[jaunty/amd64] flash plugin no longer has any sound

[Bogdan Butnaru]

Binary package hint: flashplugin-nonfree

Hello!

I'm running an up-to-date Jaunty (except for the xorg packages, I still use the last version that supported the nvidia binary driver) on amd64 (actually, it's an Intel core2 duo).

The last update of the flashplugin-nonfree package (10.0.15.3ubuntu2) completely removed all sound from Flash. It seems to work correctly (videos play, etc), but I get no sound at all. I'm using the standard PulseAudio from the repositories. Nothing appears in pavucontrol when Flash is running.

I remember seeing in the release notes something about re-enabling nspluginwrapper, but I have no idea if that caused it. Before this update Flash 10 seemed to be working perfectly.

[Bogdan Butnaru (bogdanb)]

By the way, I've just installed http://download.macromedia.com/pub/labs/flashplayer10/libflashplayer-10.0.d21.1.linux-x86_64.so.tar.gz manually and it works perfectly. It appears as "ALSA plug-in [firefox]" in pavucontrol.

[Daniel T Chen (crimsun)]

Very quickly, I'll summarise what I told Alexander and Luke some time ago:

The most recent upload to jaunty of flashplugin-nonfree (to use nspluginwrapper for both i386 and amd64) has one important regression-related ramification for amd64: nspluginwrapper depends on ia32-libs, which currently conflicts with lib32asound2-plugins. This change means that PulseAudio users - i.e., all default Ubuntu 9.04 users - will have nondeterministically inaudible sound due to a race between the user's pulseaudio daemon and the Flash plugin attempting to use the hw: and default: virtual ALSA devices, respectively. In essence, this is a regression to hardy's release.

To fix the above symptom, three things need to be done:
1) pulseaudio needs to build 32-bit libs on 64-bit arches (similar to what alsa-lib does);
2) alsa-plugins needs to build 32-bit pulseaudio alsa-lib plugins based on (1);
3) ia32-libs needs to stop building 32-bit alsa-lib plugins and instead simply depend on (instead of conflicting with) lib32asound2-plugins.

Obviously, the appropriate versioned conflicts across all packages should be added.

[Steve Langasek (vorlon)]

This analysis makes no sense; flashplugin-nonfree on amd64 has *always* used ia32-libs, and it's orthogonal whether it uses it on i386 given that this bug report is against amd64. Furthermore, the ia32-libs package in both intrepid and jaunty include all of the pulseaudio libs and plugins. So what is the supposed "change" here?

[Bogdan Butnaru (bogdanb)]

I have another question: why exactly is flashplugin-nonfree using ndiswrapper on 64bit? I use the 64-bit release of Flash directly from Adobe, installed manually, and it work perfectly with Pulseaudio without ndiswrapper.

I have had zero crashes or sound problems with it ever since I removed the current version from the repositories. I can move its audio stream between sound devices, etc. (Also, the 64bit version from Ubuntu worked very well before adding ndiswrapper.) I just don't get why we wouldn't use a "clean" version for 64bit. (Actually, I'm not sure why ndiswrapper is used on 32 bits, either, but I don't have any 32-bit Jaunty machine to try it.)

[Daniel T Chen (crimsun)]

Steve, you're missing the fact that the upload of Flash 10 to jaunty (10.0.15.3ubuntu1) using native 64-bit did not, in fact, depend on nspluginwrapper for amd64. Native 64-bit plugin bypasses ia32-libs and avoids the symptom.

Currently, installing flashplugin-nonfree on amd64 installs the wrong version of the pulse pcm+ctl plugins via ia32-libs, a Depends of nspluginwrapper. This means that the pulse pcm+ctl plugins are not used. ia32-libs needs to be regenerated, and it would be swell to have the proper fix, which is to follow alsa-lib's biarch example, too.

The problem is due to lib32asound2-plugins conflicting with ia32-libs. They both contain the 32-bit alsa-plugins libraries. We can't simply have ia32-libs Depends lib32asound2-plugins, because alsa-plugins is a main source package, and ia32-libs (in universe) bits are needed to build alsa-plugins.

The change is to generate biarch libs in pulseaudio so that alsa-plugins can just Build-Depends the appropriate biarch -dev, and then ia32-libs can simply Depends the appropriate biarch lib32asound2-plugins.

[Daniel T Chen (crimsun)]

Bogdan, nspluginwrapper is used to isolate the (historically crash-happy) Flash instances. We are not permitted to distribute non-final versions of the plugin via the partner repository, which is the current download source regardless of arch.

