[LAD] OSC: Divide & Conquer, or build a Stronghold?

[Stephen Sinclair]

On Fri, Dec 18, 2009 at 9:53 AM, nescivi <nescivi at gmail.com> wrote:
> On Thursday 17 December 2009 18:41:31 Harry Van Haaren wrote:
>> Hey all,
>>
>> I've been keeping myself busy lately, mostly with Python and OSC,
>> and I'm using multiple clients/controllers to send messages to a
>> sampler with an OSC interface.
>>
>> From this experience it seems to me a good idea, to have a kind of
>>  "central"
>>
>> place where all Audio programs "Announce" thier OSC port & program name &
>> version,
>> so the linux-audio-desktop can be a little more "coherent" for lack of a
>> better word.. ;-)
>>
>> We could create a kind of "Master OSC Host" which would keep track of
>> which clients are running, and which ports they are using. This would be
>> easiest
>> to do if EACH client, would "register" itself on an agreed port.
>>
>> So if a client wants to read some information, they could query the
>> "Master OSC Host" with a standardised set of questions about the current
>> state of QTractor/Ardour/<Any OSC capable program>.
>
>
> I think the OSC protocol in and of itself is too general for this to make
> sense on a global level.
>
> It seems you have already in mind a specific subset of kinds of data to
> exchange between programs.
>
>
> There have already been several approaches for this on specific topics.
>
> Such as the Digital Orchestra Toolkit Mapping tools,
> http://idmil.org/software/mappingtools, for mapping data between digital
> instruments/controllers, and software synths. There they define how to
> announce your controller and how to announce your synth, and then you can
> create arbitrary mappings.
>
> Personally, I have worked on an osc-based data-exchange framework for
> exchanging data (such as sensor data, but basically any data) between
> different collaborators, creating light, sound, video or other media with
> their programs, that work on interactive performances or installations.
> See http://sensestage.hexagram.ca
>
> In the WONDER software (http://swonder.sourceforge.net) we created a central
> control host which transfers and translates OSC data between the different
> components of the software. Some of that concept could be generalized for
> programs to share any kind of osc messages with each other.
>
> There are also some attempts amongst livecoders to share clocks for playing
> together...
>
>
> To summarise, it could be an interesting project, but be aware that, since the
> OSC standard is so open - in principle it doesn't even specify the method of
> transport, though most often UDP (in most cases) or TCP/IP (fewer cases) is
> used, but there are also implementations that use a serial port transmission
> (very few cases) - (getting back to my sentence) and the implementation and
> use of it in various programs varies a lot, some are only taking incoming
> transmissions and don't send messages, if based on UDP and TCP/IP, some send
> from a fixed port, some don't, some can listen to multicast messages, others
> don't. You have to consider whether you want to require them to adhere to a
> specific namespace, or whether you want to work with their existing
> namespaces.
>
> That being said... a general purpose osc-mapping program could be really
> useful... something you can put between two programs (one sending, one
> receiving), that translates from one namespace to another, and possibly does
> the required translations.

Just to add...

I am one of the contributors for the Digital Orchestra Toolkit (hi
Marije!) and I am currently working on a C implementation, so this
might be of interest once it's ready.  (It's coming along well, should
be public in January.. but I can make sources public sooner if there's
any interest.)

This solution is somewhat domain-specific, designed for creating
arbitrary mappings between controllers (input devices) and
synthesizers.  So, not necessarily an answer to the proposal of this
thread, but perhaps interesting nonetheless.

Basically we found ourselves in a situation where we were making
Max/MSP patches to read input from digital gestural controllers,
usually by reading serial data from Arduinos, and we wished to perform
signal condition and connect these values to inputs to sensors.

We decided to try and design an approach which would _not_ require
maintaining some kind of central database, and that would work across
a network.

So the Digital Orchestra Mapper as we'll call it is basically a set of
standard messages which are communicated on a multicast port.  Each
device (either controller or synth) announces itself on this common
bus, and then we made a Max/MSP GUI which can be used to assign
connections between outputs and inputs for known devices.  These
connections may have arbitrary operations applied to them.  From the
controller's point of view it is simply outputting signals, but they
pass through a "router" which converts these signals into different
messages before actually sending them on the network.  There is one
router per connection, and the router is assigned to a specific
destination address.  Therefore the multicast bus is only used to set
up connections and describe OSC namespaces, but the signal connections
are peer-to-peer.

Since I am more of a Linux guy, I am interested in using this outside
Max/MSP so I am porting it to C and hoping to provide SWIG bindings to
several languages.  The API will look something like:

    m = mapper_device_new();
    sig = mapper_signal_new(... description ..);
    mapper_device_add_signal(m, sig);

and internally in response to requested connections..

    r = router_new(address..);
    mapper_add_router(m, r);
    ...
    ... TODO: haven't finished designing the internal API for
              specifying signal conditioning

and repeatedly,

    mapper_update_signal(m, sig, value);

Essentially the user program just creates a "device", assigns
"signals", and repeatedly updates them, and then the API takes care of
communicating with the network, finding a unique network name, making
connections, etc.
On the synth-side, I think a callback interface will work for
responding to updated signals, similar to LibLo's message handlers.

Steve