Reproducing is really easy (if you have the hw):

  videotestsrc ! v4l2videoMtransform ! fakesink

Created attachment 328264 [details] [review]
gstv4l2transform: format fixation for preferring passthrough

* If outgoing format is unfixated, try to set it to input format.

* Call gst_caps_fixate () at end of fixation routine

Review of attachment 328264 [details] [review]:

Looks good to me, the extra code to fixate to input format seems useful considering we also do scaling. In videoconvert they got one extra step which I don't yet understand:

+  if (direction == GST_PAD_SINK) {
+    if (gst_caps_is_subset (caps, othercaps)) {
+      gst_caps_replace (&othercaps, caps);
+    }
+  }


I'd like to know why they do that and if we should be doing the same here.

Is this related to https://bugzilla.gnome.org/show_bug.cgi?id=748141 ?

Review of attachment 328264 [details] [review]:

commit 5829f32

Should go in 1.8, as it fixes a regression, I had added the other change as a separate patch.

commit f9e01a5

*** Bug 766588 has been marked as a duplicate of this bug. ***

Branch: 1.8
Commit: bca2655f4a38a6a6dd06dd7ee22d80c9e1b2591e