Some thoughts on how to organize Demand CFA for Koka
Static analysis in Koka. I'm looking through the Koka codebase to figure out the right strategy for including demand control flow analysis into the code. There are a number of CorePhase passes in src/Compile/Compile.hs. It seems like there could be a number of places where additional static information on demand could be useful (such as the various inlining or specialization passes). As such I might add the machinery for querying control flow information into the CorePhase Monad's State which is called CPState. It seems as if the CorePhase Monad never includes state currently, instead just transferring defs to other defs. It seems as if I just add some state available in all of those passes, it should allow for many different queries to be useful. Some queries might take some time, so I need the ability to timeout, or an effort parameter. Any particular information however that I gain from the analysis, should be cached. Thinking about it some more, though, flowsets and abstractions might need to be different for integers versus closures, or different queries. Additionally some inlining / specialization might cause different flow, so it might be best to just make the machinery available in the CorePhase Monad, but don't do any caching. I'll start with that approach.