I am learning Haskell using Learn You a Haskell. On page 54 is an
implementation of take like so:
take' :: (Num i, Ord i) => i -> [a] -> [a]
take' n _
| n <= 0 = []
take' _ [] = []
take' n (x:xs) = x : take' (n-1) xs
I understand all the code apart from the first line.
The :: part I understand as meaning this is a type definition?
(Num i, Ord i) is a tuple. The first element of the tuple has to be
numeric, fair enough. the second param has to be able to be ordered.
The parameter is the same - both are i. This means that the types
have to be the same?
Why is it not (Num i, Ord j)? Isn't the 2nd tuple element
referring to the list? Which could be of any type?
What does => signify?
i -> [a] -> [a] means first parameter is numeric? 2nd param is any
type list, 3rd param is any type list. So this is saying first param
numeric, 2nd param a list of any type and it returns a list of any
type. Well that is understandable I suppose.
The stuff before the => are constraints. (Num i, Ord i) isn't really a tuple in the usual sense. It specifies a requirement that type class instances exist for whatever specific type i you call the function with.
So what this type signature is actually saying is that the type of take' is i -> [a] -> [a], but with the additional restriction that i must have Num and Ord instances, which amounts to requiring that you can do some basic arithmetic (Num being short for "numeric" I suppose) and compare which value is larger (Ord meaning values of that type have an ordering defined, i.e. you can sort them or such).
In this particular case, the comparison n <= 0 is what uses Ord, while the subtraction and numeric literals use Num.
This is covered in chapter 3 of LYAH, and these specific type classes (among others) are mentioned in the Typeclasses 101 section.
(Num i, Ord i) is not meant to be a tuple. Read the type signature as: take' is a function of the type i -> [a] -> [a] where i has to be a type of the typeclasses Num and Ord.
You might want to read the chapter "Typeclasses 101" again.
Related
This is an example in Learn You A Haskell, chapter on higher order functions:
compareWithHundred :: (Num a, Ord a) => a -> Ordering
compareWithHundred x = compare 100 x
While the idea of the function is clear for me, I'm not sure why type signature is (Num a, Ord a). We only pass integer that is to be compared to the function, of type Int. What Ord stands for here, and why is implicitly passed argument in type signature?
That's not the only possible signature for this signature. It happens to be the most general one. compareWithHundred :: Int -> Ordering is actually a possible instantiation – the polymorphic a argument can be instatiated with any orderable number type, which does sure enough include Int, but also Integer, Rational, Double...
Prelude> let compareWithHundred :: (Num a, Ord a) => a -> Ordering; compareWithHundred x = compare 100 x
Prelude> compareWithHundred (99 :: Int)
GT
Prelude> compareWithHundred (100.3 :: Double)
LT
Not all number types permit you to order-compare them though – the classical example where this is not possible are complex numbers (which have “more than one direction” in which you could order them).
Prelude Data.Complex> compareWithHundred (100 :+ 30 :: Complex Double)
<interactive>:10:1:
No instance for (Ord (Complex Double))
arising from a use of ‘compareWithHundred’
In the expression: compareWithHundred (100 :+ 30 :: Complex Double)
In an equation for ‘it’:
it = compareWithHundred (100 :+ 30 :: Complex Double)
Hence you need to require both that the argument is a number (so there exists a value 100 which to compare with) and that the argument is in the Ord class. This combined constrained is written (Num a, Ord a).
I have something to add, in case you couldn't gather something from leftaroundabout's thorough answer.
Everything to the left of => in a type signature is a constraint. Read the type like this:
compareWithHundred :: (Num a, Ord a) => a -> Ordering
^^^^^^^^^^^^^^ ^ ^^^^^^^^
constraints | |
argument type |
result type
So you only pass one argument to the function because there is only one argument in the type signature, a. a is a type variable, and can be replaced with any type as long as that type satisfies the constraints.
The Num a says that whatever you replace a with has to be numeric (so it can be Int, Integer, Double, ...), and the Ord a says that it has to be comparable. leftroundabout's answer goes into more detail about why you need both, I just wanted to make sure you knew how to read the signature.
So it's perfectly legal in one sense to say compareWithHundred "foobar", the type checker says that that expression's type is Ordering, but then it will fail later when it tries to check that there is a Num String instance.
I hope this helps.
I am trying to figure out the way Haskell determines type of a function. I wrote a sample code:
compareAndIncrease a b =
if a > b then a+1:b:[]
else a:b:[]
which constructs a list basing on the a > b comparison. Then i checked its type with :t command:
compareAndIncrease :: (Ord a, Num a) => a -> a -> [a]
OK, so I need a typeclass Ord for comparison, Num for numerical computations (like a+1). Then I take parameters a and b and get a list in return (a->a->[a]). Everything seems fine. But then I found somewhere a function to replicate the number:
replicate' a b
| a ==0 = []
| a>0 = b:replicate(a-1) b
Note that normal, library replicate function is used inside, not the replicate' one. It should be similar to compareAndIncrease, because it uses comparison, numerical operations and returns a list, so I thought it would work like this:
replicate' :: (Ord a, Num a) => a -> a -> [a]
However, when I checked with :t, I got this result:
replicate' :: Int -> t -> [t]
I continued fiddling with this function and changed it's name to repval, so now it is:
Could anyone explain to me what is happening?
