equinox at opensourcerouting
Aug 16, 2012, 4:43 AM
Post #4 of 4
On Mon, Jul 30, 2012 at 11:31:17AM -0300, Renato Westphal wrote:
> 2012/7/30 David Lamparter <equinox [at] opensourcerouting>:
> > On Thu, Jul 19, 2012 at 03:01:09PM +0200, David Lamparter wrote:
> >> Hi everyone,
... aaand I got another two weeks of healthfail. Sorry.
> >> since I've accumulated a large backlog, I'm now doing a merge sprint.
> >> I'll post a summary at the end listing what I've merged instead of
> >> individually ACKing by mail. I expect to put a few days into this, so
> >> the summary will come next week.
> > Tree status: assert failure in bgpd in attr_unintern after
> > hash_release(). Currently hunting this down; might be an issue with an
> > older commit not part of last week's commits.
> Yes, this is an old and big problem. I remember hitting this bug in
> February this year.
> I remember trying to fix this before but had no luck. The solution I
> found is to comment the assert, without this bgpd is almost unusable
> for me.
Grah. I have a backtrace, but it's a late-effect bug, the assert
happens way after the setup code did something wrong...
Either way fixing this is the first thing for me to do right now, we
can't have bgpd asserting at random.
> The real problem is that all the reference counting stuff in bgpd is
> very confusing, IMHO this need a major cleanup.
Almost all of Quagga needs a major cleanup... except we need to retain
the accumulated "wisdom" of soft knowledge and workarounds...
Quagga-dev mailing list
Quagga-dev [at] lists