Consider a situation where a video decoder is very very slow, *much* slower than the desired framerate. The decoder supports QoS, yet still only the very first frame is ever rendered; no 'emergency renderings' are done, because all buffers but the very first are late and the code in gst_base_sink_is_too_late() goes like this:

     /* !!emergency!!, if we did not receive anything valid for more than a 
      * second, render it anyway so the user sees something */
     if (priv->last_in_time && start - priv->last_in_time > GST_SECOND) {
      late = FALSE;
      GST_DEBUG_OBJECT (basesink,
          "**emergency** last buffer at %" GST_TIME_FORMAT " > GST_SECOND",
          GST_TIME_ARGS (priv->last_in_time));
    }

The problem is the priv->last_in_time != 0 check here. priv->last_in_time is updated whenever a frame/buffer is actually rendered (normal or emergency), so if only the very first frame with a timestamp of 0 ever got rendered, no emergency renderings will be done ever again, unless a buffer actually arrives in time, which would set last_in_time to something non-0.

I think this should be:

     /* !!emergency!!, if we did not receive anything valid for more than a 
      * second, render it anyway so the user sees something */
-    if (priv->last_in_time && start - priv->last_in_time > GST_SECOND) {
+    if (priv->last_in_time != -1 && start - priv->last_in_time > GST_SECOND) {

instead (or maybe the last_in_time check should just be dropped at all?).

I will attach two logs that demonstrate the problem.