On 02/13/2018 11:58 AM, Georg Chini wrote: > On 12.02.2018 17:23, Tanu Kaskinen wrote: >> On Sat, 2018-02-10 at 23:08 +0100, Georg Chini wrote: >>> On 10.02.2018 23:04, Raman Shishniou wrote: >>>> On 02/11/2018 12:43 AM, Georg Chini wrote: >>>>> On 10.02.2018 22:25, Raman Shuishniou wrote: >>>>>> 10.02.2018 23:59, Georg Chini пиÑ?еÑ?: >>>>>>> On 08.02.2018 17:58, Raman Shyshniou wrote: >>>>>>>> Make pipe-source suspended if all writers closed fifo. >>>>>>>> Source will be automatically unsuspended if any data will >>>>>>>> be written to pipe and suspended again when last writer >>>>>>>> closed fifo. >>>>>>>> --- >>>>>>>> src/modules/module-pipe-source.c | 114 +++++++++++++++++++++++++++++++++++++-- >>>>>>>> 1 file changed, 109 insertions(+), 5 deletions(-) >>>>>>>> >>>>>>> What is the motivation/use case for the patch? Normally, if there >>>>>>> are no writers, the source should deliver silence like any other >>>>>>> source without input signal. >>>>>> pipe-source module can't generate silence if no writers connected >>>>>> because it has no clock. Aactually incoming data has some clock, but >>>>>> not pipe-source itself. Use the system clock to generate silence >>>>>> is a bad idea because it may differ from incoming data clock. >>>>> Take a look at the recent patches for pipe-sink that introduced using >>>>> system-clock timing. I guess something similar should be possible >>>>> for the pipe-source. Why would it be a problem, if the timing of the >>>>> silence is slightly different from that of the writer? Different writers >>>>> may have different timing anyway, so if one writer disconnects >>>>> and another connects, timing may change. >>>>> module-loopback is able to deal with sample rate changes on the >>>>> input side and will adapt the sample rate of the output side so that >>>>> it matches the incoming rate to keep a constant latency. >>>>> >>>> I seen the last patches for pipe-sink module. I think there is no >>>> reason to generate silence in pipe-source module with system clock. >>>> The source outputs will do read zeros (resample, convert) - just a >>>> waste of cpu time. >>>> >>> I see your point. The reason for generating silence would >>> be to have consistent behavior for all sources. I'll ask Tanu >>> for his opinion. >> My opinion: it would be nice to generate silence by default if someone >> is willing to implement that, but suspending is better than the current >> behaviour, if the current behaviour is to have the source state as >> RUNNING while not producing any data. >> > Raman, are you willing to implement generating silence? This > could also cover the case where a writer stays connected but > does not provide any data. > > If not, I will proceed reviewing your patch. > Ok. I'll implement silence generator, but I want to leave an autosuspend option. -- Raman
