I'm using "hint", but since it's basically a thin wrapper around the GHC API, this is probably a GHC api question too. Maybe this should go to cvs-ghc? Let me know and I'll go subscribe over there. It's my impression from the documentation that I should be able to load a module interpreted, make changes to it, and then reload it. This is, after all what ghci does. It's also my impression that the other imported modules should be loaded as object files, if the .hi and .o exist, since this is also what ghci does. However, if I load a module and run code like so (using hint): GHC.loadModules ["Cmd.LanguageEnviron"] GHC.setTopLevelModules ["Cmd.LanguageEnviron"] GHC.setImports ["Prelude"] cmd_func <- GHC.interpret (mangle_code text) (GHC.as :: LangType) It works fine until I change LanguageEnviron. If I make a change to a function, I don't see my changes in the output, as if the session is only getting partially reset. If I insert a syntax error, then I do see it, so it is recompiling the file in some way. However, if I *rename* the function and call it with the new name, I get a GhcException: During interactive linking, GHCi couldn't find the following symbol: ... etc. So I examined the code in hint for loadModules and the code in ghci/InteractiveUI.hs:/loadModule, and they do look like they're doing basically the same things, except a call to rts_revertCAFs, which I called too just for good measure but it didn't help (I can't find its source anywhere, but the ghci docs imply it's optional, so I suspect it's a red herring). Here's a condensed summary of what hint is doing: -- reset GHC.setContext session [] [] GHC.setTargets session [] GHC.load session GHC.LoadAllTargets -- rts_revertCAFs -- load targets <- mapM (\f -> GHC.guessTarget f Nothing) fs GHC.setTargets session targets GHC.load session GHC.LoadAllTargets -- interpret let expr_typesig = "($expr) :: xyz" expr_val <- GHC.compileExpr session expr_typesig return (GHC.Exts.unsafeCorce# expr_val :: a) -- GHC.compileExpr maybe_stuff <- hscStmt hsc_env ("let __cmCompileExpr = "++expr) ([n],[hv]) <- (unsafeCoerce# hval) :: IO [HValue] return (Just hv) and then ghci does: -- load GHC.setTargets session [] GHC.load session LoadAllTargets targets <- io (mapM (uncurry GHC.guessTarget) files') GHC.setTargets session targets GHC.load session LoadAllTargets rts_revertCAFs putStrLn "Ok, modules loaded: $modules" -- interpret GHC.runStmt session stmt step -- GHC.runStmt Just (ids, hval) <- hscStmt hsc_env' expr coerce hval to (IO [HValue]) and run it carefully So it *looks* like I'm doing basically the same thing as ghci... except obviously I'm not because ghci reloads modules without any trouble. Before I go start trying to make hint even more identical to ghci, is there anything obviously wrong here that I'm doing?