Have FFmpeg Decoders handle 0 byte buffers instead of FFmpeg

media/filters/ffmpeg_video_decoder.cc

Index: media/filters/ffmpeg_video_decoder.cc
diff --git a/media/filters/ffmpeg_video_decoder.cc b/media/filters/ffmpeg_video_decoder.cc
index 094720566e1eaa7cb03ff02a8784fec0c325231d..cf7795a337f116d8257c4aa99b32d8a10e84e843 100644
--- a/media/filters/ffmpeg_video_decoder.cc
+++ b/media/filters/ffmpeg_video_decoder.cc
@@ -349,6 +349,13 @@ bool FFmpegVideoDecoder::FFmpegDecode(
     packet.data = const_cast<uint8_t*>(buffer->data());
     packet.size = buffer->data_size();
 
+    // Starting in M60, avcodec_decode_video2() generates an error if an
+    // empty buffer is provided. Since we're not at EOS and there is no data
+    // available in the current buffer, simply return and let the caller
+    // provide more data. crbug.com/663438 has more context on 0-byte buffers.
+    if (!packet.size)
+      return true;
+
     // Let FFmpeg handle presentation timestamp reordering.
     codec_context_->reordered_opaque = buffer->timestamp().InMicroseconds();
   }