June 2014 - Haskell-Cafe

1 = 0?
by Alexey Muranov 17 Jun '14

17 Jun '14

Is this the expected behavior that 1 = 0 does not raise any error? What does this mean? Thanks. Alexey.

8 8

data/newtype vs. class
by martin 16 Jun '14

16 Jun '14

Hello all, I find myself trying to define a class, where a newtype would suffice. E.g. once I was pondering over "named" things and my initial thought was "well that's a class, which asks for a function 'getName'". But then here http://www.haskell.org/pipermail/haskell-cafe/2010-June/078803.html someone showed me a solution without any classes. Recently I was pondering over "timed things" (things which change at specific ponts in time) and again my initial thought was "that's a class, which asks for a method 'at'". Then I remembered things like State, i.e. data types which wrap around functions. State is not a class, which asks for a function 'nextState', it is all defined with data. Can someone give me some guidance, when to use classes and when to use data? Are the two concepts interchangeable in some scenarios? Martin

4 4

Re: [Haskell-cafe] Uploading to hackage fails: bad file names in tarball
by Niklas Larsson 16 Jun '14

16 Jun '14

Hi! For extremely ancient and cludgy reasons AUX isn't a valid filename on windows, it's a DOS device. The other similarly reserved name are CON, PRN and NUL. Niklas ----- Ursprungligt meddelande ----- Från: "Arie Peterson" <ariep(a)xs4all.nl> Skickat: ‎2014-‎06-‎16 10:45 Till: "haskell-cafe(a)haskell.org" <haskell-cafe(a)haskell.org> Ämne: [Haskell-cafe] Uploading to hackage fails: bad file names in tarball Since some time, I have been unable to upload packages to Hackage, via either the 'cabal upload' command, or using the web form. The error message it gives is this: ==== ✂ ==== Hackage username: AriePeterson Hackage password: Uploading dist/np-linear-0.1.1.1.tar.gz... Error: dist/np-linear-0.1.1.1.tar.gz: 400 Bad Request Error: Invalid package Invalid windows file name in tar archive: "np-linear-0.1.1.1\\src\\Aux.hs". For portability, hackage requires that file names be valid on both Unix and Windows systems, and not refer outside of the tarball. ==== ✂ ==== The tarball is created by 'cabal sdist': cabal-install version 1.18.0.2 using version 1.18.1.1 of the Cabal library. I also installed the newest cabal-install, on another machine, but this did not help (same error). I also tried to create a tarball by hand, using 'tar --format=ustar', but this again resulted in the same error message. By the way, I'm on linux, not Windows, so it is not clear how the backslashes get in the file names. What could be going on? _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe(a)haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

4 5

What is GHC 7.8.2 doing here (typed holes, data+poly kinds, type families etc.)
by Dominic Mulligan 16 Jun '14

16 Jun '14

Hi, I've come across some behaviour in GHC 7.8.2 which seems strange, and may be a bug, but I wish to check before reporting it. Consider the minimum example pasted at the bottom of this message. At the line plusComm (SS m') n' = _ GHC tells me that I have a single hole in the file with type 'S (m1 :+: n) :~: (n :+: 'S m1) which is straightforward to provide, by rewriting with the inductive hypothesis (a recursive call of plusComm) and then some fiddling. However, replacing this line with the refinement plusComm (SS m') n' = substR (plusComm m' n') _ GHC tells me that the remaining hole now has type 'S (m1 :+: n) :~: (m1 :+: n) which seems completely incorrect to me. Shouldn't this hole have type 'S (n :+: m1) :~: (n :+: S m1) ? Can anybody give any clues as to what is happening? Thanks, Dominic {-# LANGUAGE DataKinds #-} {-# LANGUAGE GADTs #-} {-# LANGUAGE KindSignatures #-} {-# LANGUAGE PolyKinds #-} {-# LANGUAGE TypeFamilies #-} {-# LANGUAGE TypeOperators #-} module MinimumExample where data (:~:) (a :: k) (b :: k) where Refl :: a :~: a substR :: a :~: b -> f a -> f b substR Refl x = x substL :: a :~: b -> f b -> f a substL Refl x = x cong :: a :~: b -> f a :~: f b cong Refl = Refl data Nat where Z :: Nat S :: Nat -> Nat data NatS (m :: Nat) where ZS :: NatS Z SS :: NatS m -> NatS (S m) type family (:+:) (m :: Nat) (n :: Nat) :: Nat where Z :+: n = n (S m) :+: n = S (m :+: n) plusZRightNeutral :: NatS m -> (m :+: Z) :~: m plusZRightNeutral ZS = Refl plusZRightNeutral (SS m') = cong (plusZRightNeutral m') plusComm :: NatS m -> NatS n -> (m :+: n) :~: (n :+: m) plusComm ZS n' = substL (plusZRightNeutral n') Refl -- everything seems correct here plusComm (SS m') n = _ -- but things look completely incorrect, here -- plusComm (SS m') n' = substR (plusComm m' n') _

