[PATCH] do not call _dbus_warn_check_failed on checks

Timo Hoenig thoenig at suse.de
Tue Nov 14 04:51:05 PST 2006


On Tue, 2006-11-14 at 13:10 +0100, Thiago Macieira wrote:

> Daniel Stone wrote:
> >> For another, there might be a lot of state kept in the applications
> >> that may or may not be transferred to the new connection. This is
> >> especially true for the unique connection names, which will be
> >> restarted. Other problems also include race conditions when a call
> >> comes in just after restart and the target wasn't yet present, etc.
> >
> >It's difficult, and you can never get it absolutely, perfectly,
> >bullet-proof, 100% bug-free.  But you can optimise for the common case
> >and get it right 99.9% of the time.  And when the planets align and
> >conspire against you to cause one tiny minor failure in an extreme
> >corner case, then it's still better than exit(3), which works for
> >exactly 0.0% of the cases.
> That is true.
> But to reduce chance of error, the bus shouldn't be restarted unless 
> absolutely necessary.

That is for security updates.  In case D-Bus is being used on a high
availability system you want the system bus being restarted in such

But all this has been thoroughly been discussed for all important
applications using D-Bus.

There are distributions out there which get along quite nicely with
system bus restarts once D-Bus is being updated.

However, I do not share Frederic's POV when it comes down to updating
applications which use the system bus.  The dnotify support in D-Bus
works nicely and I have never seen problems with regard to updating the
bus policy at runtime (read: without restarting the system bus in %


More information about the dbus mailing list