[Bug 760778] New: avenc_* audio encoders have (bad) channel-mask for mono audio in sink pad template

[GStreamer (GNOME Bugzilla)]

https://bugzilla.gnome.org/show_bug.cgi?id=760778

            Bug ID: 760778
           Summary: avenc_* audio encoders have (bad) channel-mask for
                    mono audio in sink pad template
    Classification: Platform
           Product: GStreamer
           Version: 1.4.5
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: Normal
         Component: gst-libav
          Assignee: gstreamer-bugs at lists.freedesktop.org
          Reporter: nyushchenko at dev.rtsoft.ru
        QA Contact: gstreamer-bugs at lists.freedesktop.org
     GNOME version: ---

$ gst-inspect avenc_mp2
...
  SINK template: 'sink'
    Availability: Always
    Capabilities:
      audio/x-raw
           channel-mask: 0x0000000000000004
               channels: 1
                   rate: { 48000, 44100, 32000, 24000, 22050, 16000 }
                 layout: interleaved
                 format: S16LE
...

This combination (channels=1, channel-mask=0x4) is not pleased by alsasrc:

$ gst-launch alsasrc ! audioconvert ! avenc_mp2 ! filesink location=/dev/null
Setting pipeline to PAUSED ...
Pipeline is live and does not need PREROLL ...
Setting pipeline to PLAYING ...
New clock: GstAudioSrcClock

** (gst-launch-1.0:492): CRITICAL **: file
/home/jenkins/nikita/tke-build/build/
tmp/work/cortexa9hf-vfp-neon-mx6qdl-poky-linux-gnueabi/gstreamer1.0-plugins-base
/1.4.5-r0/gst-plugins-base-1.4.5/gst-libs/gst/audio/gstaudioringbuffer.c: line
1
963 (gst_audio_ring_buffer_set_channel_positions): should not be reached

What happens is:

- on target board audio input is single-channel,

- at caps negotiation time, gst_audio_ring_buffer_parse_caps() for alsasrc is
called with caps originating from avenc_mp2 sink pad template,

- because channel-mask is defined, gst_audio_channel_positions_from_mask() is
used to compute channel positions for audio ringbuffer; with channel-mask=0x4
this results into position[0] set to GST_AUDIO_CHANNEL_POSITION_FRONT_CENTER,

- later, gst_audio_ring_buffer_set_channel_positions() is called from
alsasrc_parse_spec(); at this time alsa_position[alsa->channels-1] ipoint to
GST_AUDIO_CHANNEL_POSITION_MONO and rignbuffer's position[0] is
GST_AUDIO_CHANNEL_POSITION_FRONT_CENTER,

- gst_audio_ring_buffer_set_channel_positions() sees the difference nad calls
gst_audio_get_channel_reorder_map(),

- gst_audio_get_channel_reorder_map() returns FALSE because it refuces to work
with GST_AUDIO_CHANNEL_POSITION_MONO,

- this FALSE return causes the assertion failure.


As far as I understand, root cause of entire thing is that channel-mask should
not be defined for channels=1.

Channel-mask property on avenc_mp2's sink pad template originates from
gst_ff_aud_caps_new() in gst-libav:ext/libav/gstavcodecmap.c

This routine has two points where channel-mask property is added. First of
these is under "if (context->channels > 1)" condition but second is not.

I guess something like below patch is needed

diff --git a/ext/libav/gstavcodecmap.c b/ext/libav/gstavcodecmap.c
index 0c53d10..398aedd 100644
--- a/ext/libav/gstavcodecmap.c
+++ b/ext/libav/gstavcodecmap.c
@@ -634,7 +634,8 @@ gst_ff_aud_caps_new (AVCodecContext * context, AVCodec *
codec,
       while (*layouts) {
         gint nbits_set = get_nbits_set (*layouts);

-        if (gst_ffmpeg_channel_layout_to_gst (*layouts, nbits_set, pos)) {
+        if (nbits_set > 1 &&
+         gst_ffmpeg_channel_layout_to_gst (*layouts, nbits_set, pos)) {
           guint64 mask;

           if (gst_audio_channel_positions_to_mask (pos, nbits_set, FALSE,

... although not sure.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.