Hello! I found out, that GHC implements typeclasses as an extra argument, a record that stores all functions of the typeclass. So I was wondering, is there a way (apart from using newtype) to pass a custom record as the typeclass record, to modify the behavior of the typeclass? I thought about something like this: f :: Show a => [a] -> String f = (>>= show) -- So, f becomes something like this? __f :: ClassShow a -> [a] -> String __f (ClassShow __show) x = x >>= __show -- And if I call the function, it looks somewhat like this: g :: [Int] -> String g = f __g = __f instanceShowInt -- But is it possible to do something like this? g2 = __f (ClassShow (return . fromEnum)) Tis is just a random thought, some compilers like JHC implement them by another way. But would this theoretically be possible? Yours, FUZxxl
Robert Clausecker
I found out, that GHC implements typeclasses as an extra argument, a record that stores all functions of the typeclass. So I was wondering, is there a way (apart from using newtype) to pass a custom record as the typeclass record, to modify the behavior of the typeclass? I thought about something like this:
f :: Show a => [a] -> String f = (>>= show)
-- So, f becomes something like this? __f :: ClassShow a -> [a] -> String __f (ClassShow __show) x = x >>= __show
-- And if I call the function, it looks somewhat like this: g :: [Int] -> String g = f
__g = __f instanceShowInt
-- But is it possible to do something like this? g2 = __f (ClassShow (return . fromEnum))
Tis is just a random thought, some compilers like JHC implement them by another way. But would this theoretically be possible?
If I understand you right, you would like to decide about the instance at run-time instead of at compile-time. This is actually possible in practice. A suitable search term is "implicit configurations", in particular "reification". Greets, Ertugrul -- nightmare = unsafePerformIO (getWrongWife >>= sex) http://ertes.de/
On Mon, May 16, 2011 at 8:10 AM, Robert Clausecker
I found out, that GHC implements typeclasses as an extra argument, a record that stores all functions of the typeclass. So I was wondering, is there a way (apart from using newtype) to pass a custom record as the typeclass record, to modify the behavior of the typeclass? I thought about something like this:
Would GHC's implicit parameter extension possibly suit your purposes here? Your example would translate as: {-# LANGUAGE ImplicitParams #-} type ShowClass a = a -> String f :: (?showC :: ShowClass a) => [a] -> String f x = x >>= ?showC g :: [Int] -> String g = let ?showC = show in f g2 :: [Int] -> String g2 = let ?showC = (return . toEnum) in f ...where: > g [72, 97, 115, 107, 101, 108, 108] "7297115107101108108" > g2 [72, 97, 115, 107, 101, 108, 108] "Haskell" Clearly this doesn't allow you retrofit such functionality onto existing code using existing type classes, but I'd be wary of doing that anyway--type class instances are not something that code will expect to have changing out from under it. Otherwise, this seems to be exactly what the implicit parameters extension is designed for, judging from the way the GHC user's guide describes it. - C.
Robert,
2011/5/16 Robert Clausecker
I found out, that GHC implements typeclasses as an extra argument, a record that stores all functions of the typeclass. So I was wondering, is there a way (apart from using newtype) to pass a custom record as the typeclass record, to modify the behavior of the typeclass? I thought about something like this:
You may be interested in Agda's upcoming "instance arguments" (inspired upon Scala implicits and Agda's implicit arguments). These will be available in Agda 2.2.12 (you may find references to an older name "non-canonical implicit arguments"). The new type of function arguments are automatically inferred from call-site scope unless they are explicitly provided. Type classes are directly (not just under the hood) modelled as records, and you can do what you suggest. You can also define "local instances", and there are other advantages. We have chosen a more limited-power instance search though. More discussion online. http://wiki.portal.chalmers.se/agda/pmwiki.php?n=ReferenceManual.InstanceArg... http://people.cs.kuleuven.be/~dominique.devriese/agda-instance-arguments/ I believe a similar Haskell extension (perhaps with a less principled instance search) would improve and simplify Haskell's type class system. By the way, Kahl and Scheffczyk proposed extending Haskell with "named instances" in 2001 which allowed something like this to a limited extent. Look for "Named instances for Haskell Type Classes" in Google Scholar. Dominique
participants (4)
-
Casey McCann -
Dominique Devriese -
Ertugrul Soeylemez -
Robert Clausecker