September 2007 - Haskell-Cafe

Building "production stable" software in Haskell
by Peter Verswyvelen 18 Sep '07

18 Sep '07

The way I see it as a newcomer, Haskell shifts the typical imperical programming bugs like null pointers and buffer overruns towards "space/time leaks", causing programs that either take exponentially long to complete, stack overflow, or fill up the swap file on disc because they consume gigabytes of memory. Other bugs I found are incomplete pattern matches at runtime, but I already got an email of how to fix this using an external tool, although it would be nice if this is part of the compiler/linker itself. How well and how can a Haskell program be tested to make sure it does not cause these space/time bugs? What tools are typically used? Thanks, Peter

22 49

yi-gtk
by clisper 18 Sep '07

18 Sep '07

who knows how to compile yi-gtk? i tried,but it told me mine miss gtk. 2007-09-18 clisper

2 1

Type-Marking finite/infinte lists?
by Joachim Breitner 18 Sep '07

18 Sep '07

Hi, today while mowing the lawn, I thought how to statically prevent some problems with infinte lists. I was wondering if it is possible to somehow mark a list as one of finite/infinite/unknown and to mark list-processing functions as whether they can handle infinte lists. For example, it is known that a list produced by repeat or cycle is always infinte, that foldl can’t handle infinte lists (while foldr might). length should never be called on an infinte list, and map’s output is infinite iff the input list is. So the compiler might use this information to emit errors like “line 42: program will not terminate: length applied to infinite list” or warnings “line 23: termination not guaranteed: foldl applied to possible infinte list”. I guess you can’t know it for all lists (halting problem...), but for those where it is possible, it would be nice to tell. Or am I missing something? Can this be implemented using some guru type system hacking? Greetings, Joachim -- Joachim "nomeata" Breitner mail: mail(a)joachim-breitner.de | ICQ# 74513189 | GPG-Key: 4743206C JID: joachimbreitner(a)amessage.de | http://www.joachim-breitner.de/ Debian Developer: nomeata(a)debian.org

7 9

Re: [Haskell] Blocked STM & GC question
by Jules Bean 18 Sep '07

18 Sep '07

Simon Marlow wrote: > Ashley Yakeley wrote: >> If I have a thread that's blocked on an STM retry or TChan read, and >> none of its TVars are referenced elsewhere, will it get stopped and >> garbage-collected? >> >> I have in mind a pump thread that eternally reads off a TChan and >> pushes the result to some function. If the TChan is forgotten >> elsewhere, will the permanently blocked thread still sit around using >> up some small amount of memory, or will it be reaped by the garbage >> collector? > > In this case, your thread should receive the BlockedIndefinitely exception: > > http://www.haskell.org/ghc/docs/latest/html/libraries/base/Control-Exceptio… > > > If the system is idle for a certain amount of time (default 0.3s, change > it with the +RTS -I option) a full GC is triggered, which will detect > any threads that are blocked on unreachable objects, and arrange to send > them the BlockedIndefinitely exception. Including MVars? Your quoted text suggests 'Yes' but the docs you link to suggest 'No'. Jules

2 1

unique id for data types
by Barney Hilken 18 Sep '07

18 Sep '07

In order to make my records system practically useable, I need a type family > type family NameCmp n m which totally orders datatypes. More precisely, it should return one of the following types: > data NameLT = NameLT > data NameEQ = NameEQ > data NameGT = NameGT for each pair of datatypes n & m, according to whether n < m, n = m, or n > m in some global ordering. This ordering needs to be independent of the context, so it can't be affected by whatever imports there are in the current module. What I want to know is: does GHC give datatypes any global id I could use to generate such an ordering? Would fully qualified names work? Secondly (assuming it's possible) how easy would it be for me to write a patch to add NameCmp to GHC? Where in the source should I start looking? Thanks, Barney.

2 2

Observations from ListLike
by John Goerzen 18 Sep '07

18 Sep '07

Hi folks, I just announced ListLike[1] on haskell(a)haskell.org. Rather than repeat that announcement here, I'd like to make a few observations that came out of the development of this program: * I wrote extensive QuickCheck cases for this and wrapped them in HUnit for better display and future extensibility. Two bugs in the GHC base libraries were discovered as a result (one has been fixed already in HEAD, one not) * Hugs programs that use cpphs can't use ByteString * It would be really nice if QuickCheck supported I/O and some version of HUnit's TestLabel to generate hierarchical names when failures occur. * It would also be much nicer if QuickCheck defined Arbitrary instances for Word8 and Char. On Hugs, QuickCheck also lacks an instance of Maybe, but GHC has it for some reason. * Testing functions like foldl in Hugs, where QuickCheck is supplying the list and the function, cause Hugs to crash with stack errors. Testing some functions over ByteStrings cause Hugs to crash complaining of being out of ForeignPtrs. I have yet to figure out how to fix either problem; raising the heap size doesn't help. * I had some cases in String.hs[2] where Hugs did not permit me to define the default functions within the class declaration. However, I could define the functions outside, with the exact same type signature as they'd have inside, and within the class declaration, just say something like "words = myWords" to refer to the external declaration. GHC worked fine both ways. It seems that Hugs has a number of issues with code like ListLike that uses some more advanced Haskell features, even though Hugs nominally supports them. If you look through my Mercurial changelogs, you'll see "fix blah for hugs" messages all over. Which is too bad, because Hugs seems more portable (especially to embedded devices) than GHC. [1] http://software.complete.org/listlike [2] http://software.complete.org/listlike/browser/src/Data/ListLike/String.hs

3 2

Israel Haskell Programmers
by B K 18 Sep '07

18 Sep '07

Hello, Are there any Haskell Hackers on this mailing list who live in Israel? I am interested in starting an Israel Haskell User Group.

