Actually the snapshot format is ZLib, not raw streams so that makes sense.
Okay so profiling works better, although there is lost information! I need to set a better tree root.
The self time and such for nodes is definitely not correct.
Okay so, for some reason the total thread time is way off which is why the percentages are probably messed up.
Seems like if the first instruction is the target of a jump that it will have a stack map entry defined for it.