Well the other issue is of course that Android being available on a wide variety of phones, not all of which run ARM (the phone I am about to get for example has a custom built CPU), although I guess one could use a "generic" ASM branch for "mobile" devices (if one exists). btw the phone I am about to receive is a Samsung Galaxy-S, which has a hummingbird chip (no idea what Assembler Instruction set it uses, I believe its a closed chip) In my opinion the most "reliable" approach would actually to produce the C that wraps around NDK (for any code that could be possible) which would obviously interface with the Java libraries. Probably the biggest bane of Android is the fact that its able to run on almost all machines means that there *would* be issues using LLVM to just produce code for the generic ARM. Using the NDK + Java would mean any app written in Haskell would work on any android device. Of course as mentioned above, there are issues with this. Its a lot of work for one thing, and the GC for Java is probably not the best suited for Haskell structured programs. Also iirc, in order to make "official" Android apps which can go on the market place, you are basically forced to use the Java API + NDK for any native code, and you have to use the Java API to interface with the android GUI/World (you can't make an proper Android app just using NDK which complicates things further). The GC issue I guess could be solved by generating JVM code (instead of Java "code") which would give more freedom for GHC to generate code more suited to Haskell If this ever ended up happening, it does have the advantage that when one would develop an app for Android using Haskell, GHC (or whatever compiler we would use) can use NDK for as much code as possible and only resorting to Java libraries when required which can of course equate to creating very fast Android Apps using a productive language (Haskell) As Don mentioned through, we need a java runtime for GHC.