- Haskell-Cafe - Haskell.org

Semi-closed handles and I/O errors: misleading documentation?
by Daniel Díaz 29 Nov '15

29 Nov '15

In the documentation for System.IO <http://hackage.haskell.org/package/base-4.8.1.0/docs/System-IO.html>, we find the following assertion: Any I/O errors encountered while a handle is semi-closed are simply discarded. also this other one: Encoding and decoding errors are always detected and reported, except during lazy I/O (hGetContents, getContents, and readFile), where a decoding error merely results in termination of the character stream, as with other I/O errors. Is this really correct, though? This SO question <http://stackoverflow.com/questions/16257037/detecting-i-o-exceptions-in-a-l…> says that exceptions found during lazy I/O are thrown from pure code, and a small experiment <http://stackoverflow.com/a/33985047/1364288> seems to confirm it. I personally prefer that behaviour, by the way. Perhaps the documentation should be updated?

1 0

Re: [Haskell-cafe] Compatibility etiquette for apps, with cabal sandboxes and `stack`
by Paolo Giarrusso 29 Nov '15

29 Nov '15

On Sunday, November 29, 2015 at 1:23:58 PM UTC+1, Michael Snoyman wrote: > > For your particular example, you can use the transformers-compat package. > Thanks, that's useful! But to answer your general question: for applications, I strongly advocate > supporting just a single set of package version combos for dependencies. > Stack does do this by default, but cabal freeze files can get you most of > the way there too (barring some corner cases with different OSes and > OS-specific packages). > > I'm sure others will disagree with this recommendation, but I see no > reason to absorb a large amount of compatibility work just so that two > installations of the same application may end up behaving differently at > runtime because of differences in the behavior of a dependency. > Personally, I switched all the way to stack. I added mention of cabal sandboxes to address anybody who didn't switch yet or who prefers to stick to cabal.

1 0

Compatibility etiquette for apps, with cabal sandboxes and `stack`
by Paolo Giarrusso 29 Nov '15

29 Nov '15

Hi all, IIUC, people used to spend nontrivial effort making their Haskell tools work across a range of dependencies, and be careful about dropping support for older ones. Do cabal sandboxes or Stack reduce that need, at least for applications?* Or conversely, how bad is it to restrict support to users having them? I guess I am asking about common policies, but this probably depends on adoption of those tools. As in: a) without sandboxes or Stack, packages need to build with whatever environment is there. b) with either of sandboxes or Stack, you can just specify the environment and get it. With Stack, you can even specify whichever (supported) GHC you want, without impacting the environment. Concretely, Control.Monad.Trans.Error is deprecated in transformers-0.4.3.0, but the replacement Control.Monad.Trans.Except does not exist in transformers-0.3.x, thus in the one-but-last Haskell Platform (2014.2.0.0). Hence, fixing the deprecation would make this application hard to install** for some users, including past me (around two months ago).*** Would you apply such a patch? Would you keep it out? Or would you even use CPP (something I'd really like to avoid)? Cheers, Paolo * I take for granted libraries are a different matter — composing libraries means intersecting their supported versions. ** I'll omit detailing "hard to install", because that's a matter of opinion; I considered `constraint: transformers installed` necessary in my cabal config (https://www.vex.net/~trebla/haskell/cabal-cabal.xhtml) and I've read that's becoming the default in the HP, so I think users are entitled to this opinion. *** This particular app has a tiny userbase, so my question is a curiosity.

2 1

How to create a Wiki account?
by Jeffrey Brown 29 Nov '15

29 Nov '15

This [1] is as close as I can seem to get. I noticed [2] had obsolete information; it says Either is not by default a Monad. [1] https://wiki.haskell.org/HaskellWiki:New_accounts [2] https://wiki.haskell.org/Failure -- Jeffrey Benjamin Brown

3 2

Why was this function that fast ?
by Romain Gehrig 28 Nov '15

28 Nov '15

Hi, I implemented a function to count how many steps it takes for a number to be reduced to 1 using the Collatz Conjecture (also known as the 3n + 1 problem) and I was very surprised to see it working instantaneously for integers as big as 10^100000. The said function: collatzSteps :: Integer -> Integer collatzSteps n = steps 1 n where steps acc 1 = acc steps acc n' | even n' = steps (acc+1) (n' `div` 2) steps acc n' = steps (acc+1) (n' * 3 + 1) You can try in GHCi (apparently with an upper limit of 10^199669): GHCi> collatzSteps 10^199668 [big num...] GHCi> length $ show $ collatzSteps 10^199668 168740 GHCi> collatzSteps 10^199669 -- Oops [1] 36128 bus error ghci Notice the number 168740: it is the number of *digits* for the number of steps. It implies that the `acc` is ~= *10^168740* when the function ends. Without optimisation at some point, it would mean there were as much (acc+1) calls ! My understanding completely stops there; I can't comprehend why a function that shouldn't return anything even long after the end of the universe just finished before my eyes. Can someone shed some light on how this peculiar function was optimised by GHC? Many thanks, (still dazed) Romain

