- Haskell-Cafe - Haskell.org

Any interested authors?
by Antonio Cangiano 04 Mar '07

04 Mar '07

Dave Thomas (from The Pragmatic Programmers, not from Wendy's :)) has announced an Erlang book (http://www.pragmaticprogrammer.com/titles/jaerlang/) in the Ruby mailing list. A few responses showed an interest in a similar book for Haskell. You can see the thread here: http://tinyurl.com/2ecpsn. Dave answered such requests by saying: "I'd love to do one. It's a question of finding the right author.". If you think this would be up your alley, perhaps you could contact Dave in regards to this matter. Just a heads up, from a potential reader. :) Regards, Antonio -- http://antoniocangiano.com Zen and the Art of Ruby Programming

1 0

Safe top-level IORefs
by Roberto Zunino 04 Mar '07

04 Mar '07

I'm posting the code of a module, IORefs, allowing top-level IORefs to be safely declared and used. Usafety reports are welcome. Tested in GHC 6.6. ** Features: 1) IORef a provided for any Typeable a 2) An unbounded number of IORef's can be declared 3) An IORef declaration is 3 lines long (+ optional type signature) 4) Each IORef has its own starting value 5) Referential transparency: no NOINLINE, no unsafePerformIO is needed in the user module 6) Negligible overhead: O(1) in the usual usage pattern (neglecting a O(log n) setup phase) ** Usage: import Data.Typeable import Data.IORef import IORefs data X deriving Typeable -- the actual name of the IORef instance IORefDefault X Int where ioRefDefault _ = 42 -- type+default x = ioRef (undefined :: X) -- a convenient name for the IORef x :: IORef Int -- optional signature -- Still the same IORef as x ! y :: IORef Int y = ioRef (undefined :: X) main = do let printX = readIORef x >>= print printX -- 42 writeIORef x 3 printX -- 3 modifyIORef x succ printX -- 4 modifyIORef y succ -- y is equal to x, so... printX -- 5 -- ... the above is actually equivalent to modifyIORef (ioRef (undefined :: X) :: IORef Int) succ printX -- 6 Passing a non _|_ value to ioRef does not break the abstraction: ioRef ignores this value. Similarly, ioRef always calls ioRefDefault with _|_, so writing (ioRef (X1 :: X)) and (ioRef (X2 :: X)) will not cause the initial value to be ill-defined, i.e. depending on which expression is evaluated first. On performance: ioRef takes O(log n) to return, where n is the number of refs previous created by ioRef. In the common usage pattern, ioRef is used only in a top-level definition. If no inlining happens, we pay only a startup cost: then all IORefs are available in O(1). If inlining happens, or if we use ioRef as in the last lines of main above, we pay the log(n) price. Note that inlining only affects the performance, and not the semantics. Assumptions/known glitches: 1) We rely on cast from Typeable 2) GHCi is known not to reinitialize the refs on reload. 3) No multithreading support for now. 4) The IORefs module uses a memoization technique, relying on a "classic" top-level IORef declared through NOINLINE + unsafePerformIO. Regards, Zun. =========================================================================== {-# OPTIONS_GHC -Wall -fglasgow-exts #-} module IORefs (ioRef, IORefDefault, ioRefDefault) where import qualified Data.Map as M import System.IO.Unsafe import Data.IORef import Data.Typeable class (Typeable a, Typeable b) => IORefDefault a b | a -> b where ioRefDefault :: a -> b data Ref = forall a . Typeable a => Ref (IORef a) type RefMap = M.Map TypeRep Ref {-# NOINLINE refs #-} -- This is crucial refs :: IORef RefMap refs = unsafePerformIO $ newIORef M.empty -- This is like a memoized function, so inlining this should be safe. -- (Needs locking for multithread, though.) ioRef :: forall a b . IORefDefault a b => a -> IORef b ioRef x = unsafePerformIO $ do rs <- readIORef refs case typeOf x `M.lookup` rs of Nothing -> do ref <- newIORef $ ioRefDefault (undefined :: a) writeIORef refs $ M.insert (typeOf x) (Ref ref) rs return ref Just (Ref aRef) -> case cast aRef of Nothing -> error $ "ioRef: impossible!" Just ref -> return ref -- Should be in Data.Typeable instance Ord TypeRep where compare x y = compare (show x) (show y)

2 2

Non-exported type classes?
by Roberto Zunino 04 Mar '07

04 Mar '07

What is the effect of declaring a class in a module and not exporting it? Would that prevent to add more instances to that class other than those already defined in the module? More in detail, consider this module: === module Peano (Z,S,C) where data Z data S a -- Not exported class Peano b where instance Peano Z where instance Peano a => Peano (S a) where -- Exported class Peano a => C a where === Would that limit the instances of the class C to the Peano type naturals? Thanks, Roberto Zunino.

3 2

is there already a function that does this?
by Jared Jennings 03 Mar '07

03 Mar '07

-- Sort the [a]'s by the [b]'s. sortByKeys :: (Ord b) => [b] -> [a] -> [a] sortByKeys keys values = map snd $ sortBy compareFst (zip keys values) where compareFst x y = compare (fst x) (fst y) -- Sort a list by an IO-returning compare function. sortByM :: (Monad m, Ord b) => (a -> m b) -> [a] -> m [a] sortByM fun lis = mapM fun lis >>= \keys -> return $ sortByKeys keys lis Is there a function that can make sortByM out of sortByKeys? I don't know of one, but having been writing haskell for a few weeks, it sort of smells to me like there probably is one.

