Asynchronous notification in ALSA - I/O Plugin - Pulseaudio

Asynchronous notification in ALSA - I/O Plugin - Pulseaudio

[Juha Erkkilä]

I'm trying to get Abode Flash Player working properly on a thin client,
with sound, using Pulseaudio as a sound server.  I'm using ALSA
version 1.10.15 and Pulseaudio version 0.9.10 on Ubuntu Hardy, with
Pulseaudio server running as a system-wide daemon on the thin client.
I already outlined the issues on Pulseaudio mailing list on this post:
https://tango.0pointer.de/pipermail/pulseaudio-discuss/2008-May/001793.html,
see also Lennart Poettering's answer:
https://tango.0pointer.de/pipermail/pulseaudio-discuss/2008-May/001796.html.

I feel the proper solution would involve writing support for
asynchronous notification into ALSA library and I/O-plugin layer.
Currently, Pulse I/O-plugin does not work in the Flash case, because
Flash requires the asynchronous API.  On the other hand, Pulse requires
the I/O-plugin layer to work.  See the posts by Alexander Indenbaum,
http://mailman.alsa-project.org/pipermail/alsa-devel/2008-April/007033.html,
and Takashi Iwai,
http://mailman.alsa-project.org/pipermail/alsa-devel/2008-April/007335.html,
on this subject.

I tried to tackle this issue, and did get the signal handling code to
call the Flash audio callback, but I'm new to both ALSA and Pulseaudio,
and it appears there are some issues with it I don't fully understand.
Thus I'm asking: are there any other people who have been looking into
this, or anyone who can tell whether this is a sane thing to do or not,
will it require some nasty hack, or can help with some pointers to
relevant information?

I'm hoping that implementing the asynchronous API on I/O-plugin layer
would not only fix Flash audio, but might help some other programs too
to achieve network transparent audio.

Juha

Re: Asynchronous notification in ALSA - I/O Plugin - Pulseaudio

[Pacho Ramos]

Have you tried with flashplayer 10 beta?

Re: Asynchronous notification in ALSA - I/O Plugin - Pulseaudio

[Takashi Iwai]

At Mon, 19 May 2008 12:38:39 +0300 (EEST),
Juha Erkkilä wrote:
> 
> I'm trying to get Abode Flash Player working properly on a thin client,
> with sound, using Pulseaudio as a sound server.  I'm using ALSA
> version 1.10.15 and Pulseaudio version 0.9.10 on Ubuntu Hardy, with
> Pulseaudio server running as a system-wide daemon on the thin client.
> I already outlined the issues on Pulseaudio mailing list on this post:
> https://tango.0pointer.de/pipermail/pulseaudio-discuss/2008-May/001793.html,
> see also Lennart Poettering's answer:
> https://tango.0pointer.de/pipermail/pulseaudio-discuss/2008-May/001796.html.
> 
> I feel the proper solution would involve writing support for
> asynchronous notification into ALSA library and I/O-plugin layer.
> Currently, Pulse I/O-plugin does not work in the Flash case, because
> Flash requires the asynchronous API.  On the other hand, Pulse requires
> the I/O-plugin layer to work.  See the posts by Alexander Indenbaum,
> http://mailman.alsa-project.org/pipermail/alsa-devel/2008-April/007033.html,
> and Takashi Iwai,
> http://mailman.alsa-project.org/pipermail/alsa-devel/2008-April/007335.html,
> on this subject.
> 
> I tried to tackle this issue, and did get the signal handling code to
> call the Flash audio callback, but I'm new to both ALSA and Pulseaudio,
> and it appears there are some issues with it I don't fully understand.
> Thus I'm asking: are there any other people who have been looking into
> this, or anyone who can tell whether this is a sane thing to do or not,
> will it require some nasty hack, or can help with some pointers to
> relevant information?
> 
> I'm hoping that implementing the asynchronous API on I/O-plugin layer
> would not only fix Flash audio, but might help some other programs too
> to achieve network transparent audio.

Yeah, the async support is a big missing piece in ioplug.  I didn't
think it be so important until someone told me that flash player uses
it.  Sigh.