2 2

Re: [Haskell-cafe] Uploading to hackage fails: bad file names intarball
by Niklas Larsson 16 Jun '14

16 Jun '14

The extension wasn't really part of the filename in those days. So aux.c and aux.hs is interpreted as aux. The funny thing is that Microsoft soon recognized how bad it was and added a flag to make it unix-like (\dev\nul etc) and free the names in dos 2.0, but the default stayed as it was. Niklas ----- Ursprungligt meddelande ----- Från: "Frerich Raabe" <raabe(a)froglogic.com> Skickat: ‎2014-‎06-‎16 13:15 Till: "haskell-cafe(a)haskell.org" <haskell-cafe(a)haskell.org> Ämne: Re: [Haskell-cafe] Uploading to hackage fails: bad file names intarball Hi, There are a couple more special files, most (all?) of which are documented at http://msdn.microsoft.com/en-us/library/windows/desktop/aa365247(v=vs.85).a… That page also says "Also avoid these names followed immediately by an extension; for example, NUL.txt is not recommended." and I suspect that it may be this recommendation based on which the error is raised. "NUL.txt" as well as "Aux.hs" are valid file names though. There is no file named 'AUX' though, and "Aux.hs" is a valid filename. On 2014-06-16 11:45, Niklas Larsson wrote: > Hi! > > For extremely ancient and cludgy reasons AUX isn't a valid filename on > windows, it's a DOS device. The other similarly reserved name are CON, > PRN and NUL. > > Niklas > > ----- Ursprungligt meddelande ----- > Från: "Arie Peterson" <ariep(a)xs4all.nl> > Skickat: ‎2014-‎06-‎16 10:45 > Till: "haskell-cafe(a)haskell.org" <haskell-cafe(a)haskell.org> > Ämne: [Haskell-cafe] Uploading to hackage fails: bad file names in tarball > > Since some time, I have been unable to upload packages to Hackage, via > either > the 'cabal upload' command, or using the web form. > > The error message it gives is this: > > ==== ✂ ==== > Hackage username: AriePeterson > Hackage password: > Uploading dist/np-linear-0.1.1.1.tar.gz... > Error: dist/np-linear-0.1.1.1.tar.gz: 400 Bad Request > Error: Invalid package > > Invalid windows file name in tar archive: > "np-linear-0.1.1.1\\src\\Aux.hs". > For portability, hackage requires that file names be valid on both Unix > and > Windows systems, and not refer outside of the tarball. > ==== ✂ ==== > > The tarball is created by 'cabal sdist': > cabal-install version 1.18.0.2 > using version 1.18.1.1 of the Cabal library. > > I also installed the newest cabal-install, on another machine, but this > did > not help (same error). > > I also tried to create a tarball by hand, using 'tar --format=ustar', but > this > again resulted in the same error message. > > By the way, I'm on linux, not Windows, so it is not clear how the > backslashes > get in the file names. > > > What could be going on? > > _______________________________________________ > Haskell-Cafe mailing list > Haskell-Cafe(a)haskell.org > http://www.haskell.org/mailman/listinfo/haskell-cafe > > _______________________________________________ > Haskell-Cafe mailing list > Haskell-Cafe(a)haskell.org > http://www.haskell.org/mailman/listinfo/haskell-cafe -- Frerich Raabe - raabe(a)froglogic.com www.froglogic.com - Multi-Platform GUI Testing _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe(a)haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

1 0

Fwd: Problem with loading unix-compat in GHCi on Windows: is suggested solution acceptable?
by German Zhivotnikov 16 Jun '14

16 Jun '14

Hi all, There is a minor problem with unix-compat module on Windows: when one tries to load it (or the module that depends on it) into GHCi, i.e. in REPL mode, - GHCi refuses to load unix-compat with a message like that: ghc.exe: (skipped).cabal-sandbox\x86_64-windows-ghc-7.8.2\unix-compat-0.4.1.1\HSunix-compat-0.4.1.1.o: unknown symbol 'snprintf' ghc.exe: unable to load package 'unix-compat-0.4.1.1' Note that it only concerns REPL mode: when GHC is building a binary that uses unix-compat, - it produces correct executable that works fine. My hypothese is that this is because in typical Win32 build environment snprintf resides in libmingwex.a, which is static library, which GHCi is incapable of loading. Please correct me if I'm wrong. There is a solution - to make unix-compat depend on MSVCRT.DLL (instead of static mingwex.a) and replace 'snprintf' with Microsoft's '_snprintf' (and there's also stat/_stat to be replaced). However this introduces at least new dependency on MSVCRT.DLL, and maybe something else, which might be unexpected. Therefore, before accepting the solution into official codebase, it would be great to have an input - if anyone can foresee possible unintended side effects from suggested solution for code that depends on unix-compat module. Corresponding pull request resides at https://github.com/jystic/unix-compat/pull/12/files - could you possibly take a look if you've got something to say on that issue. Thank you in advance.

