swh / lv2 Goto Github PK

The vynil plugin is making noise with crackle even when all playback is stopped and all plugin parameters are set to default.

The screenshot was taken when the playback was stopped in Ardour.

I'm using a version of plugins from Ubuntu package in case that's relevant. Trying to load LV2 version of Higher Quality Pitch Shifter results in this error:

LADSPA version loads without issues.

After building the latest snapshot, I get undefined symbols. See:
https://gist.github.com/135612/c06b3bca480b0f46fadad14377b2675b3ab290e6

This patch resolves this issue:
http://gist.github.com/135618/d7f133882dc8ae6e67f0118549320e59c3cb2987

(It is a pain to add attachments to bug reports here. Is there an easier way?)

A second patch I made corrects permissions (and timestamps) of the installed files.

http://github.com/oget/lv2/commit/8dacda442dd10b5269288aee034d4956926dd4de

I'd be happy if you merge this too.

Many of the routines in the plugins allocate memory in the activate() method that is freed in the cleanup() method. If the activate() method is never called (say, because the plugin was instantiated, but never used), then the program that loaded the plugin will crash.

For example, in 'plugins/allpass-swh.lv2/plugin.xml', the activate() method allocates memory:

plugin_data->buffer = calloc(size, sizeof(float));

... which isn't cleaned up until cleanup() is called:

free(plugin_data->buffer);

Note that this strategy also causes the plugins to leak memory, as a plugin could be activated and deactivated multiple times before cleanup() is called.

One of the following should occur:

1.) The memory should be allocated when the instance is first created.
2.) The memory should be freed in the deactivate() method.
3.) The cleanup() method should check to make sure that memory was actually allocated before freeing that memory.

(1) and (2) are ideal.

I have swh-lv2-1.0.16 in gentoo, and use global make options -j3 -l1.8.
Reported to them as https://bugs.gentoo.org/show_bug.cgi?id=687924

Initially I thought it is failure to handle multiple jobs, but then made experiments with fresh git trunk, and found that only some -j -l values fail.

In particular, -j3 -l1.8, -j3 -l3 and -j10 -l2.
However, simple -jN (with any N) and -j3 -l5 are ok.

When I tried to install releases, I made two attempts. First time it failed missing gverb/gverbdsp.o, and second time - at util/iir.o. But when I experimented with git trunk, only util/iir.o was missed all times.

I originally filed this bug to zynjacku:
https://gna.org/bugs/?13688

The problem was, loading some of the lv2-swh plugins via zynjacku resulted in segmentation faults, while the plugins work fine with lv2_jack_host (from slv2) or ardour.

I have been asked to file this bug here. I went through all the plugins one-by-one, and meanwhile I noticed that some of them crash with even lv2_jack_host (which I misreported before). What is not listed in the below list works fine.

Note that, this test is made against the latest trunk snapshot + the patch I submitted in
http://github.com/swh/lv2/issues#issue/1
(otherwise more plugins will fail due to missing symbols)

• Allpass delay line, cubic spline interpolation:

• crashes with both lv2_jack_host, zynjacku

  [Switching to Thread 0x7fffed4bb910 (LWP 27663)]

  0x00007fffec3a0d14 in runAllpass_c (instance=0x1967a20, sample_count=)

  at plugins/allpass-swh.lv2/plugin.c:637

637 float read = cube_interp (frac,

• Allpass delay line, linear interpolation:
• crashes with both lv2_jack_host, zynjacku

[Switching to Thread 0x7fffed4bb910 (LWP 28763)]

0x00007fffec3a2433 in runAllpass_l (instance=0x196a050, sample_count=)

at plugins/allpass-swh.lv2/plugin.c:429

429 float r1 = buffer[read_phase & buffer_mask];

• Allpass delay line, noninterpolating:
• crashes with both lv2_jack_host, zynjacku
  [Switching to Thread 0x7fffed4bb910 (LWP 30129)]

runAllpass_n (instance=0x1968710, sample_count=1024) at plugins/allpass-swh.lv2/plugin.c:192

192 float read = *(readptr++);

• Bode frequency shifter (CV):
• crashes with zynjacku only

[Switching to Thread 0x7fffed4bb910 (LWP 2471)]

runBodeShifterCV (instance=0x173af90, sample_count=)

at plugins/bode_shifter_cv-swh.lv2/plugin.c:205

205 buffer_write(mixout[pos], (dout[pos] - uout[pos]) * mixc + uout[pos]);

• Comb delay line, cubic spline interpolation:
• crashes with both lv2_jack_host, zynjacku

[Switching to Thread 0x7fffed4bb910 (LWP 4036)]

0x00007fffec19ecb4 in runComb_c (instance=0x19692b0, sample_count=)

at plugins/comb-swh.lv2/plugin.c:633

633 float read = cube_interp (frac,

• Comb delay line, linear interpolation:
• crashes with both lv2_jack_host, zynjacku

