On 8/25/06, Garrett Rooney <rooneg@electricjellyfish.net> wrote:
> On 8/25/06, Vlad Georgescu <vgeorgescu@gmail.com> wrote:
>
> > Hi Garrett,
> >
> > Could you test the following patch?
> >
> > It solves the problem by maintaining a count of how many sasl contexts
> > were created, and only calling sasl_done() when all of them are
> > destroyed (the count starts from 1, because we also decrement it in
> > sasl_done_cb). It seems to work for me (i.e. the test runs
> > successfully).
>
> Unfortunately, that's not going to work. The problem isn't that we
> call sasl_done(), then later on call sasl_dispose() on some existing
> sasl context (well, that's probably A problem, but it's not THE
> problem). The problem is that if we do any sasl stuff after sasl_pool
> is destroyed then the mutex locking/unlocking that happens in the
> subsequent calls to sasl_done() or sasl_dispose() will try to
> lock/unlock mutexes allocated in memory that's potentially been reused
> because it was allocated out of sasl_pool.
Ok, but this patch by itself is still valid, right? After all, it
somehow manages to stop svn from crashing on my machine, despite the
other issues :)
> I'm thinking that we may be able to get away with sticking if
> (sasl_done) return; in various places, since this will only ever
> happen at global shutdown time, at which point we've already assumed
> that we're in single threaded mode, so locking isn't needed...
I'll add those checks. Should I submit a separate patch, or
incorporate them in the other one?
--
Vlad
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Aug 25 18:30:10 2006