I have experimental patches to add async support to ioplug.  With a
quick test using jack plugin, it seems working somehow.  Will post
them alsa-devel now.


thanks,

Takashi

Re: Asynchronous notification in ALSA - I/O Plugin - Pulseaudio

[Juha Erkkilä]

----- "Takashi Iwai" <tiwai@suse.de> kirjoitti:
> Yeah, the async support is a big missing piece in ioplug.  I didn't
> think it be so important until someone told me that flash player uses
> it.  Sigh.

It turns out Flash version 10 beta no longer uses the asynchronous API,
so using it is an option.

> I have experimental patches to add async support to ioplug.  With a
> quick test using jack plugin, it seems working somehow.  Will post
> them alsa-devel now.

Thank you!  I tested these with Firefox + Flash 9 + Pulseaudio, and while
sound output works, quitting any Flash application will freeze firefox.

Juha

Re: Asynchronous notification in ALSA - I/O Plugin - Pulseaudio

[Takashi Iwai]

At Wed, 21 May 2008 09:46:04 +0300 (EEST),
Juha Erkkilä wrote:
> 
> ----- "Takashi Iwai" <tiwai@suse.de> kirjoitti:
> > Yeah, the async support is a big missing piece in ioplug.  I didn't
> > think it be so important until someone told me that flash player uses
> > it.  Sigh.
> 
> It turns out Flash version 10 beta no longer uses the asynchronous API,
> so using it is an option.

Ah, that's good to know.

> > I have experimental patches to add async support to ioplug.  With a
> > quick test using jack plugin, it seems working somehow.  Will post
> > them alsa-devel now.
> 
> Thank you!  I tested these with Firefox + Flash 9 + Pulseaudio, and while
> sound output works, quitting any Flash application will freeze firefox.

It's a showstopper, then.  Could you figure out in which code does it
freeze?  My change is just to add async call, so it sounds like that
another hidden bug is hit by my change...


thanks,

Takashi

Re: [alsa-devel] Asynchronous notification in ALSA - I/O Plugin - Pulseaudio

[Lennart Poettering]

On Wed, 21.05.08 12:47, Takashi Iwai (tiwai@suse.de) wrote:

> > > I have experimental patches to add async support to ioplug.  With a
> > > quick test using jack plugin, it seems working somehow.  Will post
> > > them alsa-devel now.
> > 
> > Thank you!  I tested these with Firefox + Flash 9 + Pulseaudio, and while
> > sound output works, quitting any Flash application will freeze firefox.
> 
> It's a showstopper, then.  Could you figure out in which code does it
> freeze?  My change is just to add async call, so it sounds like that
> another hidden bug is hit by my change...

That freeze is an unfixable race condition that Flash has when
shutting down audio.

See the explanation down on http://pulseaudio.org/ticket/267

Lennart

-- 
Lennart Poettering                        Red Hat, Inc.
lennart [at] poettering [dot] net         ICQ# 11060553
http://0pointer.net/lennart/           GnuPG 0x1A015CC4

Re: [alsa-devel] Asynchronous notification in ALSA - I/O Plugin - Pulseaudio

[Colin Guthrie]

Lennart Poettering wrote:
> On Wed, 21.05.08 12:47, Takashi Iwai (tiwai@suse.de) wrote:
> 
>>>> I have experimental patches to add async support to ioplug.  With a
>>>> quick test using jack plugin, it seems working somehow.  Will post
>>>> them alsa-devel now.
>>> Thank you!  I tested these with Firefox + Flash 9 + Pulseaudio, and while
>>> sound output works, quitting any Flash application will freeze firefox.
>> It's a showstopper, then.  Could you figure out in which code does it
>> freeze?  My change is just to add async call, so it sounds like that
>> another hidden bug is hit by my change...
> 
> That freeze is an unfixable race condition that Flash has when
> shutting down audio.
> 
> See the explanation down on http://pulseaudio.org/ticket/267

I thought the whole point was that by adding async code to ioplug, the 
libflashsupport stuff wasn't needed and flash could just use it's native 
alsa support + alsa's pulse plugin.... or am I missing the point here?

