Difference between revisions of "Type signatures as good style"
(mention rank n types) |
(type inference not always decidable) |
||
| Line 19: | Line 19: | ||
Where <hask>ShowS</hask> is <hask>String -> String</hask> rather than <hask>a -> a</hask>. | Where <hask>ShowS</hask> is <hask>String -> String</hask> rather than <hask>a -> a</hask>. | ||
| − | + | Even more, for some type extensions the automatic inference fails, | |
| − | + | e.g. the higher-order types used by <hask>Control.Monad.ST.runST</hask> | |
| − | |||
<haskell> | <haskell> | ||
runST :: (forall s . ST s a) -> a | runST :: (forall s . ST s a) -> a | ||
Revision as of 10:40, 25 December 2008
Question
Since Haskell type checkers can automatically derive types of expressions why shall I put explicit type signatures in my programs?
Answer
Using explicit type signatures is good style and GHC with option -Wall warns about missing signatures.
Signatures are a good documentation and not all Haskell program readers have a type inference algorithm built-in.
There are also some cases where the infered signature is too general for your purposes.
E.g. the infered (most general) type for asTypeOf is a -> b -> a,
but the purpose of asTypeOf is to unify the types of both operands.
The more special signature a -> a -> a is what you want and it cannot be infered automatically.
Another example:
emptyString :: ShowS
emptyString = id
Where ShowS is String -> String rather than a -> a.
Even more, for some type extensions the automatic inference fails,
e.g. the higher-order types used by Control.Monad.ST.runST
runST :: (forall s . ST s a) -> a
cannot be inferred in general, because the problem is undecidable. In GHC, they are enabled with the language pragma RankNTypes.
How to add a bunch of signatures?
Ok, this convinced me. How can I add all the signatures I did not write so far?