[Switching to Thread 0x7fffed4bb910 (LWP 6545)]

runComb_l (instance=0x1968050, sample_count=)

at plugins/comb-swh.lv2/plugin.c:425

425 float r1 = buffer[read_phase & buffer_mask];

• Comb delay line, noninterpolating:
• crashes with both lv2_jack_host, zynjacku

[Switching to Thread 0x7fffed4bb910 (LWP 8806)]

runComb_n (instance=0x1966bf0, sample_count=)

at plugins/comb-swh.lv2/plugin.c:189

189 float read = *(readptr++);

• Crossfade
• crashes with zynjacku only

[Switching to Thread 0x7fffed4bb910 (LWP 10143)]

runXfade (instance=0x1a428d0, sample_count=)

at plugins/xfade-swh.lv2/plugin.c:83

83 buffer_write(outputL[pos], inputLA[pos] * coefA + inputLB[pos] * coefB);

• Crossfade (4 outs)
• crashes with zynjacku only

[Switching to Thread 0x7fffed4bb910 (LWP 9841)]

runXfade4 (instance=0x196a000, sample_count=)

at plugins/xfade-swh.lv2/plugin.c:194

194 buffer_write(outputLB[pos], inputLB[pos] * coefB);

• Glame bandpass analog filter:
• crashes with both lv2_jack_host, zynjacku

[Switching to Thread 0x7fffed4bb910 (LWP 10800)]

calc_2polebandpass (iirf=0x736e6967756c702d, gt=0x736170646e61622f,
fc=0.112575002, bw=0.225050002, sample_rate=44100) at util/iir.c:102

102 if ( (gt->fc==fc) && (gt->bw==bw) )

• Glame bandpass filter:
• crashes with both lv2_jack_host, zynjacku

[Switching to Thread 0x7fffed4bb910 (LWP 13459)]

chebyshev (iirf=0x0, gt=0x0, n=2, mode=1, fc=2.55158739e-06, pr=0.5) at util/iir.c:213

213 if ( (gt->fc==fc) && (gt->np==n) && (gt->ppr=pr) )

• GLAME butterworth lowpass/highpass/X-over Filter:
• crashes with both lv2_jack_host, zynjacku

[Switching to Thread 0x7fffed4bb910 (LWP 16152)]

butterworth_stage (sample_rate=, r=, f=, mode=, gt=)

at util/iir.h:130

130 gt->coeff[0][0] = a1;

• Glame Highpass/Lowpass Filter:
• crashes with both lv2_jack_host, zynjacku

[Switching to Thread 0x7fffed4bb910 (LWP 18643)]

chebyshev (iirf=0x736170686769682f, gt=0x7875007269695f73, n=2, mode=1, fc=2.55272107e-06,
pr=0.5) at util/iir.c:213

213 if ( (gt->fc==fc) && (gt->np==n) && (gt->ppr=pr) )

• Reverse delay (5s max):
• crashes with both lv2_jack_host, zynjacku

[Switching to Thread 0x7fffed4bb910 (LWP 22139)]

0x00007fffdc5c3903 in runRevdelay (instance=0x1b4cbe0, sample_count=1024)

at plugins/revdelay-swh.lv2/plugin.c:167

167 read = (wet * buffer[read_phase]) + (dry * insamp);

• SC3
• crashes with zynjacku only

[Switching to Thread 0x7fffed4bb910 (LWP 25952)]

runSc3 (instance=0x16c6710, sample_count=)

at plugins/sc3-swh.lv2/plugin.c:175

175 + chain_bal * sidechain[pos];

• Simple delay line, cubic spline interpolation:
• crashes with both lv2_jack_host, zynjacku
  [Switching to Thread 0x7fffed4bb910 (LWP 27996)]

runDelay_c (instance=0x1747d60, sample_count=)

at plugins/delay-swh.lv2/plugin.c:520

520 float read = cube_interp (frac,

• Simple delay line, linear interpolation:
• crashes with both lv2_jack_host, zynjacku

[Switching to Thread 0x7fffed4bb910 (LWP 30899)]

0x00007fffec3a1cfe in runDelay_l (instance=0x1968000, sample_count=)

at plugins/delay-swh.lv2/plugin.c:338

338 read = LIN_INTERP (frac,

• Simple delay line, noninterpolating:
• crashes with both lv2_jack_host, zynjacku

[Switching to Thread 0x7fffed4bb910 (LWP 32669)]

runDelay_n (instance=0x1969510, sample_count=)

at plugins/delay-swh.lv2/plugin.c:155

155 float read = *(readptr++);

• Surround matrix encoder
• crashes with zynjacku only

[Switching to Thread 0x7fffed4bb910 (LWP 1423)]

runSurroundEncoder (instance=, sample_count=)

at plugins/surround_encoder-swh.lv2/plugin.c:158

158 delay[dptr] = s[pos];

Please tag a stable release.
This would make it easier to package the plugins.
Some package management systems (e.g. homebrew) do not accept packages with no tagged versions.

