Consider this pipeline:

 gst-launch-0.10 filesrc location=/some/file blocksize=100 ! identity datarate=100 ! fakesink sync=true -v

Despite timestamps and durations being set correctly, fakesink will not sync against the clock correctly from the second buffer onwards.

This, however, will work just fine:

 gst-launch-0.10 fakesrc sizemax=100 sizetype=fixed ! identity datarate=100 ! fakesink sync=true -v

The reason for this is that filesrc sends a newsegment event in BYTES format with a stop position that equals the filesize.

Now, gst_base_sink_get_sync_times() in do_times calls gst_segment_to_stream_time() and gst_segment_to_running_time() with the timestamps and GST_FORMAT_BYTES as format.

gst_segment_to_*_time(), however, will not pay attention to the fact that segment->format != format and goes on to return -1 here:

  if (segment->rate > 0.0) {
    /* outside of the segment boundary stop */
    if (G_UNLIKELY (segment->stop != -1 && position > segment->stop))
      return -1;

because the timestamp integer value is > the file size in bytes, so basesink gets unsyncable positions returned and skips the sync.


(This can all be made to work by using single-segment=true on identity, but that's not the point).