Request for clarification: can ref initialization code (lazy or not) contain side effects?

New docs on avoiding recreating ref contents has two examples:

Example1:

const playerRef = useRef(new VideoPlayer());

Example2:

const playerRef = useRef(null);
if (playerRef.current === null) {
  playerRef.current = new VideoPlayer();
}

Even the old docs have similar examples:

Example3:

const ref = useRef(new IntersectionObserver(onIntersect));

Example4:

const ref = useRef(null);
if (ref.current === null) {
   ref.current = new IntersectionObserver(onIntersect);
}

Both new VideoPlayer() and new IntersectionObserver(onIntersect) look like calls that might contain side effects inside.
My question is it allowed in any of the above examples, for these calls to contain side effects?

• imho in Examples 1/3 they should not because they are executed on each render; although only values from initial render are kept.

• But examples 2/4 suggest a pattern which make reading and writing to refs ok - which normally is an impure operation already. So I thought maybe in those examples at least, it is ok for new VideoPlayer() and new IntersectionObserver(onIntersect) to contain side effects?

It would be helpful if the docs clarified this.

Inizia con la sezione attuale su useRef, “avoiding recreating the ref contents”, e con la sezione collegata della Legacy Hooks FAQ sulla creazione lazy di oggetti costosi. Confronta tutti e quattro gli esempi e verifica le indicazioni di React sugli effetti collaterali durante il rendering. Il lavoro è completato quando la documentazione spiega chiaramente se questi pattern di inizializzazione possono contenere effetti collaterali e distingue il loro comportamento.