Am I correct in thinking it’s possible for `by lazy(LazyThreadSafetyMode.NONE)` to NPE in the case w...

# getting-started

[cdurham]

Am I correct in thinking it’s possible for by lazy(LazyThreadSafetyMode.NONE) to NPE in the case where two threads race to initialize the value? I thought the worst case was multiple creations 😕

override val value: T
        get() {
            if (_value === UNINITIALIZED_VALUE) {
                _value = initializer!!()
                initializer = null
            }
            @Suppress("UNCHECKED_CAST")
            return _value as T
        }

[cdurham]

EG:

Both Thread A and Thread B enter the if condition before _value is set, one of the threads sets _value and then nulls the initializer before the second thread sets value, resulting in a NPE

[cdurham]

I suppose what I want is PUBLICATION?

[xoangon]

In your example, wouldn't the call to by lazy { } in the second thread just return the memoized value without calling the initializer function?

[Landry Norris]

This is the code for how a lazy getter works. I have seen NPEs when accessing lazy vars in weird ways (especially when 2 lazy vars accidentally depend on each other). It should NPE in the following example:

Thread 1             Thread 2
check initialized    check initialized
call initializer     OS schedules other work
set _value
nullify initializer
return value        call initializer (NPE)

[Landry Norris]

You'll want to set up some sort of a thread safe guard for any potential uses, or instead, use the variable once early on to make sure it gets initialized properly.

[xoangon]

That’s interesting to know. I hadn’t seen any case like that before. Not a 100% sure, but I think that using synchronized could fix it: