Re: [Jack-Devel] non-callback API

On Wed, Apr 20, 2011 at 8:14 AM, Dan Muresan <[hidden]> wrote:
>>> But I wonder what happens if the thread doesn't call
>>> jack_cycle_wait() "often enough". That cannot occur in
>> It's the same as if the application takes too long in the
>> process() function:  You will cause xruns and your
>> application will probably be disconnected from JACK for
>> being a zombie/deadbeat.
>
> OK. Actually process() taking too long would be equivalent to the code
> between cycle_wait() and cycle_signal() taking too long, right?
> Instead I'm asking about the other half -- the code between
> cycle_signal() and cycle_wait() taking too long.
>
> But you probably meant that the effects are the same -- which seems reasonable.

pretty much.

the server will not be waiting on the graph until (at latest) your
client returns from jack_cycle_wait(). the server will wakeup again
(at the earliest) when your client calls jack_cycle_signal() (these
limit cases are where yours is the only client).

if your client continues to do work in between the last call to
jack_cycle_signal() and jack_cycle_wait(), to the extent that it
delays its call to jack_cycle_wait(), then you've effectively stolen
time from that available to run the stuff between wait + signal (the
"process callback"). if you delay for too long, its identical to a
process callback taking too long and the effects will the same - the
server will wakeup, discover that you're not done (i.e. your client
has not called jack_cycle_signal()) in time and then the usual stuff
will occur.

1303305343.12698_0.ltw:2,a <BANLkTin6PfhWcuw9-o+JT_scxjV_0QJyxg at mail dot gmail dot com>