On Dec 19, 2006, at 10:11 PM, Dan Weston wrote:
> instance CommandFunction (Sh st ()) st where
^
I think your first argument (on which the second has a functional
dependence) does not determine the second argument, since it makes
use of st in the first argument. This strikes me as a likely place
to begin.
No, I'm pretty sure this isn't a problem. The second argument is
determined _because_ it is mentioned in the first. The functional
dependencies and instance declarations work, as long as I can make
the compiler accept them. They are only being rejected by the
termination-checking part of the algorithm.
That said, I'm open to the idea of reformulating these instances. In
fact, I don't really like the fact that I need FDs. It seems to me
that I should somehow be able to eliminate the second argument
altogether and thus the FD, but I can't seem to figure it out.
Dan
Robert wrote:
Fellow Haskellers,
I have a package that uses some light typeclass hackery to
automaticly
build parsing algorithms based on the type of a function.
I was recently informed that my package doesn't compile on GHC 6.6
due
to the new restrictions on FD resolution; in particular I have
instance
declarations which fail the coverage condition. I can use
undecidable
instances to make the package compile again, but I'd prefer not to
if I
can avoid it.
class CommandFunction f st | f -> st where
parseCommand :: String -> f -> CommandParser st
commandSyntax :: f -> [Doc]
instance CommandFunction (Sh st ()) st where
parseCommand wbc m str =
-- list monad
do (x,[]) <- runRegex (maybeSpaceBefore (Epsilon
(CompleteParse
m))) str
return x
commandSyntax _ = []
instance CommandFunction r st
=> CommandFunction (Int -> r) st where
parseCommand = doParseCommand Nothing intRegex id
commandSyntax f = text (show intRegex) : commandSyntax (f
undefined)
instance CommandFunction r st
=> CommandFunction (Integer -> r) st where
parseCommand = doParseCommand Nothing intRegex id
commandSyntax f = text (show intRegex) : commandSyntax (f
undefined)
Speak softly and drive a Sherman tank.
Laugh hard; it's a long way to the bank.
-- TMBG
_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe
