July 2009 - Haskell-Cafe

Re: [Haskell-cafe] [Haskell Cafe] Data construction: how to avoid boilerplate code?
by Luke Palmer 30 Jul '09

30 Jul '09

On Wed, Jul 29, 2009 at 10:15 AM, Paul Sujkov<psujkov(a)gmail.com> wrote: > Hi Luke, > > I'm not pretty sure. The thing I don't like is the need to copy-and-paste > all the code with enumeration constructors. So, now I have two types to make > Type data from, but they could be many, so I'll have many almost identical > makeTypeFromFoo-functions. The thing I need is something like (*): > > makeType :: ? -> Type > makeType c = case c of > ("-$-" or 240) -> Status > ("-M-" or 64) -> Message > ("-W-" or 32) -> Warning Well, you could write a helper like this: matchType :: (Eq a) => (a,a,a) -> a -> Type matchType (status,message,warning) x | x == status = Status | x == message = Message | x == warning = Warning To reduce the size of your specifications: makeTypeStr = matchType ("-$-", "-M-", "-W-") makeTypeInt = matchType (240, 64, 32) There are trade-offs to doing something like this. It's smaller, but harder to read as specification. But, because it uses a tuple, it will catch you if you add a new case but forget to add it to one of the makeType*s (providing you remember to change matchType). What you're asking for puts all the conversions in the same place, which forbids them from being split out, decoupled, and modularlized. What if, instead of simple values, you had a more involved parser for these things? Even though it's kind of verbose, I think what you already have is fine. You do have to repeat the names, but it is still content code, and the relationship of the content to the names is explicit. You might find tables like this in a user's manual for your software... Luke > > then I have all this fabric code in one place. I could use something like > Either for this example, but it will scale fine up to two types to build > value from; while I want to have arbitrary numbers of them > > And from another point, it would be great if I could avoid doubling the code > for datatype declaration with empty (no parameter) constructors, and actual > "constructing" code from the different types. So it could look somewhat > alike this (pseudo-code) (**): > > data Type = Status <-- (String "-$-", Integer 240) > | Message <-- (String "-M-", Integer 64) > | Warning <-- (String "-W-", Integer 32) > > however, I'm not sure this one is really needed. Something for the (*) is > much more interesting > > 2009/7/29 Luke Palmer <lrpalmer(a)gmail.com> >> >> On Wed, Jul 29, 2009 at 6:27 AM, Paul Sujkov<psujkov(a)gmail.com> wrote: >> > Hi haskellers, >> > >> > I have a datatype of this sort: >> > >> > data Type = Status >> > | Message >> > | Warning >> > | Error >> > | StepIn >> > | StepOut deriving (Eq, Show) >> > >> > and (at this moment) two fabric-like functions: >> > >> > makeType :: String -> Type >> > makeType c = case c of >> > "-$-" -> Status >> > "-M-" -> Message >> > "-W-" -> Warning >> > "-E-" -> Error >> > "->>" -> StepIn >> > "<<-" -> StepOut >> > otherwise -> error "Uknown Message Type" >> > >> > deduceType :: Integer -> Type >> > deduceType n = case n of >> > 240 -> Status >> > 64 -> Message >> > 32 -> Warning >> > 8 -> StepOut >> > 4 -> StepIn >> > 2 -> Error >> > otherwise -> error "Unknown Message Type" >> > >> > how can I avoid boilerplate code at this stage? The thing that I really >> > need >> > is some n-type constructor, kind of a fabric for a variety of types with >> > the >> > possibility to add them on the fly (I have simple Integer and String >> > here, >> > but they could be much more sophisticated). I don't need the possibility >> > to >> > unpack the original value (e.g. "-$-" or 240) once the Type is created - >> > in >> > this case I can always have some sort of mapping to deduce it at any >> > moment, >> > so it will be redundant to carry it through the code explicitly >> >> I am not sure what you're asking. Are you saying that what you have >> written is boilerplate? What code are you writing that could be >> automatically deduced with enough smartness? >> >> Give an example of what you might like the solution to look like. >> >> Luke >> >> > The example is rather short and simple, but I have some more places in >> > code, >> > where the same problem is observed. Is there any generic solution? >> > Thanks in >> > advance >> > >> > -- >> > Regards, Paul Sujkov >> > >> > _______________________________________________ >> > Haskell-Cafe mailing list >> > Haskell-Cafe(a)haskell.org >> > http://www.haskell.org/mailman/listinfo/haskell-cafe >> > >> > > > > > -- > Regards, Paul Sujkov >

2 1

[Haskell Cafe] Data construction: how to avoid boilerplate code?
by Paul Sujkov 29 Jul '09

29 Jul '09

Hi haskellers, I have a datatype of this sort: data Type = Status | Message | Warning | Error | StepIn | StepOut deriving (Eq, Show) and (at this moment) two fabric-like functions: makeType :: String -> Type makeType c = case c of "-$-" -> Status "-M-" -> Message "-W-" -> Warning "-E-" -> Error "->>" -> StepIn "<<-" -> StepOut otherwise -> error "Uknown Message Type" deduceType :: Integer -> Type deduceType n = case n of 240 -> Status 64 -> Message 32 -> Warning 8 -> StepOut 4 -> StepIn 2 -> Error otherwise -> error "Unknown Message Type" how can I avoid boilerplate code at this stage? The thing that I really need is some n-type constructor, kind of a fabric for a variety of types with the possibility to add them on the fly (I have simple Integer and String here, but they could be much more sophisticated). I don't need the possibility to unpack the original value (e.g. "-$-" or 240) once the Type is created - in this case I can always have some sort of mapping to deduce it at any moment, so it will be redundant to carry it through the code explicitly The example is rather short and simple, but I have some more places in code, where the same problem is observed. Is there any generic solution? Thanks in advance -- Regards, Paul Sujkov

2 1

Re: Re: Proposal: TypeDirectedNameResolution
by Johannes Waldmann 29 Jul '09

29 Jul '09

While the discussion centers around overload resolution let me re-iterate a point that (e.g.,) Java does nicely: for their "x.f", if x :: T, then you write .f (unqualified) instead of .T.f (qualified), even if you needed qualification to declare the type of x, as in "foo.bar.T x;" This requires a mixture of type checking (for x) and name resolution (for f), which probably wouldn't fit with Haskell's current module and type system. Too bad ... Best - Johannes.

