OK, so we still need to recreate the pipeline. Do we recreate the whole wg_parser or only recreate decodebin pipeline?
Without having tried it, the former sounds easier; it requires less infrastructure, and it's what we already do when changing the format in transforms.
Another question: if we assume that the caps configure only happens during setting up the reader, before decoding anything, and ignore the mid-stream situation for now. Do we still need to do the recreate things when caps configure happens?
We currently create the parser when the stream is opened, and we can't create it any later (since we need to query things like stream count and stream formats). On the other hand, toggling decoding is done on the stream interfaces, which has to be done after the stream is opened. So we definitely need to recreate the parser, one way or another.