Re: Latest dbus update may hang kde4 in NetBSD amd64

[Sverre Froyen <sverre%viewmark.com@localhost>]

On Tuesday 14 August 2012 13:19:02 David Brownlee wrote:
> On 14 August 2012 17:56, Sverre Froyen <sverre%viewmark.com@localhost> wrote:
> > On Saturday 04 August 2012 04:30:27 Sverre Froyen wrote:
> > ...
> > 
> >> I'm seeing the same thing after an "make replace" in sysutils/dbus. A
> >> ktrace of the kded4 process shows a stream of
> >> 
> >>    205      1 kdeinit4 CALL  poll(0x7f7fffffa2d0,1,0xfffeb2c8)
> >>    205      1 kdeinit4 RET   poll -1 errno 22 Invalid argument
> >>    205      1 kdeinit4 CALL  __gettimeofday50(0x7f7fffffa340,0)
> >>    205      1 kdeinit4 RET   __gettimeofday50 0
> >> 
> >> calls.
> > 
> > Setting the timeout value in the argument to poll to -1 whenever it is
> > less that -1 works around the issue. Update patch-ab with:
> > 
> > diff -u -r1.19 patch-ab
> > --- patch-ab    1 Aug 2012 17:52:22 -0000       1.19
> > +++ patch-ab    14 Aug 2012 16:53:39 -0000
> > @@ -134,3 +134,14 @@
> > 
> >       _dbus_verbose ("Socket credentials not supported on this OS\n");
> >   
> >   #endif
> >   
> >     }
> > 
> > +@@ -2525,6 +2581,10 @@ _dbus_poll (DBusPollFD *fds,
> > +       _DBUS_STRUCT_OFFSET (DBusPollFD, revents) ==
> > +       _DBUS_STRUCT_OFFSET (struct pollfd, revents))
> > +     {
> > ++      if (timeout_milliseconds < -1) {
> > ++        _dbus_warn("_dbus_poll: timeout = %d (fixed)\n",
> > timeout_milliseconds);
> > ++        timeout_milliseconds = -1;
> > ++      }
> > +       return poll ((struct pollfd*) fds,
> > +                    n_fds,
> > +                    timeout_milliseconds);
> 
> Wonderful - thanks. Fixes the issue for me - dbus pkgsrc entry updated!

We should probably try to figure out why the timeout values go negative. It 
might be good to see if this also happens on other platforms  -- particularly 
Linux. With the updated patch-ab, it is easy to tell. Just grep for _dbus_poll 
in the xwindows log. On my NetBSD amd64 I get:

_dbus_poll: timeout = -529 (fixed)

If it is a general issue I think we should file an upstream dbus bug report.

Regards,
Sverre