In playsink reconfigure function there is some code to reenable the video chain if it already exists. In the case of a pipeline with audio, video and subtitles the ghostpad for video is the one from the text chain, for audio video only the ghost pad should be the one from the video chain.

During reconfigure playsink tries to set the ghost correctly depending on the new conditions. I don't understand that block of code though :

      if (!need_vis && !need_text && !playsink->textchain) {
        GST_DEBUG_OBJECT (playsink, "ghosting video sinkpad");
        gst_ghost_pad_set_target (GST_GHOST_PAD_CAST (playsink->video_pad),
            playsink->videochain->sinkpad);
      }

Indeed it does not make sense (IMHO) to check if there's a textchain because if there was one it might be cleaned later on. Can't we just check for need_vis and need_text ?

The consequence of this, as observed on my player, is that playing a A/V file after a A/V/T file is just not working because the ghostpad points to a non existing chain.

The ghostpad could be reset correctly when removing the text chain but just changing the condition to this, makes it work for me :

      if (!need_vis && !need_text) {
        GST_DEBUG_OBJECT (playsink, "ghosting video sinkpad");
        gst_ghost_pad_set_target (GST_GHOST_PAD_CAST (playsink->video_pad),
            playsink->videochain->sinkpad);
      }