Hi guys, has something changed in 1.6.21 regarding nullables? code posted in thread.

# getting-started

[wck]

Hi guys, has something changed in 1.6.21 regarding nullables? code posted in thread.

[wck]

class Test {
    fun <T: Any?> nullableFun(): T {
        return null as T
    }
    @Test
    fun testNull() {
        assertTrue {
            val a = nullableFun<String>()
            a?.length == null
        }
    }
    @Test
    fun testNull2() {
        assertTrue {
            nullableFun<String>()?.length == null
        }
    }

}

[wck]

both tests passed in 1.6.10, however testNull failed with NPE in 1.6.21, testNull2 still passed

[ephemient]

it could be a side effect of the definitely not-null feature, however your code is bad in all cases, it should either be

fun <T: Any> nullableFun(): T?
fun <T> nullableFun(): T?

because what you currently have implies that nullableFun<String>(): String which is wrong (but can't be checked in the function itself due to generics)

[wck]

The code is made intentionally to reproduce the NPE

[wck]

I'm just curious on whats the difference between those, it crashes if it's assigned to a val

[wck]

Shouldn't the behavior of those be the same

[ephemient]

in the first case, it's equivalent to

val a: String =

for which the compiler will insert a check in various cases

[wck]

thanks a lot, that makes sense. So what about the second case? couldn’t find too much information on this one, reading the Definitely non-nullable types doesn’t seem to be quite related

[ephemient]

I agree the compiler isn't being consistent and that definitely non-nullable types doesn't directly impact this case, but it's the only type-related change I spotted in the changelog so it's possible this was a side effect

[wck]

@ephemient https://youtrack.jetbrains.com/issue/KT-52743
This is now confirmed as a major bug if you are still interested