Re: [libvirt] [PATCH] Revert "dbus: correctly build reply message"

On Mon, Sep 09, 2019 at 03:47:21PM +0400, Marc-André Lureau wrote: > Hi > > On Fri, Sep 6, 2019 at 7:37 PM Michal Privoznik <mprivozn(a)redhat.com&gt; wrote: > > > > This reverts commit 39dded7bb61444bb608fadd3f82f6fe93d08fd0e. > > > > This commit broke virpolkittest on Ubuntu 18 which has an old > > dbus (v1.12.2). Any other distro with the recent one works > > (v1.12.16) which hints its a bug in dbus somewhere. Revert the > > commit to stop tickling it. > > > > Signed-off-by: Michal Privoznik <mprivozn(a)redhat.com&gt; > > This patch is no longer necessary for dbus-vmstate in this series. I > found this "bug" when implementing an alternative solution. So > reverting it is fine by me. > > However, it may be revealing a transient bug in > virDBusMessageIterDecode() rather than libdbus. It would be worth to > have VIR_DEBUG() output from the failing test. It is not really a bug - rather its due to the short cut taken by the test suite. virpolkittest is mocking the dbus_connection_send_with_reply_and_block method. This method accepts a DBusMessage input (the MethodCall) and has to create DBusMessage as output (the MethodReply) The DBusMessage input has no serial number set, since this is the job of the real dbus_connection_send_with_reply_and_block() method impl. The dbus_message_new_method_return() impl expects the original message to have a valid serial number though. Thus when we create the DBusMessage output, we hit an assertion failure when calling dbus_message_new_method_return(). This is in fact why the code intentionally uses dbus_message_new(DBUS_MESSAGE_TYPE_METHOD_RETURN) in the first place. What's strange though is why it ever succeeds on any distro. The assertion we're seeing should *always* hit us on every distro.