[Bug 29056] New: Program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

Bug #: 29056
           Summary: Program freezes after sndPlaySound
           Product: Wine
           Version: 1.3.32
          Platform: x86-64
        OS/Version: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: winmm&mci
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]
    Classification: Unclassified

Created attachment 37441
  --> http://bugs.winehq.org/attachment.cgi?id=37441
test program for issue

1cde966c35dcb0b5dc7bc5a6c42b788f6d8f561c introduces a regression for the test
program attached.

Before this, Wine would repeatedly play the sound when the button is clicked.
After this commit, the first click plays the sound correctly. The second click
freezes the program.

Cheers

[Bug 29056] Program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

Ross Lagerwall  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]
                   |                            |t
    Regression SHA1|                            |1cde966c35dcb0b5dc7bc5a6c42
                   |                            |b788f6d8f561c

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

Dmitry Timoshkov  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Program freezes after       |Test program freezes after
                   |sndPlaySound                |sndPlaySound

--- Comment #1 from Dmitry Timoshkov  2011-11-11 01:08:54 CST ---
Jörg Höhle     
Mon, 19 Sep 2011 12:30:58 +0000 (14:30 +0200)

winmm: Never write silence, mmdevapi must handle underruns.

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

Ross Lagerwall  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                URL|                            |http://numberwise.com/files
                   |                            |/Numberwise-1.13.exe

--- Comment #2 from Ross Lagerwall  2011-11-11 02:07:44 CST ---
Note that this error actually occurs in a real program: Numberwise -
http://numberwise.com/files/Numberwise-1.13.exe

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

GyB  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #3 from GyB  2011-11-11 02:51:47 CST ---
No freezing occurs here in the test program when Alsa is in use.
When PulseAudio is running there is also no problem here.
The only way to reproduce the hanging issues is to suspend PA prior to the
test: pasuspender wine Project1.exe. Afterwards the test program indeed
freezes. To recover from this is to restart the PA daemon.
Tested with 1.3.32 and wine-1.3.32-122-g2e5f73e

Fedora 16 x86
Alsa 1.0.24
PulseAudio 0.9.23
Kernel 3.1.0-7.fc16.i686.PAE
Audio device: nVidia Corporation MCP61 High Definition Audio (rev a2)

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

Henri Verbeet  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |regression

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

--- Comment #4 from Jörg Höhle  2011-11-11 06:48:53 CST ---
Is the sound to play a tiny one, shorter than ALSA's period size?
If so, then it's my prediction from bug #27087, comment #17 come true.
> The only way to reproduce the hanging issues is to suspend PA

Like in bug #27087, comment #12?

We need to make progress with bug #28723 in order to decide how to handle
ALSA's period.  Once that is done, we may pad small initial writes with silence
to force ALSA to start (requires ALSA's period <= mmdevapi buffer size).

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

--- Comment #5 from GyB  2011-11-11 07:27:36 CST ---
(In reply to comment #4)
> Is the sound to play a tiny one, shorter than ALSA's period size?
> If so, then it's my prediction from bug #27087, comment #17 come true.
> > The only way to reproduce the hanging issues is to suspend PA
>
> Like in bug #27087, comment #12?

Yes, it seems to be the same issue: yesterday I tested the file
UIButtonPress.wav (in bug #27087) with your wintest.exe mcishell, and it
behaved in the same way: no freezing when Alsa is in use, no freezing when PA
is running. It does freezes when PA is suspended.

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

--- Comment #6 from Ross Lagerwall  2011-11-11 10:03:36 CST ---
My system is:
OpenSUSE 11.4 x86_64 (with 32bit libs)
Alsa 1.0.24
Pulseaudio 0.9.22
Kernel 2.6.37.6-0.9-desktop
Audio device: Intel Corporation N10/ICH 7 Family High Definition Audio
Controller (rev 01)

Running with pasuspender results in no sound the first click and a freeze on
the second click, even for versions *before* 1cde966c35.
Running without pasuspender results in sound for the first click and a freeze
on the second click for versions 1cde966c35 and after. It results in correct
behavior for versions before 1cde966c35.

This seems to be regardless of the length of the sound.

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

--- Comment #7 from GyB  2011-11-11 10:22:47 CST ---
(In reply to comment #6)
> Running with pasuspender results in no sound the first click and a freeze on
> the second click, even for versions *before* 1cde966c35.
> Running without pasuspender results in sound for the first click and a freeze
> on the second click for versions 1cde966c35 and after. It results in correct
> behavior for versions before 1cde966c35.
>
> This seems to be regardless of the length of the sound.