1 0

Uploading to hackage fails: bad file names in tarball
by Arie Peterson 16 Jun '14

16 Jun '14

Since some time, I have been unable to upload packages to Hackage, via either the 'cabal upload' command, or using the web form. The error message it gives is this: ==== ✂ ==== Hackage username: AriePeterson Hackage password: Uploading dist/np-linear-0.1.1.1.tar.gz... Error: dist/np-linear-0.1.1.1.tar.gz: 400 Bad Request Error: Invalid package Invalid windows file name in tar archive: "np-linear-0.1.1.1\\src\\Aux.hs". For portability, hackage requires that file names be valid on both Unix and Windows systems, and not refer outside of the tarball. ==== ✂ ==== The tarball is created by 'cabal sdist': cabal-install version 1.18.0.2 using version 1.18.1.1 of the Cabal library. I also installed the newest cabal-install, on another machine, but this did not help (same error). I also tried to create a tarball by hand, using 'tar --format=ustar', but this again resulted in the same error message. By the way, I'm on linux, not Windows, so it is not clear how the backslashes get in the file names. What could be going on?

3 3

improved jhc manual
by John Meacham 16 Jun '14

16 Jun '14

Hi all, I just wanted to show off the new jhc manual http://repetae.net/computer/jhc/manual.html If anyone has any feedback, that would be great, even better would be patches of course :) Manual writing is not my strong point. The entire manual is in the darcs repo as markdown formatted files or as literate code embedded in the source files. Something I need to figure out is how to get haddock to not barf on jhc's input files. A problem with haddock is that it tries to replicate ghc's module chasing algorithm and falls down when anything is slightly off, jhc's ability to handle recursive modules throws it for a loop. Now I can easily get jhc to spit out fully resolved module names, it does so with the --annotate-source option, but need some sort of hook into haddock saying "don't try to resolve module names, here they all are." Even haddock just having a 'keep going' option would be great to get something out there. Or even improving haddock to handle recursive module imports. John -- John Meacham - http://notanumber.net/

4 7

type families and type classes
by Gautier DI FOLCO 15 Jun '14

15 Jun '14

Hi all, I have a bunch of data types which are "divided" in other data types as following: type FileName = String data FileWish = File FileName deriving (Show, Eq) data FileFact = FileExist | FileDontExist deriving (Show, Eq) data FileAction = FileCreate deriving (Show, Eq) observe :: FileWish -> [FileFact] observe :: undefined plan :: FileWish -> FileFact -> [FileAction] plan (File _) FileDontExist = [FileCreate] plan (File _) FileExist = [] perform :: FileWish -> [FileAction] -> IO () perform :: undefined I want to have a model as extensible as possible, even if I'm new to Haskell, I think that it's preferable to base parts of system on behavior (type classes) rather than data types. So I have File that give a Wish, a Fact (via observe) and a Action (via plan). I want to put observe, plan and perform in different type classes (let's say Observable, Plannable and Performable). When I create a new thing (like File) I want, for example, that : observe :: Wish e -> [Fact e] where e is File. But I want (Fact e) be an instance of Plannable. And so on. At first glance, I think I have to use this: https://www.fpcomplete.com/school/to-infinity-and-beyond/pick-of-the-week/t… But it is coupled to data type and one type class. If you have any hints for me. Maybe I don't take the right way to solve my problem. Thanks in advance.

1 0

Taking a number to a power and taking its mod
by Rafael Almeida 15 Jun '14

15 Jun '14

Hello, I've been trying to do a problem from SPOJ which requires me to find prime numbers in a range. I've been battling with several possibility, so far I have had no luck. I remember primality tests from college. So I looked up the one I knew: fermat's little theorem. It requires me to do this check: a^(p-1) `mod` p == 1 However, taking a number to a large power is very slow. So it was even worse than the solution I was already trying. I'm not here for your guys to give me the solution. But I'm here to understand how can I do that operation quickly. This wiki entry implements powMod http://www.haskell.org/haskellwiki/Testing_primality Calling that funcion with powMod 999999527 2 (999999527-1) is very fast. What's going on? Can someone shed me some light? []'s Rafael

2 2