How should one reason about function evaluation in examples like the following in Haskell:
let f x = ...
x = ...
in map (g (f x)) xs
In GHC, sometimes (f x) is evaluated only once, and sometimes once for each element in xs, depending on what exactly f and g are. This can be important when f x is an expensive computation. It has just tripped a Haskell beginner I was helping and I didn't know what to tell him other than that it is up to the compiler. Is there a better story?
Update
In the following example (f x) will be evaluated 4 times:
let f x = trace "!" $ zip x x
x = "abc"
in map (\i -> lookup i (f x)) "abcd"
With language extensions, we can create situations where f x must be evaluated repeatedly:
{-# LANGUAGE GADTs, Rank2Types #-}
module MultiEvG where
data BI where
B :: (Bounded b, Integral b) => b -> BI
foo :: [BI] -> [Integer]
foo xs = let f :: (Integral c, Bounded c) => c -> c
f x = maxBound - x
g :: (forall a. (Integral a, Bounded a) => a) -> BI -> Integer
g m (B y) = toInteger (m + y)
x :: (Integral i) => i
x = 3
in map (g (f x)) xs
The crux is to have f x polymorphic even as the argument of g, and we must create a situation where the type(s) at which it is needed can't be predicted (my first stab used an Either a b instead of BI, but when optimising, that of course led to only two evaluations of f x at most).
A polymorphic expression must be evaluated at least once for each type it is used at. That's one reason for the monomorphism restriction. However, when the range of types it can be needed at is restricted, it is possible to memoise the values at each type, and in some circumstances GHC does that (needs optimising, and I expect the number of types involved mustn't be too large). Here we confront it with what is basically an inhomogeneous list, so in each invocation of g (f x), it can be needed at an arbitrary type satisfying the constraints, so the computation cannot be lifted outside the map (technically, the compiler could still build a cache of the values at each used type, so it would be evaluated only once per type, but GHC doesn't, in all likelihood it wouldn't be worth the trouble).
Monomorphic expressions need only be evaluated once, they can be shared. Whether they are is up to the implementation; by purity, it doesn't change the semantics of the programme. If the expression is bound to a name, in practice you can rely on it being shared, since it's easy and obviously what the programmer wants. If it isn't bound to a name, it's a question of optimisation. With the bytecode generator or without optimisations, the expression will often be evaluated repeatedly, but with optimisations repeated evaluation would indicate a compiler bug.
Polymorphic expressions must be evaluated at least once for every type they're used at, but with optimisations, when GHC can see that it may be used multiple times at the same type, it will (usually) still be shared for that type during a larger computation.
Bottom line: Always compile with optimisations, help the compiler by binding expressions you want shared to a name, and give monomorphic type signatures where possible.
Your examples are indeed quite different.
In the first example, the argument to map is g (f x) and is passed once to map most likely as partially applied function.
Should g (f x), when applied to an argument within map evaluate its first argument, then this will be done only once and then the thunk (f x) will be updated with the result.
Hence, in your first example, f xwill be evaluated at most 1 time.
Your second example requires a deeper analysis before the compiler can arrive at the conclusion that (f x) is always constant in the lambda expression. Perhaps it will never optimize it at all, because it may have knowledge that trace is not quite kosher. So, this may evaluate 4 times when tracing, and 4 times or 1 time when not tracing.
This is really dependent on GHC's optimizations, as you've been able to tell.
The best thing to do is to study the GHC core that you get after optimizing the program. I would look at the generated Core and examine whether f x had its own let statement outside the map or not.
If you want to be sure, then you should factor f x out into its own variable assigned in a let, but there's not really a guaranteed way to figure it out other than reading through Core.
All that said, with the exception of things like trace that use unsafePerformIO, this will never change the semantics of your program: how it actually behaves.
In GHC without optimizations, the body of a function is evaluated every time the function is called. (A "call" means the function is applied to arguments and the result is evaluated.) In the following example, f x is inside a function, so it will execute each time the function is called.
(GHC may optimize this expression as discussed in the FAQ [1].)
let f x = trace "!" $ zip x x
x = "abc"
in map (\i -> lookup i (f x)) "abcd"
However, if we move f x out of the function, it will execute only once.
let f x = trace "!" $ zip x x
x = "abc"
in map ((\f_x i -> lookup i f_x) (f x)) "abcd"
This can be rewritten more readably as
let f x = trace "!" $ zip x x
x = "abc"
g f_x i = lookup i f_x
in map (g (f x)) "abcd"
The general rule is that, each time a function is applied to an argument, a new "copy" of the function body is created. Function application is the only thing that may cause an expression to re-execute. However, be warned that some functions and function calls do not look like functions syntactically.
