- jhc - Haskell.org

Re: [jhc] Monad.ST
by Henning Thielemann 15 Nov '09

15 Nov '09

On Sun, 15 Nov 2009, Isaac Dupree wrote: > Henning Thielemann wrote: >> >> It compiles, but I have not tested any program that uses it. Even with >> simplest examples using storablevector it depends largely on fortune whether >> they run or crash. So no chance to make reliable statements about something >> more complex. > > hmm, I looked at the code. I don't quite understand lazy ST, but I don't > think that version you made is correct. The writes (and reads) really may > not be re-ordered based on which data is demanded when. You are right. I only thought about newSTRef's for which the order is not important. Yes, we also have to assert that writes are in order and reads remain between adjacent writes. Once I solved a similar problem in the 'lazyio' package. I defined IO actions in a way, that triggering one action also requires to execute all previous actions. This should also work for ST but it is somehow overly strict. E.g. do a <- newSTRef 'a' b <- newSTRef 'b' writeSTRef a 'c' writeSTRef b 'd' can be re-ordered to do b <- newSTRef 'b' writeSTRef b 'd' a <- newSTRef 'a' writeSTRef a 'c' That is, maximum laziness would be achieved, if there would be a state-thread for every reference.

1 0

strange crashes and changes in performance
by Henning Thielemann 15 Nov '09

15 Nov '09

I'm playing around with a simple program that fills a buffer and writes that to disk. The first variant is strange since it becomes much more than twice as slow, when the test runs twice: http://code.haskell.org/storablevector/speedtest/SpeedTestExpSlow.hs The second variant is strange, because if I use allocaArray instead of allocaBytes then it crashes. However, when I run the test twice, it works. http://code.haskell.org/storablevector/speedtest/SpeedTestExpCrashAllocaArr…

1 0

allowing unqualified method names in instance declarations
by Henning Thielemann 14 Nov '09

14 Nov '09

The attached patch shall help accepting an instance declaration like > import qualified Custom > > instance Custom.Class T where > methodA = ... > methodB = ... by translating it to > instance Custom.Class T where > Custom.methodA = ... > Custom.methodB = ... I don't know, whether this also works if you do > import qualified Custom hiding (methodA, methodB, ) John recently sent two patches to jhc(a)haskell.org, that are not present in the darcs repository. Do they need some confirmation? They work for me, as far as I can judge.

3 4

ignore optimizer phase information for INLINE pragmas as available in GHC
by Henning Thielemann 14 Nov '09

14 Nov '09

In low-level packages like bytestring and storablevector we have pragmas like {-# INLINE [1] func #-} The attached patch ignores the number, that specifies the optimizer phase where inlining shall happen. Ignoring that information is certainly not the best way, since an "inline only in phase x" may be meant as "better inline too seldom than too often". But this way I might get bytestring compiled without many modifications. At least I like to give a warning, when phase information is ignored, but I could not get that for now. Now when compiling Data.ByteString I am lost at: ./Data/ByteString.hs:1942 - Error: Unknown name: illegalOperationErrorType ./Data/ByteString.hs:1942 - Error: Unknown name: mkIOError cabal: Error: some packages failed to install: bytestring-0.9.1.5 failed during the building phase. The exception was: I don't know what the current state of exceptions in JHC is.

2 3

darcs patch: fix rewriting of OPTIONS pragma generated by new versi...
by John Meacham 14 Nov '09

14 Nov '09

Sat Nov 14 14:54:22 PST 2009 John Meacham <john(a)repetae.net> * fix rewriting of OPTIONS pragma generated by new versions of happy

1 0

transformers package on JHC
by Henning Thielemann 14 Nov '09

14 Nov '09

For those people who want to use the 'transformers' package on JHC I have made a patch: http://code.haskell.org/~thielema/transformers-jhc.patch I hesitate to push this to the 'transformers' main branch, since it contains only several syntactic replacements like: do ~(a,b) <- act replaced by do ab <- act let (a,b) = ab How about replacing 'mtl' dependency by 'transformers' in JHC's code? This would free us from some functional dependencies and might bring us closer to a JHC compiled JHC.

