Bug 202204 – REGRESSION(r246399): [GStreamer] Problems playing AAC streams

• Component: WebKitGTK
• Status: RESOLVED FIXED

[Alberto Garcia]

This stream fails to play, or plays only for a few seconds:

http://tmp.xaq.nl/test-aac.html

These two work fine, but will occasionally fail if played *after* the AAC stream:

http://tmp.xaq.nl/test-mp3.html
http://tmp.xaq.nl/test-hls.html

Reportedly WebKitGTK 2.24.0 was the last one that didn't have problems with this.

I can reproduce the problem with 2.24.4, and it seems to fail with 2.26.1 as well.

Originally reported here: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=941137

[Carlos Alberto Lopez Perez]

I can't reproduce with 2.24.2, but I can with 2.24.3.
I can also reproduce with trunk.

So I guess the problem is caused by one of the patches back-ported to 2.24.3.
Looking at the list https://trac.webkit.org/wiki/WebKitGTK/2.24.x#Proposedmergesfor2.24.3 I see several GStreamer related patches there.

[Carlos Alberto Lopez Perez]

It seems it is caused by r246399

If I revert it on my local checkout (I'm on trunk@250519) I can't longer reproduce the issue.

[Philippe Normand]

I'll try to check it this week, thanks for the bisect Carlos!

[Philippe Normand]

Strange that commit introduced the regression for sure?

Here what I observe is that the internal queue gets drained just before the next data chunk arrives and that triggers EOS. However it shouldn't be triggered because this is a "live" non-seekable stream. I have a simple fix but I'll try to write a layout test as well, which will delay the upstreaming of the fix.

So here's the fix, in case downstream users are in urgent need of it:

diff --git a/Source/WebCore/platform/graphics/gstreamer/WebKitWebSourceGStreamer.cpp b/Source/WebCore/platform/graphics/gstreamer/WebKitWebSourceGStreamer.cpp
index 9810c257b37..965b6e276d0 100644
--- a/Source/WebCore/platform/graphics/gstreamer/WebKitWebSourceGStreamer.cpp
+++ b/Source/WebCore/platform/graphics/gstreamer/WebKitWebSourceGStreamer.cpp
@@ -451,11 +451,11 @@ static GstFlowReturn webKitWebSrcCreate(GstPushSrc* pushSrc, GstBuffer** buffer)
             GST_BUFFER_OFFSET(*buffer) = baseSrc->segment.position;
             GST_BUFFER_OFFSET_END(*buffer) = GST_BUFFER_OFFSET(*buffer) + size;
             GST_TRACE_OBJECT(src, "Buffer bounds set to %" G_GUINT64_FORMAT "-%" G_GUINT64_FORMAT, GST_BUFFER_OFFSET(*buffer), GST_BUFFER_OFFSET_END(*buffer));
-            GST_TRACE_OBJECT(src, "doesHaveEOS: %s, wasSeeking: %s, seeking: %s, size: %u", boolForPrinting(priv->doesHaveEOS), boolForPrinting(priv->wasSeeking), boolForPrinting(priv->isSeeking), size);
+            GST_TRACE_OBJECT(src, "doesHaveEOS: %s, wasSeeking: %s, seeking: %s, buffer size: %u, size: %" G_GUINT64_FORMAT, boolForPrinting(priv->doesHaveEOS), boolForPrinting(priv->wasSeeking), boolForPrinting(priv->isSeeking), size, priv->size);
             if (priv->haveSize && GST_BUFFER_OFFSET_END(*buffer) >= priv->size) {
                 if (priv->wasSeeking)
                     priv->wasSeeking = false;
-                else
+                else if (priv->isSeekable)
                     priv->doesHaveEOS = true;
             } else if (priv->wasSeeking)
                 priv->wasSeeking = false;

[Carlos Alberto Lopez Perez]

(In reply to Philippe Normand from comment #4)
> Strange that commit introduced the regression for sure?
> 

I'm not sure if it was the one that introduced the regression, or the one that seems to work-around the issue if reverted. I didn't a full bisect.
I just tested to revert it on my local branch, and observed that after reverting it the issue seems fixed.

