LMMS freezes when you try to render audio twice

[PaulBatchelor]

Sorry folks, I found another bug which I can reproduce. LMMS really doesn't like this laptop of mine!

When I bounce an audio file, then go to bounce it again, it freezes. The GDB backtrace would seem to suggest that it's some sort of threading race condition.

gdb.txt

On a related but unrelated note, is there someone I can talk to off github whose knowledgeable about the part s of the codebase in charge of offline rendering? My email is thisispaulbatchelor at gmail dot com.

[musikBear]

When I bounce an audio file, then go to bounce it again, it freezes

Can you explain 'Bounce' -What does you do exactly?

[PaulBatchelor]

Sorry, render audio.

[mikobuntu]

I have been having a similar render issue for quite some time too on master branch ( crash on render ). I haven't delved into debugging the cause yet. From memory I think it may be GUI related ( not sure, but will work on this tonight ) I have been building against QT5 on Linux , so to rule that out I will build with the defaults in another folder too. * note to myself ...test with all possible audio backends and config settings *

[PaulBatchelor]

@zonkmachine has been running into export crashes with #3403, which I believe are related to this issue.

I think the issue with this, (and any exporter related issues) is the renderer class itself, which I believe is causing a race condition with the audio mixer.

[softrabbit]

The discussion in #2457 might contain some hints.

[PhysSong]

@PaulBatchelor Can you still reproduce this issue with latest stable-1.2?

[PhysSong]

If you still able to do so, could you please post backtrace with all threads(using thread apply all bt) and the problematic project file?

[PaulBatchelor]

I am still able to reproduce this bug. I have attached both the project file and the backtrace below:
LMMS_3397.zip

[musikBear]

I cant reproduce with your file on win32 rc4 (std settings wave)
@PaulBatchelor
I miss flg. info:

• what do you export to
• do you use two different names, or do owerwrite with same name?

In fact, could you list your exact reprocurable method?

[PaulBatchelor]

I'm running on Slackware Linux 14.1 64-bit, and I am using JACK version 0.124.1.

Here are the steps I do to get the bug:

1. start LMMS
2. open song one_note.mmpz
3. select export from menu
4. save as "one_note.wav"
5. leave all default settings (on my machine: 24-bit 44.1Khz, sinc Medium, no oversampling). hit start.
6. select export from menu again
7. save as "one_note.wav". yes to overwrite.
8. leave all default settings. hit start.

At this point, LMMS freezes. I have also tried variations where I save to different filenames. Same results.

[PhysSong]

@PaulBatchelor Does LMMS still hangs if you use other audio backend?

[PaulBatchelor]

@PhysSong seems to work fine using ALSA.

[musikBear]

Linux issue, i step out

[PhysSong]

@PaulBatchelor Are you able to get backtrace using gdb with thread apply all bt? Providing other threads' state would be more useful.

seems to work fine using ALSA.

Anyway, does the first export breaks playback as well as the second export? Those results can be useful when investigating this issue.

[zonkmachine]

@PhysSong There's a backtrace zipped in this comment above: #3397 (comment)

[PhysSong]

Ah, thanks!

[PhysSong]

lmms/src/core/audio/AudioJack.cpp

Lines 399 to 403 in 55076d0

	m_framesToDoInCurBuf = getNextBuffer( m_outBuf );
	if( !m_framesToDoInCurBuf )
	{
	m_stopped = true;
	}

After rendering finishes, AudioJack's m_stopped is set to false. However, Mixer's FIFO buffer is empty at the same time. So JACK backend stops.

I think this condition should be removed and replaced properly.

[PhysSong]

FYI, #3207 might be fixed with the same way...