3 2

Dynamic thread management?
by Hugh Perkins 18 Sep '07

18 Sep '07

Haskell/FP seems to have solved the hardest bit of threading, which is making it obvious which bits of a program are parallelizable, and which are not. Remains to actually parallelize out the programs. Am I being naive or is this trivial? There has been a lot of talk about parallelizing out a program statically, at compile time, but this would seem to suffer from the halting problem? The only way to know how long a function, or one of it's sub-functions, will take to run is to actually run it? Is there some reason why we cant just start a function running in a single thread, whilst running profiling, then after a while we check which bits of the function are taking time to run, and are parellizable, and we parallelize those out? This sounds reasonably trivial and workable? Basically, "parallizable" in a Haskell context means, as a first approximation, any map, foldr, foldl or derivatives, and any independent let assignments, then we can always add in extra parallizable cases later. Profiling is already built into Haskell AFAIK? so the profiling information is already available? Thoughts? Is there some reason why such an approach has been disregarded/is harder than it sounds? (By the way, can someone provide me a non-google link to the SPJ video talking about nested data parallism? google video is unavailable from my current location, and although I did watch this video once, it was on my second day of doing Haskell, a few weeks ago, so much of it was incomprehensible to me ;-) )

15 29

Re: [Haskell] Swapping parameters and type classes
by Bas van Dijk 17 Sep '07

17 Sep '07

On 9/16/07, Mads Lindstrøm <mads_lindstroem(a)yahoo.dk> wrote: > Hi all > > If I have this type: > > data Foo a b = ... > > and this class > > class Bar (x :: * -> *) where ... > > I can imagine two ways to make Foo an instance of Bar. Either I must > "apply" the 'a' or the 'b' in (Foo a b). Otherwise it will not have the > right kind. To "apply" the 'a' I can do: > > instance Bar (Foo a) where ... > > But what if I want to "apply" the 'b' ? How do I do that ? > > > Greetings, > > Mads Lindstrøm > > > _______________________________________________ > Haskell mailing list > Haskell(a)haskell.org > http://www.haskell.org/mailman/listinfo/haskell > (Redirecting to Haskell-Cafe) I think the easiest way to make it work is with a newtype wrapper around 'Foo a b' that swaps the arguments, like Brent mentioned. Without it you need some kind of type level lambda abstraction. Something like: instance Bar (\x -> Foo x b) where ... But this is not valid Haskell. cheers, Bas

1 0

Records: Examples
by Barney Hilken 17 Sep '07

17 Sep '07

> {-# LANGUAGE TypeFamilies #-} Hi Justin, thanks for your interest. Hope this helps! > module Examples where > import Records To get started, you need to define your labels. They are just singleton datatypes: > data FirstName = FirstName deriving (Show, Eq, Ord) > data Surname = Surname deriving (Show, Eq, Ord) > data Address = Address deriving (Show, Eq, Ord) > data PhoneNo = PhoneNo deriving (Show, Eq, Ord) you can define as many as you like. Next you have to define the order on fields. At the moment you have to do this by hand, but I hope to get ghc to do this automatically: > type instance NameCmp FirstName FirstName = NameEQ > type instance NameCmp FirstName Surname = NameLT > type instance NameCmp FirstName Address = NameLT > type instance NameCmp FirstName PhoneNo = NameLT > type instance NameCmp Surname FirstName = NameGT > type instance NameCmp Surname Surname = NameEQ > type instance NameCmp Surname Address = NameLT > type instance NameCmp Surname PhoneNo = NameLT > type instance NameCmp Address FirstName = NameGT > type instance NameCmp Address Surname = NameGT > type instance NameCmp Address Address = NameEQ > type instance NameCmp Address PhoneNo = NameLT > type instance NameCmp PhoneNo FirstName = NameGT > type instance NameCmp PhoneNo Surname = NameGT > type instance NameCmp PhoneNo Address = NameGT > type instance NameCmp PhoneNo PhoneNo = NameEQ Now we are ready to play! To define records, use (=:) and (+:) > barney = FirstName =: "Barney" +: Surname =: "Hilken" +: > Address =: "Horwich" +: PhoneNo =: "697223" You can use as many or as few of the fields as you like, and you can write them in any order, but trying to use a field twice in the same record will give you a (rather incomprehensible) type error. > justin = Surname =: "Bailey" +: FirstName =: "Justin" +: Address =: "Somewhere" To extract the value at a field use (.:) > myPhone = barney.:PhoneNo To delete part of a record, use (-:) > noCallers = barney -: Address To update existing fields in a record, use (|:) > barney' = barney |: Address =: ((barney .: Address) ++ ", UK") The power of the records system is that these five operators, =: +: .: -: |: are Haskell polymorphic functions. So you can define functions like > livesWith p q = p |: Address =: (q .: Address) which returns p, but with its Address field changed to that of q. Note that this function works on any records p and q with Address fields, whatever other fields they may have. You can even define functions parametrised by field names: > labelZip n m = zipWith (\x y -> n =: x +: m =: y) then 'labelZip FirstName Surname' is a function which takes two lists and returns a list of records: > names = labelZip FirstName Surname ["Barney", "Justin"] ["Hilken", "Bailey"] of course, labelZip isn't restricted to the four labels we defined earlier, it works on anything. The system is strongly typed, so record errors (such as missing or duplicated fields) are caught at compile time. There are type operators (:=:), (:+:), (:-:), (:.:) corresponding to the record operators, and classes `Contains`, `Disjoint`, `Subrecord` which allow you to express conditions on types. Unfortunately, the type system sometimes decides that a function has a different type from the one you expect, and won't accept the header you want to give it. More experience with the system is needed before we can say whether this is a problem. Barney.

1 0