Re: [Jack-Devel] Patches for jack-0.121-3

Le 22 mars 2012 à 12:55, Fons Adriaensen a écrit :

> On Thu, Mar 22, 2012 at 10:55:18AM +0100, Stéphane Letz wrote:
> 
> 
>> Right now I see two flags that force "re-init" of the DLL: engine->first_wakeup  and timer->reset_pending. Those flags are tested in jack_run_cycle (engine->first_wakeup || timer->reset_pending).
>> 
>> 1) "first_wakeup" is set in jack_engine_new and in jack_stop_freewheeling. 
>> 
>> 2) "reset_pending" is set in jack_engine_delay
>> 
>> Would it make sense to keep only one flag? and set it in 4 places: first cycle, stop freewheel, XRun handling, buffer size change?
> 
> I didn't remove 'reset_pending' because I wasn't sure if 
> 'first_wakeup' could be used from the context that sets
> 'reset_pending'. The 'pending' part suggested something
> asynchronous, i.e. this flag implicitly being used as a
> sema in combination with some mutexes.
> 
> Apart from that there could be just one flag.
> 
> But you really want the 'perfect' way, there could be more.
> There are 3 things to initialise:
> 
> 1. next_wakeup (to the current wakeup time, as the DLL will run)
> 2. period_usecs
> 3. filter_omega
> 
> 'First cycle' and 'buffer size change' should set them all.
> For 'xrun' and 'end of freewheeling', just setting (1) is
> enough, as the current value of (2) would be more accurate
> than the initial one, and (3) is a constant.

OK, but the thing is that  'end of freewheeling' event happens outside of the audio RT thread. 

Right now JACK2 "set them all" in case of  : 'First cycle', 'buffer size change', and  'end of freewheeling' (since actually the audio driver is restarted and get back in the "First cycle'" case).

In JACK2,  Xrun case can be possibly do " just setting (1)". I'll change that.

> 
> There are two other matters to consider/check.
> 
> For the first, see the comments near the end of jack_run_cycle();

Ok for JACK1, JACK2 does that for "each callback" (real one or emulated one  with a thread + blocking backend)
> 
> The second concerns Jack1 only IIRC. If a cycle is skipped, 
> the DLL should be run anyway. Jack1-0.121.3 did the right
> thing and I don't think the patches affect this. But it's
> worth checking.
> 
> Ciao.
> 
OK for JACK1 possibly, JACK2 does not have this issue.

Stéphane

1332419512.5586_0.ltw:2, <786B68C4-6049-4448-9F25-C6314B3C23D9 at grame dot fr>