Strange INN-current crash.

Julien ÉLIE julien at trigofacile.com
Fri Apr 3 17:15:55 UTC 2009


Hi Petr,

> JL> Then, on later CHECK, it returns 238 (and not 438) because it hopes a
> JL> well-formed
> JL> article will be sent for this article (it was maybe an error the first
> JL> time).
>
> I believe that this is an error. Very little chance of receiving
> well-formed article.. Only Cyclone drop empty header, but I not know
> what Cyclone does with spaces in header..
> And this wasting bandwith, especially for binaries groups.

OK, I will try to make innd remember the message-ID.


> Invalid Cc: header without spaces - ("Cc:").
>
> No log in "news" about this (this is error, and i not know your "new
> patch"), and this "bad" article not remember (this is also wrong?).

Yes, this is now logged with my new patch and I still need to remember
the message-ID.


> And if "Cc:" after Message-ID:
>
> Log:
> Apr  3 14:37:35.832 - npc5.stu.neva.ru <a1 at 54> 439 No colon-space in "Cc:"
> header
>
> log is fine, but also without history update.

Yes, that was exactly the bug I saw:  as the message-ID was unknown, it was
not logged.
In such cases, we might for instance use the message-ID provided in the
argument of TAKETHIS/IHAVE; I have not checked yet whether it is easy
to do.


> If I get a lot of such articles from many peers, may be a problem with
> logging (perfomance, for example), because I have one log message for each
> peer for SAME article.
> I think better to remember those bad articles.

Yes, I understand.


> I think accept "Distribution: " and "Distribution:   " (I remember your
> USEPRO), but not accept "Distribution:", "Distribution", and remember this
> msgid.

That is exactly what I am going to do.
I will tell you when I commit the patch to CURRENT.

The more important thing was to fix the segfault you saw.  (Which was done.)

-- 
Julien ÉLIE

« It's documented in The Book, somewhere... » (Larry Wall) 




More information about the inn-workers mailing list