It seems to me that there was no leak here. Am I missing something ?
This patch just repeats the peercaps test which was done before already.

Comment on attachment 204618 [details] [review]
patch

I don't see the potential leak here either, and as far as I can tell the new version is functionally 100% equivalent to the old version.

What if peercaps == GST_CAPS_ANY ?

if (peercaps && !gst_caps_is_any (peercaps)) {
    /* get intersection */
    caps =
        gst_caps_intersect_full (peercaps, thiscaps, GST_CAPS_INTERSECT_FIRST);
    GST_DEBUG_OBJECT (basesrc, "intersect: %" GST_PTR_FORMAT, caps);
    gst_caps_unref (peercaps);
  } else {
    /* no peer, work with our own caps then */
    caps = gst_caps_copy (thiscaps);
  }
}

It would go into the "else", and there peercaps would not be freed?

Ah, yes. But the code upstream (both git and 0.10.35) is:

  2663	  /* get the peer caps */
  2664	  peercaps = gst_pad_peer_get_caps_reffed (GST_BASE_SRC_PAD (basesrc));
  2665	  GST_DEBUG_OBJECT (basesrc, "caps of peer: %" GST_PTR_FORMAT, peercaps);
  2666	  if (peercaps) {
  2667	    /* get intersection */
  2668	    caps =
  2669	        gst_caps_intersect_full (peercaps, thiscaps, GST_CAPS_INTERSECT_FIRST);
  2670	    GST_DEBUG_OBJECT (basesrc, "intersect: %" GST_PTR_FORMAT, caps);
  2671	    gst_caps_unref (peercaps);
  2672	  } else {
  2673	    /* no peer, work with our own caps then */
  2674	    caps = gst_caps_copy (thiscaps);
  2675	  }
  2676	  gst_caps_unref (thiscaps);

Created attachment 205151 [details] [review]
patch

basesrc: minor negotiation optimization

I see what has happen. I forgot about a local modification that speeds up caps-nego a bit (attached), and hence the diverge... Sorry about that.

in 0.11 the intersection uses refcounting instead of copy and the filter is done in the getcaps call. For 0.10, I wonder if it is still worth it..

In fact, intersect also just does a copy when caps2 is ANY so this patch doesn't reduce any complexity. closing.