On 09/09/13 13:26, Dirkjan Ochtman wrote:
On Mon, Sep 9, 2013 at 12:50 PM, Martin Sustrik <sustrik@xxxxxxxxxx> wrote:Yes, classic feature-creep problem.I worry about this as well; the current discussion seems to have veered in many directions, with no good consensus emerging... Am I wrong? If so, can someone summarize the consensus (preferably in a new thread)?
Don't worry, it's just a discussion on a rather complex topic, so it is expected to be a bit messy.
Btw, if you put aside all the complex stuff discussed above and look only on what the library itself is supposed to do, it's in fact very simple: When opened, the socket does a query and it gets back a list of nn_connects, nn_binds and nn_setsockopts it applies to itself.
That way the programmer can use the socket without being aware of actual IP addresses etc. The configuration stuff is left for the admin to specify.
I almost feel like there should be an Enhancement Proposal process similar to what Python and a bunch of other projects have these days.
Any pointers?
At the moment, the idea is to have a separate address family for the configuration service.Is it possible to add separate address families from outside the nanomsg core? If so, I think it would be preferable to develop this there at first, to see what works and to avoid contaminating nanomsg with a bunch of extra complexity during the design of the configuration service stuff.
Yes, it's definitely possible. I not sure it's worth it though given that the functionality is probably going to be something as simple as reading a config file or doing a database query.
Martin