3 2

Info about List with terminal element
by Silvio Frischknecht 26 Nov '15

26 Nov '15

Hello, I'm interested in learning more about a list-like structure with a terminal element. I.e. data List t e = Cons e (List t e) | Null t or maybe data List t e = Cons e (List t e) | Null | Terminal t An practical application could be lazy reading from a file readFile :: String -> List Error String Obviously there is a Functor instance. There is even a Monad instance. However, it is not defined as obviously because it's not clear what the terminal element should be. There is a straight forward fold like structure. My questions are. Are there more interesting generalizations of List? What rules do they follow? How should the Monad work? Why? I know this is related to the whole Iteratee discussion (I'm not completely familiar with it). Cheers Silvio

5 10

Could not deduce Monad m => MonadReader r (ReaderT r m) ??
by David Turner 26 Nov '15

26 Nov '15

Porting some code from GHC 7.8.4 to GHC 7.10.2 and have encountered this rather odd failure: Could not deduce (MonadReader (PqColumn m -> m B.ByteString) (ReaderT (PqColumn m -> m B.ByteString) m)) arising from a use of ‘ask’ from the context (Monad m) bound by the type signature for readFieldByteString :: Monad m => PqColumn m -> RowReader m B.ByteString at src/EventStore/LibPq.hs:164:24-72 In a stmt of a 'do' block: readField <- ask In the expression: do { readField <- ask; lift $ readField col } In an equation for ‘readFieldByteString’: readFieldByteString col = do { readField <- ask; lift $ readField col } I wondered if it was because the environment type (PqColumn m -> m B.ByteString) itself was an instance of MonadReader and the typechecker was getting confused, so I put the environment in a newtype and it was indeed happy. Unfortunately I can't share enough of this code to get the problematic fragment to compile, and my initial attempts at a small reproduction have been unsuccessful. It's something like the following, but this compiles just fine. class MyClass m where type MyType m defaultValue :: m (MyType m) l :: Monad m => MyType m -> ReaderT (MyType m -> m Int) m Int l n = do f <- ask lift $ f n Before I put more effort into reproducing this, does anyone recognise this failure or know what might be going on here? Cheers, David

1 0

Re: [Haskell-cafe] Type-level lazy bool operations
by Oleg 26 Nov '15

26 Nov '15

Dmitry Olshansky wrote: > There are some type-level boolean operation (If, &&, ||, Not) in > Data.Type.Bool. Are they lazy enough? > > I faced with problem and it seems that the reason is non-lazy "If". > I.e. both (on-True and on-False) types are trying to be calculated. > Does it work in this way really? Could it be changed? Just a remark: if (type-level) computations are pure, then whether they are done lazily or in any other way should not matter. The evaluation strategy invariance is the main characteristic, if not the definition, of purity. Type-level computations are not pure however: they may have an effect of divergence, or generating a type-error. Thus we really need a type-level If, which evaluates only one of the two conditional branches but never both. It can be easily implemented using the standard approach of delaying evaluation with a thunk. Please see http://okmij.org/ftp/Haskell/TTypeable/TTypeable.hs and search for ORELSE (near the end of the file). See http://okmij.org/ftp/Haskell/typeEQ.html#TTypeable for background.

2 2

regex-posix broken on 64 bit windows.
by Gershom B 26 Nov '15

26 Nov '15

Here’s a ticket reporting the issue: https://github.com/haskell/haskell-platform/issues/144 The ticket is on the platform site, but it doesn’t really belong there. On the other hand, there’s no tracker for regex-posix anywhere else. This problem has probably existed for a _long_ time, and I guess has to do with 64-bit incompatibilities in the cbits shipped with the package. I’ve cc’d Chris Kuklewicz as he is author and maintainer. But I know that he has taken a step back from these packages some time ago. Generally, I would be happy to consider them stable-ish and let the community just migrate away from them or not as open source typically evolves. But regex-posix is pretty widely used still and is in the platform, so I think it is unfortunate that it is broken in this way. It would be great if some interested volunteer were to step up and fix this issue and maybe even offer to take over maintainership (to which I’m sure Chris would have no objection?). (Relatedly, it would be good to have a future discussion on which regex library, if any, is suitable for inclusion in the platform. At this point my inclination would be regex-tdfa-text, although my sense is nearly all the libs are relatively unmaintained at this point?) —Gershom

1 0

[ANN] Linode package
by Sebastian 25 Nov '15

25 Nov '15

Hello everybody, I'm happy to announce the initial release of the Linode package. It contains bindings to the Linode API and a few user friendly routines to setup and destroy VPS instances. The next major release will cover private IPs and load balancing. Hackage: http://hackage.haskell.org/package/linode Github: https://github.com/Helkafen/haskell-linode A working example is provided in the docs. It creates one linode instance and performs a dummy install: https://hackage.haskell.org/package/linode-0.1.0.4/docs/Network-Linode.html Use cases include: - Running some daily batch, with one or several instances - Dynamically adding some resources to a long running process - Anything you would normally do with a VPS Ideas, contributions and feedback are welcome! -- Sebastian de Bellefon

2 1