Re: [Jack-Devel] jack_frames_to_time() bug

Le 3 janv. 2012 à 13:04, Fons Adriaensen a écrit :

> On Tue, Jan 03, 2012 at 10:30:18AM +0100, Stéphane Letz wrote:
> 
>> I suggest that you prepare a clean patch with all
>> you proposed "timing related changes" (for jack1
>> for instance)
> 
> I'd be happy to, provided someone can tell me _which_ changes
> are wanted.
> 
> 1. Add the functions to read current and next wakeup times
>   separately, fix jack_frames_to_time().
> 
>   Options in that case:
> 
>   A. Change the jack_nframes_t parameter or result of jack_frame_time(),
>      jack_frames_since_cycle_start(), jack_frames_to_time() and
>      jack_time_to_frames() to a double to preserve full precision.

We cannot change current API...

> 
>   B. Or add new versions keeping the current ones as deprecated.

Possibly.

> 
>   C. Fix the anomaly in jack_frames_since_cylce_start() which
>      seems to use an unfiltered current_wakeup time, making
>      it inconsistent with the rest of the timing functions. 

OK.

> 
> or
> 
> 2. Add an atomic accessor for all four of current_wakeup_time,
>   next_wakeup_time,  last_frame_time, and the current period
>   time estimate of the DLL. This allows the client to do all
>   calculations to full precision. Leave the other functions as
>   the are as a client wanting full precision is not required
>   to use them. Of course still fix jack_frames_to_time().
> 
>   Options in that case:
> 
>   A. Same as C above.
> 
> As a separate but related issue I'd like to review the DLL
> code as I'm not convinced that the current implementation
> provides full precision.

This would be helpful.


> 
> AFAIK, the decision is postponed until after the 'great
> unification', so I'm not really motivated to write a patch
> ATM.

Well not sure the 'great unification' will happen before physicists fix the issue on their side... ((-; maybe the discovery of Higgs boson will help?


> OTOH, fixing the bug in jack_frames_to_time() just requires
> a cut-and-paste of three lines (and testing it of course).      
> You don't need a patch to do that.
> 
> Ciao,
> 
> -- 

Waiting for Paul answer on the whole issue.

Stephane 

1325628273.17154_0.ltw:2,a <F89404B4-5B69-4E65-828E-8AC32C6F78BC at grame dot fr>