1 0

RFC: Unicode support in Alex
by Jean-Philippe Bernardy 29 Jul '09

29 Jul '09

Hello, I have modified the Alex lexer generator to support unicode. The general idea is that the state-machine works on the UTF8 representation of the text. I submit my work here for review in order to off-load the maintainer (Simon Marlow) as far as possible. The prototype is available on github: git://github.com/jyp/Alex.git Be sure to * checkout the "utf8" branch (so "git diff master" shows the changes) * Do a 2-stage bootstrapping before testing Caveats: * The generated code depends on some utf8 packages; * There is no attempt to fix the bytestring-based wrappers; * Left-context recognition is not table-based any more; * Presence of debug code. Bug reports, comments, and especially patches are welcome :) Thanks, -- JP

1 0

Adding a field to a data record
by Henry Laxen 29 Jul '09

29 Jul '09

Dear Group, It seems to me this should be easy, but I can't quite figure out how to do it without a lot of typing. Here is the question: Suppose you have a data type like: Data Foo = Foo { a :: Int, b :: Int, ... many other fields ... y :: Int } deriving (Eq, Read, Show, Typeable, Data) Now I would like to add a field z :: Int to the end of Foo. If I have a ton of data out on disk, which I wrote with, say writeFile "a.data" (show foo) -- where foo is a [Foo] say 1000 long, I would like to get a new "a.data" file which has a new z::Int field. So far the only way I can think of is to make a new Data Foo1, which includes the z::Int, read in a.data as a list of Foo, write a function like: fooTofoo1 :: Foo -> Foo1 fooTofoo1 xx = Foo1 {a = a xx, ... y = y xx, z = 1} then write the file back out, and perhaps use emacs to query-replace all the Foo1's back to Foo's, add the z::Int field back into Foo, and read it back. Please tell me there is a better way. Thanks in advance. Best wishes, Henry Laxen PS: I have read syb1, and syb2 a couple of times now, but so far haven't been able to connect it with this kind of problem.

8 9

ANN: The Monad.Reader - Issue 14
by Wouter Swierstra 29 Jul '09

29 Jul '09

Dear all, I am pleased to announce that a new issue of The Monad.Reader is now available: http://themonadreader.wordpress.com/ Issue 14 consists of the following three articles: * Fun with Morse Code by Heinrich Apfelmus * Hieroglyph 2: Purely Functional Information Graphics Revisited by Jefferson Heard * Lloyd Allison’s Corecursive Queues: Why Continuations Matter by Leon P Smith Please note that I've moved the Monad.Reader to a new Wordpress blog. You may want to update your bookmarks. If you’d like to write something for the next issue of The Monad.Reader, please get in touch. I haven’t fixed the deadline for the next issue just yet, but expect a deadline late 2009. Wouter

1 0

runProcess does not care about hSetBuffering?
by Magicloud Magiclouds 28 Jul '09

28 Jul '09

Hi, Code like: (or_, ow_) <- createPipe or <- fdToHandle or_ ow <- fdToHandle ow_ hSetBuffering ow LineBuffering hSetBuffering or LineBuffering h <- runProcess cmd [] Nothing Nothing Nothing (Just ow) Nothing In the cmd process, the ow is not LineBuffering.... -- 竹密岂妨流水过山高哪阻野云飞

2 1

an instance in other than the last type parameters
by Petr Pudlak 28 Jul '09

28 Jul '09

Hi, I have probably a very simple question, but I wasn't able to figure it out myself. Consider a two-parameter data type: > data X a b = X a b If I want to make it a functor in the last type variable (b), I can just define > instance Functor (X a) where > fmap f (X a b) = X a (f b) But how do I write it if I want X to be a functor in its first type variable? Is that possible at all? Something like: > instance Functor ??? where > fmap f (X a b) = X (f a) b Thanks in advance, Petr

8 12

ghci identifier listing question
by Daniel van den Eijkel 28 Jul '09

28 Jul '09

Hi, in GHCi, after pressing the tab-key, all defined identifiers are listed. If an identifier was redefined, all old and shadowed versions are listed, too. Is it possible, to reach the (shadowed) values in any way? Another question: Is it possible to show only those identifiers that are defined during the interactive session? I only can list all identifiers which start with a given prefix, but I would like to know if GHCi can show me all identifiers defined in a given module or in the current session (by pressing Tab or alike). Best regards, Daniel

2 2

A Question of Restriction
by Brian Troutwine 28 Jul '09

28 Jul '09

Hello all. I would like to define a data type that is the super-set of several types and then each of the proper subset types. For example: data Foo = One | Two | Three | Four data Odd = One | Three data Even = Two | Four This, of course, does not work. It seems that such a thing should possible to express entirely in the type system, but I cannot think of how. Would someone be so kind as to explain how this sort of thing can be accomplished? Thanks, Brian

8 9