Improve latency by waiting for the next keyframe

Instead of caching all the data from the previous keyframe till right now and
sending it to the client in a burst, since latency is important, we instead wait
for the next keyframe before sending buffers to the clients. This will lead to
a delay of upto 128 frames before the stream starts.

The keyframe distance and hence the delay can be tweaked by setting the
"keyframe-max-dist" property on vp8enc in src/encode.c:create_webm_profile()

1 files changed, 1 insertions, 2 deletions

diff --git a/src/lib.h b/src/lib.h
index db57354..6894c0b 100644
--- a/src/lib.h
+++ b/src/lib.h
@@ -34,8 +34,6 @@ struct _TranscodeServerCtx {
   SoupMessage *msg;
   GstElement *pipeline;
   GstElement *appsrc;
-  /* Contains buffers from the previous keyframe till right now */
-  GstBufferList *keyframe;
   /* If the encoding is not chunked, we'll get multiple requests
    * with separate Content-Length headers on the same path */
   SoupEncoding encoding;
@@ -59,6 +57,7 @@ struct _TranscodeClientCtx {
   GstElement *appsink; /* We hold an extra ref to this */
   GstPad *ghostsinkpad;
 
+  gboolean keyframe_found;
   guint timeout_handler_id;
   guint seconds_since_write;
   /* The transcode server context; we don't hold a ref to this */