Jukebox TileState interface is poorly defined and/or broken

Expected behavior

Jukebox TileState methods are poorly defined or have broken behavior after 1.19.4 update. It is not clear what setPlaying, setRecord, isPlaying, stopPlaying should do to the block entity. It is more complicated because paper adds non-snapshot TileStates.

Observed/Actual behavior

setPlaying(Material) and setRecord do not change the IsPlaying nbt or start the tick counter or update neighbors either for snapshots or live views.

isPlaying incorrectly only checks if the BlockState has HAS_RECORD set to true. Which is not accurate as Jukebox can have a record but not be playing.

stopPlaying only plays the stop music event for clients to stop playing the music. It does not change the IsPlaying nbt value or the various tick counters. This breaks hoppers being able to extract once it's stopped. But the javadocs say "without ejecting the record". So what does that mean? Is being collected by a hopper ejected?

Steps/models to reproduce

Just use any of the methods mentioned with a block state snapshot or live view or look at the CraftJukebox implementation.

Plugin and Datapack List

N/A

Paper version

1.19.4

Other

The one issue not mentioned here upstream has already fixed (but not in the right way IMO).

This is very complicated because these methods both affect the snapshot, and the real experience of the clients before update() is called. Should startPlaying immediately start playing the disc or should it wait for update(). stopPlaying immediately stops the music regardless. It's just a big mess of confusing behavior.

Start with the CraftJukebox implementation and inspect setPlaying, setRecord, isPlaying, and stopPlaying for both block-state snapshots and live views. First establish the intended semantics for IsPlaying, tick counters, neighbor updates, hopper extraction, and client music before deciding how update() should affect playback. Done means the methods have consistent, documented behavior across both views.