Hi Brian, 

On Fri, 2019-09-27 at 15:50 +0000, Gix, Brian wrote:
> On Fri, 2019-09-27 at 10:52 +0200, michal.lowas-rzechonek@silvair.com
>  wrote:
> > Inga, Brian,
> > 
> > Still, even if we add a method, the application is free *not* to
> > implement it, since we agreed back in the day that calls to
> > MessageReceived do not require a response, so any errors would be
> > simply
> > ignored by the daemon.
> 
> This is not an option.
> 
> A node does not get to decide which susbscriptions are "valid".  If a
> Virtual Address subscription is added to
> a node, and then a message is sent to that virtua address, the App
> needs to be able to receive it.
> 
> Yes, any discrete message may be lost, but I don't think we have the
> option of letting *all* virtual addressed
> messages to an App to be ignored.  If we add an App API, it will need
> to be mandatory, which is why I am
> against it.
> 
> I strongly believe we need:
> 
> 1. A single method for delivering (non dev key) received messages
> 2. A method that does not require dictionary parsing
> 
> How are we feeling about:
> void MessageReceived(uint16 source, uint16 key_index,
> array{byte} destination, array{byte} data)
> 
> 
> Where destination length of:
> 0 - Unicast address of element
> 2 - Group Address
> 16 - Variable Label
> 
> I think this fulfills all of our requirements.
> 

For a single MessageReceived() method, the cleanest way is to have the
subscription address parameter as a variant (suggested by Michal) or as
a dictionary.
An array introduces an extra consideration of byte ordering for group
addresses.

What I mostly about is that the represenation of the subscription
address in the MessageReceived() method corresponds to the
representation in the configuration dictionaries for the Attach() and
UpdateModelConfiguration() methods.