Effect owners: EffectOwners

Who applied an effect to whom.

EffectOwners.set(target, AwakenEffects.ENTHRALLED.get(), charmer);

if (EffectOwners.isAppliedBy(target, AwakenEffects.ENTHRALLED.get(), attacker)) {
    // the attacker is the one who charmed them
}

Why it exists

A MobEffectInstance carries a duration and an amplifier and nothing else — no owner. An effect whose whole point is the relationship needs the owner kept beside it:

• this target may not hurt the one who charmed them;
• this target's invulnerability frames do not hold off the surgeon who put them under, nor their side.

Member	Answers
set(target, effect, owner)	records it
getId(target, effect)	the owner's UUID, or null
get(target, effect)	the owner resolved to a LivingEntity, in the target's level
isAppliedBy(target, effect, candidate)	the question actually asked
clear(target, effect)	forgets it

Keyed on the target's UUID, not on the target

The hand-rolled version of this was a WeakHashMap<LivingEntity, UUID>, and that loses its entry whenever the entity object is replaced while the effect survives on the reloaded one: a player changing dimension or respawning, a mob whose chunk unloads and comes back. The effect then keeps running with its owner gone and nothing reports it — a charm that quietly stops protecting the charmer. Both places in AwakenAwakenNoMi that did this were reachable inside the effect's own duration.

It does not survive a server restart, on purpose

Persisting to the target's NBT was the shape originally proposed. Measured, the two effects that need an owner run for 6 and 20 seconds, which no restart fits inside — writing and evicting NBT for that would be cost with no case behind it. An effect that genuinely outlives a restart should carry its owner in its own persisted state rather than here.

Cleanup

A read that finds the effect gone drops its own entry, so the normal path cleans up after itself. A ten-minute age cap, swept on the next set, is only a safety valve for an entry nothing ever reads again; the whole map is dropped on ServerStoppedEvent.

What does not belong here

Per-carrier scratch state — the last tick an effect did something, the position it started from, what it has already hit. That is per-session and per-object by nature, and losing it when an entity is replaced is usually the right answer rather than a bug. Keep it in a WeakHashMap beside the effect and say so.