I agree that having native versions provides a better user experience. The mozillateam is moving in that direction, but I have no ETA.

[Bogdan Butnaru (bogdanb)]

I'm not sure I understand what you mean by distributing non-final versions via partner; isn't the Flash plugin downloaded from Adobe at install time? And anyway, is only the 64-bit version in beta?

[Daniel T Chen (crimsun)]

flashplugin-nonfree downloads only the 32-bit from archive.canonical.com in the partner repository. The 64-bit is alpha (thus non-final).

Bogdan Butnaru <email address hidden> wrote:

>I'm not sure I understand what you mean by distributing non-final
>versions via partner; isn't the Flash plugin downloaded from Adobe at
>install time? And anyway, is only the 64-bit version in beta?
>
>--
>[jaunty/amd64] flash plugin no longer has any sound
>https://bugs.launchpad.net/bugs/314739
>You received this bug notification because you are a member of Ubuntu
>Audio Team, which is subscribed to pulseaudio in ubuntu.

[Luke Yelavich (themuso)]

At this point, getting bi-arch pulseaudio libraries is going to be a lot of work. In order to prevent pulseaudio libraries being linked against libraries that aren't needed, a substantial rewrite of autoconf/automake and possibly m4 code would be needed for the pulseaudio package. Even then, the pulseaudio libraries still require bi-arch equivalents for libX11.so.6 and maybe others.

So I am of the opinion that this option is a no go.

[Steve Langasek (vorlon)]

$ ldd -d -r /usr/lib32/libpulse.so.0
 linux-gate.so.1 => (0xf7fab000)
 libSM.so.6 => /usr/lib32/libSM.so.6 (0xf7f2b000)
 libICE.so.6 => /usr/lib32/libICE.so.6 (0xf7f13000)
 libX11.so.6 => /usr/lib32/libX11.so.6 (0xf7e23000)
 libcap.so.1 => /lib32/libcap.so.1 (0xf7e1f000)
 libgdbm.so.3 => not found
       ^^^^^^^^^^^^^^^^^^^

This is the real reason the alsa pulse plugin has stopped working in ia32-libs. It has nothing at all to do with "wrong versions" of the pulseaudio plugins (which would be a bug in its own right, since that would imply missing versioned conflicts/dependencies), nor would this be resolved by a biarch packaging of pulseaudio (unless the 32-bit PA build were to drop gdbm support), because the build-dependency would be missing!

I'll look at getting the missing libgdbm added to ia32-libs for now.

[Daniel T Chen (crimsun)]

Steve, yes, you're right of course. Thanks for following up!

[Launchpad Janitor (janitor)]

This bug was fixed in the package ia32-libs - 2.7ubuntu3

---------------
ia32-libs (2.7ubuntu3) jaunty; urgency=low

  * fetch-and-build:
    - add libgdbm3, needed as a dependency of libpulse. LP: #314739.
    - drop libxcb-xlib0, which is now obsolete.
    - replace libpulsecore8 with libpulsecore9

 -- Steve Langasek <email address hidden> Tue, 27 Jan 2009 00:34:20 +0000

[Id2ndR (id2ndr)]

This doesn't work for me anymore, but I use PulseAudio from PPA, do I think the bug is different. I switch to flash 64bits from adobe labs witch works better than the 32bits release for me.

[Iyeru (iyeru42)]

the white flash object issue is NOT a pulseaudio issue, it happens with esound as well. I'm currently running Jaunty 9.04

it happens on homestarrunner.com too. There are no reproduction steps, you just have to wait for it to become white.

Closing Firefox then re-opening firefox fixes the white box issue, but there has to be an easier way.

[Kip Warner (kip)]

I am getting the occasional empty white box as well. Usually restarting firefox fixes it temporarily. But as Iyeru mentioned, there has to be an easier way. Hopefully this will get patched soon if we can narrow down the problem (firefox, pulse, gstreamer, flash-nonfree, 64-bit emulation, etc.).

Kip

[Arne Nordmann (launchpad-norro)]

I can confirm this problem. After some time, there are just white boyes instead of the flash video objects. The time it occurs seems not reproducable to me. Neither it seems to be a certain time, nor a certain number of played flash videos nor a certain website/video.
Restarting firefox helps.

My system is an amd64 with up-to-date Jaunty, Firefox 3.5.3 and flashplugin-nonfree.