Is there any interest in implementing a top level "<-" to run monadic code? Currently this sort of thing is done with unsafePerformIO and switching off inlining with some pragma. Indeed, the 'atomically' haddock actually advises doing this to declare top-level TVars. The same trick is used in the source of Data.Unique and System.Random. This is bad Haskell. To avoid observation of effects, etc., we would need a new monad rather than IO. There would be equivalents of IO functions such as these: newIORef newMVar newTVarIO newUnique I can think of two uses: 1. Global mutable state. For instance, here's the count variable for Data.Unique rewritten: uniqSource :: MVar Integer uniqSource <- newMVarTL 0 Isn't that much nicer? <http://haskell.org/haskellwiki/Top_level_mutable_state> 2. Solving the expression problem using open witnesses, a recent hobby-horse of mine. For instance, here's a simple scheme for extensible exceptions using top-level "<-": -- declare exception carrying an Int myException :: Exn Int <- newExn -- throw with 5 foo = do ... throw myException 5 -- catch, print the Int bar = catch foo myException (\i -> putStrLn (show i)) I already have code for this, except with an unsafe hack to do the "<-" declaration: <http://hackage.haskell.org/packages/archive/open-witness/0.1.1/doc/html/Data-OpenWitness-Exception.html> My open-witness package, which also shows how to do Typeable/Dynamic safely: <http://hackage.haskell.org/cgi-bin/hackage-scripts/package/open-witness> -- Ashley Yakeley Seattle, WA