2 3

bug or feature: generated constraints of derived Show instance
by Henning Thielemann 14 Nov '09

14 Nov '09

With the patch about unqualified methods in instance declarations I can compile most of my xml-basic package. However it stops at data ProcessingInstruction name string = Known [Attr.T name string] | Unknown String deriving (Eq, Ord, Show ) since the Show instance of Attr.T needs an (Name.Attribute name) constraint. JHC does not seem to generate that automatically. GHC does so, but I do not know, whether this is Haskell 98. I can fix this by writing a custom Show instance, but there are more types with that problem. The file is http://code.haskell.org/~thielema/xml-basic/src/Text/XML/Basic/ProcessingIn… and the compiler message is: Typechecking... [1 of 4] Text.XML.Basic.ProcessingInstruction(....................user error ( What: failure Why: Signature too Weak: [Jhc.Show.Show Text.XML.Basic.ProcessingInstruction.v186,Jhc.Show.Show Text.XML.Basic.ProcessingInstruction.v187] does not imply [Text.XML.Basic.Name.Attribute Text.XML.Basic.ProcessingInstruction.v186] Where: on line 11 in <unknown> in the explicitly typed Instance@.iJhc.Show.showsPrec.Text.XML.Basic.ProcessingInstruction.T Text.XML.Basic.ProcessingInstruction.73_d (Text.XML.Basic.ProcessingInstruction.Known Text.XML.Basic.ProcessingInstruction.74_aa) = Jhc.Show.showParen (Text.XML.Basic.ProcessingInstruction.73_d Jhc.Order.>= 10) (Jhc.Show.showString "Known" Jhc.Basics.. Jhc.Show.showChar ' ' Jhc.Basics.. Jhc.Show.showsPrec 10 Text.XML.Basic.ProcessingInstruction.74_aa) Instance@.iJhc.Show.showsPrec.Text.XML.Basic.ProcessingInstruction.T Text.XML.Basic.ProcessingInstruction.75_d (Text.XML.Basic.ProcessingInstruction.Unknown Text.XML.Basic.ProcessingInstruction.76_aa) = Jhc.Show.showParen (Text.XML.Basic.ProcessingInstruction.75_d Jhc.Order.>= 10) (Jhc.Show.showString "Unknown" Jhc.Basics.. Jhc.Show.showChar ' ' Jhc.Basics.. Jhc.Show.showsPrec 10 Text.XML.Basic.ProcessingInstruction.76_aa) {- on line 11 -} Compilation of module: Text.XML.Basic.ProcessingInstruction)

