Forgot to do simple math stuff for the various CMP types.
Okay the stack state stuff is far too convoluted, like it takes up so much space and is hard to read.
Okay so, I see that no counting is set for integers..... this should never be set for non-objects because it does not make sense.
And doing that actually removed a transition which was flagged, so that is rather nice.
Okay so, stack poisons are only really needed for instructions which are jumped to from a future instruction. For jump forwards, there just needs transitions to be done which can be derived from the stack state.
Actually, transition to the next instruction should be jumped to by the Java label stuff. This is just in case that the target state does not match and it needs to be adjusted. This of course would mean that the entry poison is not needed. So would say if the directly following instruction (if we flow naturally) has a different state and it has not explicitely been poisoned then we do a transition to it.
I should sleep. But I have my first instance of going to another state for transitioning. Right now it is just a simple uncount and a forget about a bunch of other locals.