> Here what I observe is that the internal queue gets drained just before the
> next data chunk arrives and that triggers EOS. However it shouldn't be
> triggered because this is a "live" non-seekable stream. I have a simple fix
> but I'll try to write a layout test as well, which will delay the
> upstreaming of the fix.
> 
> So here's the fix, in case downstream users are in urgent need of it:
> 
> diff --git
> a/Source/WebCore/platform/graphics/gstreamer/WebKitWebSourceGStreamer.cpp
> b/Source/WebCore/platform/graphics/gstreamer/WebKitWebSourceGStreamer.cpp
> index 9810c257b37..965b6e276d0 100644
> --- a/Source/WebCore/platform/graphics/gstreamer/WebKitWebSourceGStreamer.cpp
> +++ b/Source/WebCore/platform/graphics/gstreamer/WebKitWebSourceGStreamer.cpp
> @@ -451,11 +451,11 @@ static GstFlowReturn webKitWebSrcCreate(GstPushSrc*
> pushSrc, GstBuffer** buffer)
>              GST_BUFFER_OFFSET(*buffer) = baseSrc->segment.position;
>              GST_BUFFER_OFFSET_END(*buffer) = GST_BUFFER_OFFSET(*buffer) +
> size;
>              GST_TRACE_OBJECT(src, "Buffer bounds set to %" G_GUINT64_FORMAT
> "-%" G_GUINT64_FORMAT, GST_BUFFER_OFFSET(*buffer),
> GST_BUFFER_OFFSET_END(*buffer));
> -            GST_TRACE_OBJECT(src, "doesHaveEOS: %s, wasSeeking: %s,
> seeking: %s, size: %u", boolForPrinting(priv->doesHaveEOS),
> boolForPrinting(priv->wasSeeking), boolForPrinting(priv->isSeeking), size);
> +            GST_TRACE_OBJECT(src, "doesHaveEOS: %s, wasSeeking: %s,
> seeking: %s, buffer size: %u, size: %" G_GUINT64_FORMAT,
> boolForPrinting(priv->doesHaveEOS), boolForPrinting(priv->wasSeeking),
> boolForPrinting(priv->isSeeking), size, priv->size);
>              if (priv->haveSize && GST_BUFFER_OFFSET_END(*buffer) >=
> priv->size) {
>                  if (priv->wasSeeking)
>                      priv->wasSeeking = false;
> -                else
> +                else if (priv->isSeekable)
>                      priv->doesHaveEOS = true;
>              } else if (priv->wasSeeking)
>                  priv->wasSeeking = false;

I tested this patch and it also fixes the issue :)

[Philippe Normand]

Created attachment 380332 [details]
Patch

[Philippe Normand]

I haven't managed to write a reliable test unfortunately.

[Philippe Normand]

Committed r250901: <https://trac.webkit.org/changeset/250901>

[Alberto Garcia]

This may not have been fixed after all (or we have a different bug).

Here's a new test case:

   http://tmp.xaq.nl/test-aac.html

Leave it playing normally, and after around 2 minutes of playback you'll notice some hiccups and it will eventually stop completely.

Reproduced with the MiniBrowser and WebKitGTK 2.26.2

---

In case the URL I pasted earlier changes, the test case contains this single <audio> element:

<audio id="player" controls>
 <source src="http://icecast.omroep.nl/radio4-eigentijds-aac" type="audio/aac" />
  <p>Your browser does not support the audio element.</p>
</audio>

[Philippe Normand]

I can't reproduce this issue in trunk. It happens in Ephy TP though, it's not related with EOS, so it's a new bug. Please file it separately and attach logs, as usual with media bugs:

GST_DEBUG_FILE=gst.log GST_DEBUG="3,webkit*:8"

[Alberto Garcia]

New bug filed as https://bugs.webkit.org/show_bug.cgi?id=204410