2 1

Re: [Haskell-cafe] Opinion about JHC
by John Meacham 14 Nov '09

14 Nov '09

Hi, I believe this bug has been fixed in the darcs version, at least, it is in my testing. The fix will be in 0.7.3, or you can try the version from the darcs repository directly. Thanks for the bug report! John On Wed, Nov 11, 2009 at 01:36:05PM -0800, Philippos Apolinarius wrote: > Hi, John. > I > am sending you this second email because the first one has a worng paste. However, if I use the right input, I gent the same error. In fact, it gives exactly the same error for any input, right or wrong. > > {- file: tree.hs -} > {- compile: jhc tree.hs -dc -o > jtree } > import System (getArgs) > > import System.IO > > import IO > > > data Op = AND | OR | NOT deriving (Show, Read) > > > > data Tree= L Int | T Op [Tree] deriving (Show, Read) > > > main= do > > putStrLn "Give me a tree:" > > s <- getLine > let xx= read s > putStrLn (show (xx::Tree)) > > > Here is what happens when I try to run it: > > philip@desktop:~/jhctut$ ./jtree > Give me a tree: > T AND [L 1, L 2] > > jtree_code.c:2670: case fell off > Aborted > > It gives the same error for any input: > > philip@desktop:~/jhctut$ ./jtestarbo > Give me a tree: > fdsfkldkl > > jtestarbo_code.c:2670: case fell off > Abortado > > > It seems that the problem is in the Read class, since it works if I use the Show class only: > > import System (getArgs) > > import System.IO > > import IO > > > data Op = AND | OR | NOT deriving (Show, Read) > > > > data Tree= L Int | T Op [Tree] deriving (Show, Read) > > > main= do > > putStrLn "Give me a tree:" > > s <- > getLine > let xx= T AND [L 1, L 2] > putStrLn (show (xx::Tree)) > > I hope you can fix it. > > > > > > --- On Wed, 11/11/09, John Meacham <john(a)repetae.net> wrote: > > From: John Meacham <john(a)repetae.net> > Subject: Re: [Haskell-cafe] Opinion about JHC > To: haskell-cafe(a)haskell.org > Received: Wednesday, November 11, 2009, 1:37 AM > > On Tue, Nov 10, 2009 at 07:41:54PM -0800, Philippos Apolinarius wrote: > > I discovered a Haskell compiler that generates very small and fast > > code. In fact, it beats Clean. It has the following properties: > > Excellent. that was my goal ;) > > > 1 --- One can cross-compile programs easily. For instance, here is how I generated code for Windows: > > > > jhc --cross -mwin32 genetic.hs -o genetic > > Yup. This was a major goal. compiling for iPhones and embedded arches is > just as easy assuming you have a gcc toolchain set up. (at least with > the hacked iPhone SDK.. I have never tried it with the official one) > > > > > 2 -- It seems to be quite complete. > > > > 3 -- However, it often compiles a file, but the program fails to run. > > > > I have the following questions about it: > > > > 1 -- How active is the team who is writing the JHC compiler? > > Hi, I am the main contributor, but others are welcome and several have > made signifigant contributions. Development tends to be spurty. A lot of > work will get done in a short amount of time, this generally corresponds > to when an external contributor gets involved and the back and forth > helps stimulate patches on my part to complement theirs. > > Although I have not been able to devote a lot of my time to jhc in the > past, hopefully this will change in the not to distant future and I will > be able to work on it full time. > > > > 2 -- Is it complete Haskell? The author claims that it is; it compiled > > all programs that I wrote, but that does not mean much, because my > > programs are quite simple. > > It does Haskell 98 and several extensions, which is pretty much what GHC > does. However, it does not implement the same set of extensions as GHC > so this causes issues as a lot of people use GHC extensions extensively. > > I plan on supporting all of Haskell' of course, and the popular GHC > extensions to help compatibility. Not all are implemented. > > > 3 -- Why the Haskell community almost never talks about JHC? > > Part of it is that I am not very good at advocacy. I don't always > post announcements on the main haskell lists figuring the interested > parties are on the jhc list already. I do try to make jhc good, fast, > and usable, I always hoped someone better at advocacy than me would join > the project :) In truth, I think the spurty nature of development also > affects this, the list will be quite for a long time with a flurry of > development lasting a few weeks occasionally inspiring some discussion > in the other groups. > > In any case, I am glad you liked what you found! please join the mailing > list for jhc if you are interested in its development or using it. > > John > > > > -- > John Meacham - ⑆repetae.net⑆john⑈ - http://notanumber.net/ > -- > Haskell-Cafe mailing list > Haskell-Cafe(a)haskell.org > http://www.haskell.org/mailman/listinfo/haskell-cafe > > > > __________________________________________________________________ > Ask a question on any topic and get answers from real people. Go to Yahoo! Answers and share what you know at http://ca.answers.yahoo.com -- John Meacham - ⑆repetae.net⑆john⑈ - http://notanumber.net/

2 1

Region inference and no GC -- what kinds of programs?
by Jason Dusek 12 Nov '09

12 Nov '09

It's my understanding that, through region inference, JHC is able to generate binaries that are statically allocated. How applicable is this feature to long running server processes? Removal of GC latency can help in maintaining SLAs. I'd like to learn more about this feature of JHC -- how to know if it has been done, how it is triggered and what are some example programs that work with it. -- Jason Dusek

1 0

Multi parameter type classes
by Henning Thielemann 12 Nov '09

12 Nov '09

Is there also anonymous access to the JHC bug+feature-tracker? I like to add a request for multi-type parameter classes. These are necessary to compile NumericPrelude, what most of my signal processing code is based on. I need them for instance for vector spaces in order to write frequency filters and interpolations that work on both mono and stereo signals. I do not need functional dependencies. numericprelude$ grep -rh LANGUAGE src | sort --unique {-# LANGUAGE FlexibleInstances #-} {-# LANGUAGE MultiParamTypeClasses #-} {-# LANGUAGE NoImplicitPrelude #-}

2 1