>>>> articles have a unique id.
>>>
>>> 3 different servers...  Individual.net, gmane and a private one.
>>>
>> My personal suspicion would be a mis-configured private server, not
>> creating article-ids automatically?
>> 
>> Does that match with what is being downloaded when the issue happens?
> 
> I'd rather guess that it's either a corner-case threading/race issue, or
> a filesystem issue.
> 
> The filesystem issue is particularly likely, I'd think, given that the
> problem went away after deleting a directory.  But there are two
> possibilities there as well.
> 
> 1)  Do a thorough fsck (filesystem check, on MS it'd be scandisk unless
> they changed the name in the decade I've been off it), and watch for
> other hints that the disk might be failing, as I'm almost positive I
> remember a similar node assertion issue (that one on Linux) from back
> possibly five years (perhaps more, IIRC it was the the old C code pan,
> 0.14.x) ago, that was the first sign of a disk that ultimately failed.
> After the disk finally failed and the reporter was running on a new
> install, the problem could not be duplicated, nor could anyone else
> duplicate it.
> 
> 2)  POSIX based filesystems have far less restrictions in the characters
> allowed in filenames than MS filesystems do.  It's possible that
> someone's using an obscure and possibly non-RFC compliant Message-ID
> generation algorithm, that's generating message-ids with characters that
> pan /should/ escape but doesn't as it's not expecting those characters
> as they're illegal in message-ids to begin with.  Since pan uses the
> message-
> id as the name of the message file in cache, this could create problems
> on MS systems with their more restricted filenames, while it would "just
> work" on POSIX filesystems.
> 
> Because enough stuff depends on message-ids that if they're not there at
> all, one would think the server would have far more problems than a few
> obscure clients such as pan choking.  But if the message-ids are there,
> but include illegal characters, many clients including pan on its native
> POSIX based platforms may not see a problem at all, even as it causes
> problems for pan on MS.
> 
> Certainly, such a problem should be traceable, but it would likely
> involve a debug session that many, including me, couldn't do without
> help.

I've seen it on 2 completely independent machines so I don't think it's 
hardware :-)

As for debugging, if I could cause the problem, I'd be happy to attempt 
debugging, but it appears to be completely random intervals and I'm not 
going to "debug" every time I use pan :-(

To Steve's question about when it happens, I can't say for sure about 
what's being downloaded when the issue happens in terms of what server(s) 
might be active at the time.  Can I set pan to only ever use 1 downloading 
thread?  That way if/when it fails, I could note which group was "active" 
at the time and perhaps narrow it down to a particular server.  Is it 
possible to get the assertion to provide a more descriptive error than "an 
assertion failed"?  Non-descriptive errors remind me of the old DOS 
"general protection fault" errors...

I can say that the last time it happened, when I discovered that "blowing 
way" the groups folder solved the problem, that there didn't appear to be 
any articles missing when it reloaded all the groups.

Bruce