[kdrakon]

I only have an amateur interest in JVM development, but doesn't this bug break some rules around GC 'object roots'? Even if 'child' and 'something' are referentially copied to local variables on the calling stack, I believe 'this' should continue to act as a root for both of them. I think.

[dietrichepp]

"this" can be collected in the body of a method, if the last reference dies before the method returns. It is treated no differently from the explicit method parameters.

[adevine]

Not necessarily. Those "getChild()" and "something()" methods could, for example, return new objects. this is only the root of child and something if this retains references (i.e. member variables) to these objects, which I'm guessing was not the case in this example.

[kdrakon]

Ah, yes, didn't consider that. But what about 'this'?; for what reason could the JVM choose to GC an object that the calling thread is currently running in? The example method exists in an instance of 'this' and isn't static.

[Dylan16807]

Execution in Java does not have a concept of "current object". 'this' is passed in a magic way, but it does not exist in a magic way. It's like any other variable, and can be collected as soon as it's not needed. Which can be extremely early, because optimizations can reduce the 'need' for a variable to only its side effects.

Go pick apart aggressively optimized and inlined code and I bet you can find situations where 'this' never even exists.