See attached test case. When running under ion/rabaldr, sometimes shows a runtime error in function 4, sometimes doesn't. I suspect we might be reading a garbage value at some point and make a decision upon it.

(same thing with function 11)

Hah, this is an invalid test case: - there's a function A that set_global 0, then calls function B - function B also set_global 0, then calls function A, leading to a runaway recursion situation and an InternalError. - when reaching the InternalError, the last set_global 0 that's been executed can be the one from A or from B. - func_4 is called after function A, and it does a get_global 0. - since the read value can be different (according to which function run last in the recursion), func_4 can have different behavior. So that's probably a bug in the fuzzer: it should ensure a deterministic global state before calling each function. I'll reuse the initial state before each function call.