==28903== Invalid free() / delete / delete[] / realloc()
==28903==    at 0x4C2AEAB: free (vg_replace_malloc.c:530)
==28903==    by 0x4A03F8DC: ??? (in /usr/lib/lv2/revdelay-swh.lv2/plugin-linux.so)
==28903==    by 0xE912082: lilv_instance_free (in /usr/lib/liblilv-0.so.0.22.0)
==28903==    by 0x6972404: ARDOUR::LV2Plugin::cleanup() (lv2_plugin.cc:1927)
==28903==    by 0x696C210: ARDOUR::LV2Plugin::~LV2Plugin() (lv2_plugin.cc:721)
==28903==    by 0x696C563: ARDOUR::LV2Plugin::~LV2Plugin() (lv2_plugin.cc:752)

==28903== Conditional jump or move depends on uninitialised value(s)
==28903==    at 0x4C2AE61: free (vg_replace_malloc.c:530)
==28903==    by 0x4A03F8DC: ??? (in /usr/lib/lv2/revdelay-swh.lv2/plugin-linux.so)
==28903==    by 0xE912082: lilv_instance_free (in /usr/lib/liblilv-0.so.0.22.0)
==28903==    by 0x6972404: ARDOUR::LV2Plugin::cleanup() (lv2_plugin.cc:1927)
==28903==    by 0x696C210: ARDOUR::LV2Plugin::~LV2Plugin() (lv2_plugin.cc:721)
==28903==    by 0x696C563: ARDOUR::LV2Plugin::~LV2Plugin() (lv2_plugin.cc:752)

Sorry no debug-build, but this really smells like some uninitialized pointer variable.
If needed I can do with a debug build, but really just calloc allocate the instance or memset will solve this.

lowpass_iir-swh.lv2 has a similar issue (no crash), valgrind reports

==28903== Conditional jump or move depends on uninitialised value(s)
==28903==    at 0xF5352EF: __sin_avx (in /lib/x86_64-linux-gnu/libm-2.21.so)
==28903==    by 0xF4F01DA: sincos (in /lib/x86_64-linux-gnu/libm-2.21.so)
==28903==    by 0x50884EE5: chebyshev_stage (in /usr/lib/lv2/lowpass_iir-swh.lv2/plugin-linux.so)
==28903==    by 0x508856C9: chebyshev (in /usr/lib/lv2/lowpass_iir-swh.lv2/plugin-linux.so)
==28903==    by 0x69688EF: lilv_instance_activate (lilv.h:1648)

Example from the generated GLAME Butterworth Lowpass ttl:

 :port [
     a :InputPort, :ControlPort ;
     :name "Cutoff Frequency (Hz)" ;
     :index 0 ;
     :symbol "cutoff" ;
     :minimum 0.0001 ;
     :maximum 0.45 ;
     
     :default 0.112575 ;
     :portProperty pprops:logarithmic ;
     :portProperty :sampleRate ;
   ] ;

Here, sampleRate tells us that minimum and maximum shall be multiplied with the processing sample rate, i.e. minimum and maximum in range [0,1] makes sense. However, default will not be multiplied, so the default value must be in range [minimum*sampleRate, maximum*sampleRate].

Thanks for @PhysSong who posted a comment about it, and to @drobilla for confirming that this default value is not valid.

Hi,

I know of your plugins via the Ardour program. Thanks for the great work!

I’m running into some issues though. I asked on IRC #ardour-osx , and they said there are some known issues with the plugins under os x.

This is what I encountered thus far:

When trying to load Multiband EQ Ardour consistently crashes:
https://gist.github.com/raw/757211/079bc782aff25f4a46216513d555fd88ed026ff5/ardour-crash.log

SC4 doesn’t affect it’s input when loaded by itself, adding a bypassed instance of SC1 changes this, like:
http://ardour.org/node/2176

I was wondering if you could look into this.
thanks, Eric

I like the "Artificial latency" plugin to eliminate latencies that were introduced during recording (multiple microphones on a drum kit, for example).

However, it would be more intuitive (to me) to
a) set the delay that I want to apply and not the delay the signal has
b) also be able to set negative values
and c) set the delay in samples because it can be measured much more accurate.

I did a quick experiment and came up with this trivial plugin. It has all the properties mentioned above and I tested it in Ardour3.

Would you be willing to include such a plug-in in your collection? I'd be happy to create a corresponding pull request if you like.

The sidechain inputs should have a lv2:portProperty lv2:isSideChain; port property.
Currently they're only in a custom group pg:inGroup swh:sc3-sidechain ; and "swh:sc3-sidechain" is not meaningful to a host.

expected (here sc3-swh.lv2/plugin.ttl).

   :port [
     a :InputPort, :AudioPort ;
     :name "Sidechain" ;
     :index 7 ;
     :symbol "sidechain" ;
     :portProperty :isSideChain;
     pg:inGroup swh:sc3-sidechain ;
     pg:role pg:centerChannel ;
   ] ;