On Sun, 2007-12-16 at 23:57 +0200, Juho Snellman wrote:
> > > I've also confirmed a small test case seems to cause this failure. Right
> > > now it requires hunchentoot and drakma (same as in the app), although I
> > > aim to attempt to minimize the test further. It can still take minutes
> > > to cause the recursive lock failure. I just wanted to get this out
> > > there...
> > Thanks.
> > I think what's going on is that the Hunchentoot reader threads are
> > somehow managing to enter a GC after the dynamic binding of
> > *CURRENT-THREAD* has been unwound.
> And indeed, the attached patch makes the problem go away. It's not the
> right long-term fix, but you might find it useful if the bug is
> causing you trouble with a production service right now.
I can confirm this. It seems like your hypothesis has been borne out.
And I now will do much rejoicing, and will be looking forward to a
principled fix (or a hotfix, if it's deemed critical enough) merged into