After the first usage of pasuspender you need to reload the PA daemon,
otherwise the freezing will occur, whether you use pasuspender in the next test
or not. 'killall pulseaudio' then 'pulseaudio -D'.
At least that's what I experienced on my system (I'm not familiar with PA,
always using ALSA).

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |download

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

--- Comment #8 from Ross Lagerwall  2011-11-12 08:23:57 CST ---
I tested the issue on Mandriva 2011 64 bit and it does not seem to be exist so
maybe it is something specific to the OpenSUSE setup.

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

Jörg Höhle  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |1.4.0

--- Comment #9 from Jörg Höhle  2011-12-12 06:52:56 CST ---
Ross, please compare your Mandriva and OpenSuse setups.  Please report PA and
alsa_plugins versions.
Use cat /proc/asound/card0/pcm0c/sub0/?w_params to see which parameters the
ALSA back end (last before speaker) uses, and the Wine log files to see which
parameters the front end (likely PulseAudio or dmix) uses.  The front end
matters actually.

I want to know whether this issue is solely about the ALSA bug that it won't
start when fed less than alsa_period_size samples (comment #4) or whether
something else is involved, e.g some strange pasuspender issue.
> This seems to be regardless of the length of the sound.

In the case with pasuspender or without?

What I believe should be observed:
- Without pasuspender: New PulseAudio (alsa plugins 1.0.24).  Might work.
- Without pasuspender: Old PulseAudio is running.  One click
  (might start but not with short samples), then hangs (PA underrun bug).
- With pasuspender: Dmix is running, no PA.  No click, hangs (never starts)
with short samples; long samples work.

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

--- Comment #10 from Ross Lagerwall  2011-12-12 21:33:43 CST ---
Hi,

I will test this when I have some time and report back the results.

Cheers
Ross

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

Jörg Höhle  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |DUPLICATE

--- Comment #11 from Jörg Höhle  2012-01-18 00:35:50 CST ---
(In reply to GyB's comment #7)
> After the first usage of pasuspender you need to reload the PA daemon,
> otherwise the freezing will occur, whether you use pasuspender in the next test
> or not. 'killall pulseaudio' then 'pulseaudio -D'.

This is indeed bogus behaviour of pasuspender observable by me in Ubuntu Lucid.
Any sound output to  "default" is dead afterwards, incl. speaker-test (whereas
speaker-test -c2 -Dhw:0 still works, the device is not locked).
In the older Ubuntu Intrepid, pasuspender works as its manpage says and you
need not restart PA.

(In reply to Ross' comment #8)
> I tested the issue on Mandriva 2011 64 bit and it does not seem to be exist so
> maybe it is something specific to the OpenSUSE setup.

Please file a bug report against PA or your Linux distribution if that bug
persists with the most recent PA/alsa_plugins.
> Running without pasuspender results in sound for the first click and a freeze
> on the second click for versions 1cde966c35 and after.

In summary, scenarios involving pasuspender (as opposed to not using PA at all)
are likely consequences of that bug and should be marked invalid here. Because
it's not cluttered by pasuspender issues, let's make wine bug #29299 the one
about the regression caused by commit 1cde966c35dcb0b5dc7bc5a6c42b788f6d8f561c
winmm: Never write silence, mmdevapi must handle underruns.
As a consequence:
- ALSA may not start short samples < period size
  This seems to only affect period-oriented devices like ALSA+dmix or hw:0,
  whereas ALSA+PA starts with the first sample written.
- and it does not play trailing samples.

*** This bug has been marked as a duplicate of bug 29299 ***

[Bug 29056] Test program freezes after sndPlaySound

http://bugs.winehq.org/show_bug.cgi?id=29056

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |CLOSED

--- Comment #12 from Austin English  2012-01-23 19:46:23 CST ---
Closing DUPLICATE.

[Bug 29056] Numberwise 1.13 freezes after sndPlaySound since Wine 1.3.32

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=29056

Anastasius Focht  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Test program freezes after  |Numberwise 1.13 freezes
                   |sndPlaySound                |after sndPlaySound since
                   |                            |Wine 1.3.32
                URL|http://numberwise.com/files |https://web.archive.org/web
                   |/Numberwise-1.13.exe        |/20210314105217/https://www
                   |                            |.numberwise.com/files/Numbe
                   |                            |rwise-1.13.exe
          Component|winmm&mci                   |winealsa.drv
                 CC|                            |[e-mail]