Hector Martin

it doesnt seem that difficult in concept, just a little complex to work out all the bugs, but they're apple, they can do a lot of testing

It's... Not that complicated given the other considerations you have to deal with anyway, like x86 binaries not having unambiguous parses

t-they don't???

so if you want some real fun... call rand test eax, 1 jz 1f+1 1: add ax, ax If rand returns an odd value, that executes "add ax, ax". If it returns an even value, it skips the first byte of the instruction, which will be 66h (operand size prefix), so it becomes "add eax, eax"

That's ... horrid. Or clever. Or something. Self-modifying code that doesn't actually modify anything.

Not self modifying, just jumping in the middle of an instruction

Utterly ancient anti-reversing technique too. You don't trust x86 disassembly unless you have evidence of it starting on an instruction boundary that actually gets jumped to.

this should be significantly easier to handle than a JIT by using the "superset disassembly" brute-force trick? JITing a JIT seems like it'd much trickier because it tends to touch the "cache invalidation" hard CS problem