17 Feb
2005
17 Feb
'05
11:41 a.m.
Le 17 févr. 05, à 16:16, Alfons Adriaensen a écrit :
On Thu, Feb 17, 2005 at 09:43:16AM -0500, Paul Davis
wrote:
Yes, this is currently done like that in jackdmp
stephane's new OSX implementation (jackdmp)
avoids both of these, and
ends up with an extra process-cycle worth of latency. he does exactly
what you suggest above. is it avoidable? i don't know. stephane didn't
seem to think so when we talked about it briefly on IRC. maybe it can
be.
Strange. The only way you could have an extra cycle of latency is by
explicitly programming for it, either inserting a buffer somewhere,
or by changing the sequence
backend wakeup -> read inputs - call graph - write outputs
to
backend wakeup -> write outputs - read inputs - call graph
which amounts to the same thing. How would this help in allowing
processing to continue during graph changes ?
I'm convinced it should be possible. The price to pay should not
not be increased audio latency,
I think it depends of what is important: a "more" robust system with
one extra buffer latency or the current model.
I will make both options available in jackdmp.
but an asynchronous interface
to the client creation / port creation and connection functions.
In other words, you will have to request something and the result
will be known not when your call returns, but by some asynchronous
event. This makes sense anyway for things that can be modified from
any number of places at the same time.
Hum.... I'm not sure asynchronous interface is needed. In my
implementation graph state changes are "seen" by the audio thread the
next cycle, but since all calls that change the graph state are
serialized by the server, clients always see a coherent state.
Stephane