On Wed, 2008-01-30 at 17:31 +0000, Bastien Nocera wrote:
> On Tue, 2008-01-29 at 18:01 +0100, Marcel Holtmann wrote:
> > Hi Bastien,
> > 
> > > Some time ago I had problems with client.c's architecture. It would
> > > allow me to easily report completed discoveries without causing crashes.
> > > 
> > > The attached patch moves all the static variables into the main
> > > BluetoothClient object, and instead of return a new object when using
> > > _new() we return an existing instance of the client.
> > > 
> > > This makes no changes to the external API for the BluetoothClient, and
> > > should work just as well as older versions did.
> > 
> > patch has been applied. Don't really see the difference, but I am okay
> > with doing it this way.
> > 
> > > A couple of bugs I found in client.c:
> > > - launch test-client with a dongle inserted, remove and reinsert the
> > > dongle, the bdaddr of the device is not set anymore
> > > - launch test-client without a dongle inserted, insert it, the bdaddr is
> > > not set and there's no children to the adapter (although there are if
> > > inserted when started)
> 
> Both of those are a race:
> When the AdapterAdded signal is launched in hcid, the address isn't
> available yet. It is available when "ModeChanged" has been launched
> though.
> 
> hcid_dbus_register_device() setups the structures, and launched
> AdapterAdded, but the address is only set when hcid_dbus_start_device()
> is called (and the ModeChanged signal is launched).
> 
> How do you think we should be handling that? I would think we should
> make GetAddress fail if the data isn't available yet, or hack around it
> in client.c and only setup the adapters when the mode changes.

The easy fix for hcid attached (the default bdaddr should be
00:00:00:00:00:00:00, not an empty string).

And also a patch for client.c to get the device address again when the
device changes mode, so we can change it in the tree from the default
all-zeros address to the real one.

I'll fix the other bugs now.

Cheers