[openbeosnetteam] Re: about net_server and food for thought

Hi Waldemar,

On 4/2/07, Waldemar Kornewald <wkornewald@xxxxxxxxxxxx> wrote:

Add-ons are indeed a nice concept, but they aren't practical or useful
for everything. Do you expect 3rd-parties to add new configuration
systems? Is there a real use for this and is it worth the programming
overhead?


 Having add-ons is a side effect. I just want the code to be modular
and pluggable. Whether  everything is linked in the same binary or we
have external add-ons doesn't really matter much to me personally.
However i do see the possibility of third parties augmenting the
system with new auto configuration facilities, and having add-ons
would allow this without replacing net_server.


I don't know what the advantage is. I don't think the user will access
that file system directly (by hand), not even for advanced stuff. If a
file system is easier for developers then I'm fine with that, but I
think a nice API would be less cumbersome than digging through
directory trees.


 My main objective is reusal of current technology vs. adding new
stuff. As there is no particular standard in this area we are open to
follow what we see as best. We could very well add an API over the
backend, but my initial point was more about the events themselves and
how to propagate them through the system.


Anyway, as a suggestion, the FS could provide most of the interface
information as attributes, so the file itself could be the direct pcap
interface and its attributes would contain name, address, etc.


 Yes. My only concern right now in terms of how to model the
directory structure is the address issue. At the moment each interface
has a single domain address and this doesn't work very well with IPv6.
In IPv6 each interface has at least a link local address and N (>= 0)
global addresses. With the current model this would require creating
virtual interfaces (something of the sorts of the suggested :0
notation) to handle all of addresses. I would rather have each
interface have a list of addresses, it would simplify the address
selection algorithms. However it isn't very easy to map a list (or
array) with attributes. Besides maybe having a single attribute packed
with all of the addresses, but then it would be hard to signal which
address was added or removed. This was another concern i wanted to
raise, if anyone has any comments, let me know.

 Hugo

Other related posts: