Currently we always throw away JIT code when we perform a compacting GC due to the fact that both are triggered for shrinking GCs. This is also the behaviour we want in practice since the purpose of shrinking GCs is to reduce memory use. However it would be nice to be able to test these separately. This would give us confidence that JIT code is correctly traced. This depends finding a way of dealing with the fact that we sometimes bake elements pointers into code (bug 979449). There are probably other issues too.