Re: Note about device activation in current trunk...

[Dan Williams <dcbw redhat com>]

On Thu, 2007-09-13 at 16:44 +0200, Helmut Schaa wrote:
> Hi,
> 
> I just tried to implement the feature "create new connection" in KNM and ran 
> into the following problem:
> 
> After entering all information for the new connection KNM sends 
> the "NewConnection"-signal to NM and afterwards calls the "Activation"-method 
> for the appropriate device.
> 
> Unfortunately NM did not get the settings (which are requested 
> during "NewConnection"-signal-processing) for this connection yet when 
> the "Activation"-call arrives and therefore fails :(

Yup.

> I'm not sure but most likely nm-applet will have the same problem (this 
> feature is not implemented in nm-applet yet).

I did this last night actually.

> Dan, Tambet, how can we avoid this issue?

Yeah, I know this is a problem at the moment, because it's unlikely NM
has gotten the configuration info for the connection yet.  This is what
I'm fixing today; I believe the issue needs to be solved in NM itself to
keep the APIs and interface behavior simple for clients like knm and
nm-applet.

NM internally will always request the connection details when a new
connection appears from the user settings daemon, so it's a matter of
teaching the activation code to wait a bit before failing the
activation, probably with a timer.  The Activate() interface was done
with object paths because NM has to have the connection object path when
it calls GetSecrets(), and its simpler in the client to have this all be
in the same place rather than special casing new NMConnections that NM
may/may not know about yet.  Assume that NM will do the right thing here
and activate a connection you tell it about, that it may not yet have.

Dan