4 3

RE: MPTCs and rigid variables
by C Rodrigues 03 Mar '07

03 Mar '07

>{-# OPTIONS_GHC -fglasgow-exts #-} > >class Foo a b | a -> b where > foo :: Foo b c => a -> Maybe c > >instance Foo String () where foo _ = Nothing >instance Foo Int String where foo 4 = Just (); foo _ = Nothing > There appears to be a type-safe way to use unsafeCoerce# for this: -------- import qualified GHC.Exts class Foo a b | a -> b where foo :: a -> FooBox b data FooBox b = forall c. Foo b c => FooBox (Maybe c) instance Foo () () instance Foo String () where foo _ = FooBox Nothing instance Foo Int String where foo 4 = FooBox $ Just (); foo _ = FooBox Nothing runFoo :: (Foo a b, Foo b c) => a -> Maybe c runFoo x = case foo x of FooBox x -> GHC.Exts.unsafeCoerce# x -------- The class constraint check of a,b,c gets moved to the runFoo method instead of the instance declarations. Between the body of foo and the call to foo in runFoo, type 'c' is unknown to the compiler, so we have to encapsulate it in a FooBox. The unsafeCoerce# is safe because the type constraint (Foo b c) that we've placed both on FooBox and on runFoo permits c to have only the type uniquely specified by b. We actually could remove the Foo constraint on FooBox, which makes the code no longer typesafe but it will still work just fine. _________________________________________________________________ Dont miss your chance to WIN 10 hours of private jet travel from Microsoft® Office Live http://clk.atdmt.com/MRT/go/mcrssaub0540002499mrt/direct/01/

2 1

Proposal: simple interface to libraries Haddock
by David House 03 Mar '07

03 Mar '07

Hey all, I'm proposing something simple that should make it easier for both Haskell programmers and Haskell tool writers to find the documentation for a function. We need a simple URL for the Haddock docs of any function. PHP has this already: just hit php.net/funcname for the documentation for funcname(). My proposal is similarly simple: http://docs.haskell.org/funcname A CGI script would find funcname in the index and redirect to the URL of the docs for funcname. The only contentious issue is what to do in case of ambiguous lookups; i.e. would docs.haskell.org/lookup be Data.Map.lookup, or Data.List.lookup? My suggestion is that we present a page saying 'which function did you mean?'. We should also allow a way to be unambiguous in the URL: http://docs.haskell.org/Data.Map.funcname Pros: 1) It's really easy to find the documentation for any function. 2) It's really easy to write a tool that finds documentation for a function (including Firefox keywords, Emacs/Vim integration, GHCi integration and so on). Any comments? -- -David House, dmhouse(a)gmail.com

5 10

instance Eq Exception??
by Isaac Dupree 03 Mar '07

03 Mar '07

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 It seems the Exception datatype is currently an instance of Eq, that does not even satisfy reflexivity (for Dynamic exceptions). Does anybody use this? What is it good for? Why is it there? What if we delete the instance because it doesn't make sense? Exception is not an instance of Ord currently; it also has an instance Show that does make a little sense - for Dynamic exceptions, it gives (in GHC, not Hugs) the information of what the exception's type looks like. Isaac -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.3 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFF6bSNHgcxvIWYTTURArNRAJwMmp/keolibaZggANLCe5IMdit7ACcDZkp 1WP7Fkdj36ZEeHxHFem1py8= =gkfv -----END PGP SIGNATURE-----

1 0

Re: [Haskell-cafe] MPTCs and rigid variables
by David House 03 Mar '07

03 Mar '07

On 03/03/07, Bryan Burgers <bryan.burgers(a)gmail.com> wrote: > As far as I know, you should be defining the type of bar in the class > declaration for Foo, not in an instance declaration. I think I over-reduced my problem. Here's a more complex issue that can't be resolved by moving bar to be a method of the class: {-# OPTIONS_GHC -fglasgow-exts #-} class Foo a b | a -> b where foo :: Foo b c => a -> Maybe c instance Foo String () where foo _ = Nothing instance Foo Int String where foo 4 = Just (); foo _ = Nothing -- -David House, dmhouse(a)gmail.com

1 0

An HSH case study
by John Goerzen 03 Mar '07

03 Mar '07

Some of you may be interested in my recent blog post[1], where I discuss porting hg-buildpackage[2] from a set of shell scripts to a set of Haskell programs that use HSH. To my surprise, I achieved a 20% *reduction* in source lines of code, while doing the exact same tasks. I guess I have to consider switching more code to HSH. Purely in the interests of science, of course ;-) [1] http://changelog.complete.org/posts/587-A-better-environment-for-shell-scri… [2] http://hg.complete.org/hg-buildpackage

1 0

ANN: HSH 1.0.0
by John Goerzen 03 Mar '07

03 Mar '07

Hello, I'm pleased to announce the first release of HSH. HSH is designed to let you mix and match shell expressions with Haskell programs. With HSH, it is possible to easily run shell commands, capture their output or provide their input, and pipe them to/from other shell commands and arbitrary Haskell functions at will. HSH makes it easy to run shell commands. But its real power is in piping. You can pipe -- arbitrarily -- between external programs, pure Haskell functions, and Haskell IO functions, mixing and matching at will. Here are a few examples to get you started: runS $ "echo /etc/pass*" -> "/etc/passwd /etc/passwd-" run $ "ls -l" -|- "wc -l" -> 12 run $ "ls -l" -|- wcL -> 12 wcL is a pure Haskell function: wcL :: [String] -> [String] wcL inp = [show $ genericLength inp] More information, including downloads and further examples, is at the HSH homepage: http://software.complete.org/hsh -- John

2 2