On Tue Feb 27 17:33:43 2024 +0000, Rémi Bernon wrote:
The output type might still contain different attributes, such as different subtype. It's not completely clear how native handles this and there's plenty of possibilities. The main reason I'm doing this is because the frame size isn't required, and might be provided by the decoder instead. I could do that completion for hand-picked attributes if that sounds better. This is probably not completely correct, as it propagates every attribute downstream when I believe some get swallowed, but it was also easier. Note that there's the same kind of thing being done in the topology loader now, for mostly the same reasons.
I updated the code to only propagate hand-picked attributes instead of everything, also made a similar change on the mf/topology_loader side. It actually fixed a couple of things where some pipelines failed to resolve (when decoder transforms are used, which is not yet the case in upstream Wine), as MF_MT_COMPRESSED_TYPE was incorrectly propagated.