Col

Re: [alsa-devel] Asynchronous notification in ALSA - I/O Plugin - Pulseaudio

[Lennart Poettering]

On Thu, 22.05.08 14:11, Colin Guthrie (gmane@colin.guthr.ie) wrote:

> > That freeze is an unfixable race condition that Flash has when
> > shutting down audio.
> > 
> > See the explanation down on http://pulseaudio.org/ticket/267
> 
> I thought the whole point was that by adding async code to ioplug, the 
> libflashsupport stuff wasn't needed and flash could just use it's native 
> alsa support + alsa's pulse plugin.... or am I missing the point here?

The destruction dead lock issue is present both in libflashsupport and
the native ALSA backend. It's just that it is almost never triggered
if a hw device is used.

Adobe released the sources of libflashsupport where you can actually
check this.

Lennart

-- 
Lennart Poettering                        Red Hat, Inc.
lennart [at] poettering [dot] net         ICQ# 11060553
http://0pointer.net/lennart/           GnuPG 0x1A015CC4

Re: [alsa-devel] Asynchronous notification in ALSA - I/O Plugin - Pulseaudio

[Colin Guthrie]

Lennart Poettering wrote:
> On Thu, 22.05.08 14:11, Colin Guthrie (gmane@colin.guthr.ie) wrote:
>> I thought the whole point was that by adding async code to ioplug, the 
>> libflashsupport stuff wasn't needed and flash could just use it's native 
>> alsa support + alsa's pulse plugin.... or am I missing the point here?
> 
> The destruction dead lock issue is present both in libflashsupport and
> the native ALSA backend. It's just that it is almost never triggered
> if a hw device is used.

Ahhhhhhhhh :)

Col

Re: [alsa-devel] Asynchronous notification in ALSA - I/O Plugin - Pulseaudio

[Lennart Poettering]

On Mon, 19.05.08 12:38, Juha Erkkilä (juha.erkkila@opinsys.fi) wrote:

> I'm trying to get Abode Flash Player working properly on a thin client,
> with sound, using Pulseaudio as a sound server.  I'm using ALSA
> version 1.10.15 and Pulseaudio version 0.9.10 on Ubuntu Hardy, with
> Pulseaudio server running as a system-wide daemon on the thin client.
> I already outlined the issues on Pulseaudio mailing list on this post:
> https://tango.0pointer.de/pipermail/pulseaudio-discuss/2008-May/001793.html,
> see also Lennart Poettering's answer:
> https://tango.0pointer.de/pipermail/pulseaudio-discuss/2008-May/001796.html.
> 
> I feel the proper solution would involve writing support for
> asynchronous notification into ALSA library and I/O-plugin layer.

I disagree. (See the other mail I just posted)

> Currently, Pulse I/O-plugin does not work in the Flash case, because
> Flash requires the asynchronous API.  On the other hand, Pulse requires
> the I/O-plugin layer to work.  See the posts by Alexander Indenbaum,
> http://mailman.alsa-project.org/pipermail/alsa-devel/2008-April/007033.html,
> and Takashi Iwai,
> http://mailman.alsa-project.org/pipermail/alsa-devel/2008-April/007335.html,
> on this subject.
> 
> I tried to tackle this issue, and did get the signal handling code to
> call the Flash audio callback, but I'm new to both ALSA and Pulseaudio,
> and it appears there are some issues with it I don't fully understand.
> Thus I'm asking: are there any other people who have been looking into
> this, or anyone who can tell whether this is a sane thing to do or not,
> will it require some nasty hack, or can help with some pointers to
> relevant information?

Actually I have been looking into this.

This doesn't really help however, since there's a unfixable race
condition in Flash shutting down audio. It's not going to be much
better than the libflashsupport code I came up with.

Adobe was informed about this. And in Flash 10 they responded to the
criticism. Flash 10 doesn't use alsa async anymore and seems to work
fine on ioplug and thus on PA.

Lennart

-- 
Lennart Poettering                        Red Hat, Inc.
lennart [at] poettering [dot] net         ICQ# 11060553
http://0pointer.net/lennart/           GnuPG 0x1A015CC4