@AndrewCMyers Is the careful thinking devising an abstraction that the memory actually implements and developing a threat model based on it?
If so, one reason there may be another attack lurking: the devised abstraction will certainly be oblivious to some classes of behavior.
@kmett@MarisaVeryMoe @codydroux @ezyang@mattmight@andrasKovacs6 Abstract GC can be used in definitional interpreters too and big step enables a much more effective mode of it. (I have a recent ESOP paper about it.) Higher precision and (not published) incredible performance gains.
@GuannanWei@ilyasergey@samth@krismicinski@mattmight@lambda_calculus @JanMidtgaard I have a draft paper connecting AAM, ADI, and PoPA constraint-based via a unifying monad exactly as Guannan suggests: monadic computations encode constraints, running it intertwines discovery and resolution. I’m happy to share to any interested.
@glaebhoerl@GuannanWei This is how CFA2 does it. I have a blog post draft that does what I describe that I’d be happy to share. Now I’m wondering if you can analyze CPS code using ADI by inserting a continuation monad in the monad stack.
@glaebhoerl If your CPS transformation distinguishes the continuation terms it introduced, you can use the @GuannanWei’s technique (related to ADI) to get precise call-return matching for the calls that existed pre-CPS. You can often recover the distinction directly from the IR too.