Testers wanted: reentrant resolver

[Brian F. Feldman]

Peter Schultz <pmes at bis.midco.net> wrote:
> Brian F. Feldman wrote:
> > Patch updated :)
> > <URL:http://green.homeunix.org/~green/reentrant_resolver.patch>
> > 
> 
> Technicalities aside, this patch rules!
> 
> Thanks,
> Pete...
> 
> P.S.  Fundakowski?!?!  What a fine name!  8^)

Thanks ;-)  Okay, I found and fixed the MAJOR bug that was screwing things 
up -- the "s" field in res_send.c was not being initialized to -1, but to 0! 
This resulted in whatever was residing at fd 0 being closed if it responded 
to kqueue() (for the read event)!  This also appears to be why I was seeing 
resolvers sometimes take five minutes.

The patch has been updated; everyone, please try it now if you saw streams 
randomly being closed in mozilla (which I did).  I am pretty confident in 
the behavior now, because I watched it with debugging printfs to make sure
res_close() never closed 0 ;-)

<URL:http://green.homeunix.org/~green/reentrant_resolver.patch>

Hopefully more people will test this now so I can feel confident and commit 
it.  It will be really nice to MFC, too, since -STABLE users also see this 
behavior just using -lc_r.

-- 
Brian Fundakowski Feldman                           \'[ FreeBSD ]''''''''''\
  <> green at FreeBSD.org                               \  The Power to Serve! \
 Opinions expressed are my own.                       \,,,,,,,,,,,,,,,,,,,,,,\