matroskademux: properly resume cluster scanning

... rather than getting offset tracking messed up, and then likely
failing a subsequent assert.
This commit is contained in:
Mark Nauwelaerts 2011-02-09 16:22:47 +01:00
parent a86bab6689
commit 9de84b45bc

View file

@ -2353,13 +2353,13 @@ gst_matroska_demux_search_cluster (GstMatroskaDemux * demux, gint64 * pos)
GST_DEBUG_OBJECT (demux, "read buffer size %d at offset %" G_GINT64_FORMAT,
GST_BUFFER_SIZE (buf), newpos);
gst_byte_reader_init_from_buffer (&reader, buf);
cluster_pos = 0;
resume:
cluster_pos = gst_byte_reader_masked_scan_uint32 (&reader, 0xffffffff,
GST_MATROSKA_ID_CLUSTER, cluster_pos,
GST_BUFFER_SIZE (buf) - cluster_pos);
GST_MATROSKA_ID_CLUSTER, 0, gst_byte_reader_get_remaining (&reader));
if (cluster_pos >= 0) {
newpos += cluster_pos;
/* prepare resuming at next byte */
gst_byte_reader_skip (&reader, cluster_pos + 1);
GST_DEBUG_OBJECT (demux,
"found cluster ebml id at offset %" G_GINT64_FORMAT, newpos);
/* extra checks whether we really sync'ed to a cluster:
@ -2399,7 +2399,7 @@ gst_matroska_demux_search_cluster (GstMatroskaDemux * demux, gint64 * pos)
goto resume;
} else {
/* partial cluster id may have been in tail of buffer */
newpos += MAX (GST_BUFFER_SIZE (buf), 4) - 3;
newpos += MAX (gst_byte_reader_get_remaining (&reader), 4) - 3;
gst_buffer_unref (buf);
buf = NULL;
}