jdk/jfr/threading/TestDeepVirtualStackTrace seems to have some assumptions about the frame state used when an allocation is performed that are not true for Graal and this leads to periodic failures where it see a shallower stack trace then it expects. If escape analysis in general permits an allocation to be performed at a location which is different than the bytecodes then in general it's permitted to perform allocation where ever it's optimal to do so. If the test were modified to introduce a side effect of some kind before the allocation then it would be performed in the right inline scope. I'm not sure what other way there is to make this pass with Graal.