On Don, 2012-12-27 at 16:06 -0500, Hans-Christoph Steiner wrote:
On Dec 27, 2012, at 3:36 PM, Roman Haefeli wrote:
On Don, 2012-12-27 at 15:29 -0500, Hans-Christoph Steiner wrote:
On Dec 27, 2012, at 2:14 PM, Roman Haefeli wrote:
On Die, 2012-12-25 at 13:00 -0800, Miller Puckette wrote:
Hi all - I'm afraid to 'fix' this riht now, but will look at it at least. Alternatively I could add a message to pd to restort DSP without stopping/starting the audio I/O, which I'm hoping will at least reduce the need to start/stop SDP all the time.
The change that caused this problem is that I fixed Pd not to have the audio system open whern DSP isn't running so that, on APIs in which audio is exclusive (e.g., ALSA) you can turn DSP off in Pd and go off and use another device with Pd remining open.
The problem arises when there is one message trying to control many different tasks:
- switch DSP computing on/off (save CPU time)
- free/occupy the audio back-end (save soundcard time)
- mute/unmute Pd (save ears)
- force recompilation of the DSP graph
An alternative would be to have Pd automatically close audio devices on ALSA, OSS, and MMIO but always keep it open when using jack. But then what about portaudio? I somply don't know the correct way to deal with this.
I'd say it's preferable if Pd's behavior is independent from the back-end currently in use. Rather, I suggest to use different messages for different tasks. IMHO, 'dsp 0' really should only turn computation of tilde-objects off as this is what it means and what one expects. What about a new message to pd 'card 0|1' to free devices which would work completely independently from 'dsp 0|1'? This would allow to record a signal to a soundfile without occupying the soundcard, for instance.
I think that having a pulseaudio backend for Pd as the default on GNU/Linux would be the best way to solve this problem. Pulse will then handle the multiple apps playing at the same time. Then for people who want to skip Pulse, then can use the ALSA blocking implementation.
An Audio API can be quite different than another, so I don't think it makes sense to treat them all the same in terms of things like whether DSP disconnects the audio API. For example Mac OS X CoreAudio, even through PortAudio, always has handled multiple apps playing back. And the "audio stuck" workaround actually only causes problems with CoreAudio.
So you agree that 'dsp 0|1' should not automatically disconnect|connect the audio API, regardless of what API is in use?
I don't know enough about the various audio APIs to say whether that's a good idea or not. Perhaps it makes sense with some audio APIs.
I'm not sure you understood my proposal. Exactly _because_ it might not make sense for all APIs, I suggested to use a separate message for API on|off, so that 'dsp 0|1' can behave exactly the same with all APIs (as it used to do anyway before 0.43). Does that make sense?
Example: I often use an idiom [dsp 0, dynamically create stuff, dsp 1] in my patches. I wouldn't want this to work nicely only with -jack while causing ugly clicks with -alsa or -oss. The patchs behaviour shouldn't be affected by the audio API currently in use.
Roman