GHCi is a great tool to use here:
*Main> :type replicate
replicate :: Int -> a -> [a]
You define replicate' in terms of replicate (I rename your variables for clarity):
replicate' n e
| -- blah blah blah
| n > 0 = e : replicate (n - 1) e
Since you call replicate (n - 1), the type checker infers that n - 1 must have type Int, from which it infers that n must have type Int, from which it infers that replicate' has type Int -> a -> [a].
If you wrote your replicate' recursively, using replicate' inside instead of replicate, then you would get
*Main> :type replicate'
replicate' :: (Ord a, Num a) => a -> a1 -> [a1]
Edit
As Ganesh Sittampalam points out, it's best to constrain the type to Integral as it doesn't really make sense to replicate a fractional number of times.
The key flaw in your reasoning is that replicate actually only takes Int for the replication count, rather than a more general numeric type.
If you instead used genericReplicate, then your argument would be roughly valid.
genericReplicate :: Integral i => i -> a -> [a]
However note that the constraint is Integral rather than Num because Num covers any kind of number including real numbers, whereas it only makes sense to repeat something an integer number of times.
As a beginner, I was working on item number 3 of the Haskell 99 problems. The first code snippet is my solution. However, I cannot add the type declaration of [a] -> Int -> a.
It produces an error: Couldn't match type 'a' with 'Int'
--My Solution
--elementAt :: [a] -> Int -> a
elementAt xs num = head . reverse $ takeWhile (<num+1) xs
Meanwhile, one of the solution in the problem 3 works perfectly with the same output.
elementAt' :: [a] -> Int -> a
elementAt' list i = list !! (i - 1)
I've used the :type in the GHCI for the elementAt without the declaration and it shows:
elementAt :: (Num c, Ord c) => [c] -> c -> c
What is the difference of the two functions?
takeWhile (<num+1) xs says that you want to take elements at the front of the list xs while the element (not the index!) is less than num + 1. Since you're comparing num (which is an Int) with the list elements, the compiler infers that the list elements have to be Int as well, which contradicts your type declaration.
Without the type declaration, GHC infers that xs's elements and num have to be the same type (since you compare list elements with num). This type has to have an Ord instance (for the comparison) as well as a Num instance (because of the addition).
The type of takeWhile (< num + 1) is [Int] -> [Int].
However the type of xs is [a].
Essentially you are trying to apply a predicate Int -> Bool to elements of type a.
If you want to filter based on the position in the list, you first need to augment the input list with the position of each element. You can do this fairly easily with zip:
zip xs [1..]
You can then drop elements based on the value of the second element of the resulting list.
I need to create a function of two parameters, an Int and a [Int], that returns a new [Int] with all occurrences of the first parameter removed.
I can create the function easily enough, both with list comprehension and list recursion. However, I do it with these parameters:
deleteAll_list_comp :: Integer -> [Integer] -> [Integer]
deleteAll_list_rec :: (Integer -> Bool) -> [Integer] -> [Integer]
For my assignment, however, my required parameters are
deleteAll_list_comp :: (Eq a) => a -> [a] -> [a]
deleteAll_list_rec :: (Eq a) => a -> [a] -> [a]
I don't know how to read this syntax. As Google has told me, (Eq a) merely explains to Haskell that a is a type that is comparable. However, I don't understand the point of this as all Ints are naturally comparable. How do I go about interpreting and implementing the methods using these parameters? What I mean is, what exactly are the parameters to begin with?
#groovy #pelotom
Thanks, this makes it very clear. I understand now that really it is only asking for two parameters as opposed to three. However, I still am running into a problem with this code.
deleteAll_list_rec :: (Eq a) => a -> [a] -> [a]
delete_list_rec toDelete [] = []
delete_list_rec toDelete (a:as) =
if(toDelete == a) then delete_list_rec toDelete as
else a:(delete_list_rec toDelete as)
This gives me a "The type signature for deleteAll_list_rec
lacks an accompanying binding" which makes no sense to me seeing as how I did bind the requirements properly, didn't I? From my small experience, (a:as) counts as a list while extracting the first element from it. Why does this generate an error but
deleteAll_list_comp :: (Eq a) => a -> [a] -> [a]
deleteAll_list_comp toDelete ls = [x | x <- ls, toDelete==x]
does not?
2/7/13 Update: For all those who might stumble upon this post in the future with the same question, I've found some good information about Haskell in general, and my question specifically, at this link : http://learnyouahaskell.com/types-and-typeclasses
"Interesting. We see a new thing here, the => symbol. Everything before the => symbol is >called a class constraint. We can read the previous type declaration like this: the >equality function takes any two values that are of the same type and returns a Bool. The >type of those two values must be a member of the Eq class (this was the class constraint).
The Eq typeclass provides an interface for testing for equality. Any type where it makes >sense to test for equality between two values of that type should be a member of the Eq >class. All standard Haskell types except for IO (the type for dealing with input and >output) and functions are a part of the Eq typeclass."
One way to think of the parameters could be:
(Eq a) => a -> [a] -> [a]
(Eq a) => means any a's in the function parameters should be members of the
class Eq, which can be evaluated as equal or unequal.*
a -> [a] means the function will have two parameters: (1) an element of
type a, and (2) a list of elements of the same type a (we know that
type a in this case should be a member of class Eq, such as Num or
String).
-> [a] means the function will return a list of elements of the same
type a; and the assignment states that this returned list should
exclude any elements that equal the first function parameter,
toDelete.
(* edited based on pelotom's comment)
What you implemented (rather, what you think you implemented) is a function that works only on lists of Integers, what the assignment wants you to do is create one that works on lists of all types provided they are equality-comparable (so that your function will also work on lists of booleans or strings). You probably don't have to change a lot: Try removing the explicit type signatures from your code and ask ghci about the type that it would infer from your code (:l yourfile.hs and then :t deleteAll_list_comp). Unless you use arithmetic operations or similar things, you will most likely find that your functions already work for all Eq a.
As a simpler example that may explain the concept: Let's say we want to write a function isequal that checks for equality (slightly useless, but hey):
isequal :: Integer -> Integer -> Bool
isequal a b = (a == b)
This is a perfectly fine definition of isequal, but the type constraints that I have manually put on it are way stronger than they have to. In fact, in the absence of the manual type signature, ghci infers:
Prelude> :t isequal
isequal :: Eq a => a -> a -> Bool
which tells us that the function will work for all input types, as long as they are deriving Eq, which means nothing more than having a proper == relation defined on them.
There is still a problem with your _rec function though, since it should do the same thing as your _comp function, the type signatures should match.
I am trying to revise for my functional programming exam, and am stumped on the first questions on past papers, and yes, we arent allowed solution sheets, here is an example of the 1st question on a past paper.
For each of the following expressions give its type in Haskell (for an expression that has many types, just give one type).
(True, "hello", 42)
[42, 4, 2]
length [True]
filter even
I think personally that the answer for one and two would be a tuple of bool, String and int and a list of ints respectively, is this correct to assume? and secondly how would you answer 3 and 4, i am sure length True just outputs a list of all elements that are of that length, and that filter even just alters a list of ints to a list that are of all even numbers, though how could i show this as an answer?
If you want to get types of variables offline with ghci you have to type
:t expression
if you want to create variables in ghci, you have to use let without using 'in' (as in do notation for monads, I don't know if you have seen them yet) :
let var = expr
If you check it all by yourself, you should be able to remember it more easily for your exams. (good luck for it ;))
length [True] will be Int, and it would return 1. You can check that with ghci or lambdabot.
filter even will be (Integral a) => [a] -> [a]
for example, [Int] -> [Int]
And I think this is kind of pointless because lambdabot can tell all those things to you.
To be precise, the type of [42, 4, 2] is going to be
Num a => [a]
This is because an integer literal in Haskell is treated as having an implicit "fromIntegral" in front of it, so the real expression is [fromIntegral 42, fromIntegral 4, fromIntegral 2].
"fromIntegral" is part of the Num class, and has the type
fromIntegral :: (Integral a, Num b) => a -> b
This says that it converts an instance of some Integral type (i.e. Int or Integer) into an arbitrary other numeric type (Int, Float, Double, Complex ...). This is why you can say something like "43.2 + 1" without getting a type error.
"length [True]" is going to have type Int, because "length" has type "[a] -> Int", and the argument (a list of Bool) is provided.
"filter even" is a little bit more complicated. Start with the type of "filter":
filter :: (a -> Bool) -> [a] -> [a]
The first parameter (the bit in brackets) is itself a function that takes a list item and returns a Bool. Remember that the "->" operator in Haskell types is right associative, so if you put in the implied brackets you see that the type is:
filter :: (a -> Bool) -> ([a] -> [a])
In other words if you give it the first argument, you get back a new function that expects the second argument. In this case the first argument is:
even :: (Integral a) => a -> Bool
This introduces a slight wrinkle: "even" requires its argument to be an Integral type (i.e. Int or Integer, as above), so this constraint has to be propagated to the result. If it were not then you could write this:
filter even "foo"
Hence the answer is:
filter even :: (Integral a) => [a] -> [a]
You can see that the Integral constraint comes from the type of "even", while the rest of the type comes from "filter".