[1] http://www.haskell.org/haskellwiki/GHC/FAQ#Subexpression_Elimination
Related
I'm using QuickCheck to generate arbitrary functions, and I'd like to generate arbitrary injective functions (i.e. f a == f b if and only if a == b).
I thought I had it figured out:
newtype Injective = Injective (Fun Word Char) deriving Show
instance Arbitrary Injective where
arbitrary = fmap Injective fun
where
fun :: Gen (Fun Word Char)
fun = do
a <- arbitrary
b <- arbitrary
arbitrary `suchThat` \(Fn f) ->
(f a /= f b) || (a == b)
But I'm seeing cases where the generated function maps distinct inputs to the same output.
What I want:
f such that for all inputs a and b, either f a does not equal f b or a equals b.
What I think I have:
f such that there exist inputs a and b where either f a does not equal f b or a equals b.
How can I fix this?
You've correctly identified the problem: what you're generating is functions with the property ∃ a≠b. f a≠f b (which is readily true for most random functions anyway), whereas what you want is ∀ a≠b. f a≠f b. That is a much more difficult property to ensure, because you need to know about all the other function values for generating each individual one.
I don't think this is possible to ensure for general input types, however for word specifically what you can do is “fake” a function by precomputing all the output values sequentially, making sure that you don't repeat one that has already been done, and then just reading off from that predetermined chart. It requires a bit of laziness fu to actually get this working:
import qualified Data.Set as Set
newtype Injective = Injective ([Char] {- simply a list without duplicates -})
deriving Show
instance Arbitrary Injective where
arbitrary = Injective . lazyNub <$> arbitrary
lazyNub :: Ord a => [a] -> [a]
lazyNub = go Set.empty
where go _ [] = []
go forbidden (x:xs)
| x `Set.member` forbidden = go forbidden xs
| otherwise = x : go (Set.insert x forbidden) xs
This is not very efficient, and may well not be ok for your application, but it's probably the best you can do.
In practice, to actually use Injective as a function, you'll want to wrap the values in a suitable structure that has only O (log n) lookup time. Unfortunately, Data.Map.Lazy is not lazy enough, you may need to hand-bake something like a list of exponentially-growing maps.
There's also the concern that for some insufficiently big result types, it is just not possible to generate injective functions because there aren't enough values available. In fact as Joseph remarked, this is the case here. The lazyNub function will go into an infinite loop in this case. I'd say for a QuickCheck this is probably ok though.
In Haskell functions always take one parameter. Multiple parameters are implemented via Currying. That being the case, I can see how a function of two parameters would be defined as "func1" below. It's a function that returns a function (closure) that adds the outer function's single parameter to the returned function's single parameter.
However, although this is how curried functions work, that's not the regular Haskell syntax for defining a two-parameter function. Instead we're taught to define such a function like "func2".
I'd like to know how Haskell understands that func2 should behave the same way as func1. There's nothing about the definition of func2 that suggest to me that it is a function that returns a function. To the contrary it actually looks like a two-parameter function, something we're told doesn't exist!
What's the trick here? Is Haskell just born knowing that we can define multi-parameter functions in this textbook way, and that they work the way we expect anyhow? That is, is this a syntax convention that doesn't seem to be clearly documented (Haskell knows what you mean and will supply the missing function return for you), or is there some other magic at work or something I'm missing?
func1 :: Int -> (Int -> Int)
func1 x = (\y -> x + y)
func2 :: Int -> Int -> Int
func2 x y = x + y
main = do
print (func1 7 9)
print (func2 7 9)
In the language itself, writing a function definition of the form f x y z = _ is equivalent to f = \x y z -> _, which is equivalent to f = \x -> \y -> \z -> _. There's no theoretical reason for this; it's just that those nested lambda abstractions are a terrible eye-/finger-sore and everyone thought that it would be fine to sacrifice a bit of pedantry to make some syntax sugar for it. That's all there is on the surface and is probably all you need to know, for now.
In the implementation of the language, though, things get trickier. In GHC, which is the most common implementation, there actually is a difference between f x y = _ and f = \x -> \y -> _. When GHC compiles Haskell, it assigns arity to declarations. The former definition of f has arity 2, and the latter has arity 0. Take (.) from GHC.Base
(.) f g = \x -> f (g x)
(.) has arity 2, even though its type ((b -> c) -> (a -> b) -> a -> c) says that it can be applied up to thrice. This affects optimization: GHC will only inline a function that is saturated, or has at least as many arguments applied as its arity. In the call (maximum .), (.) will not inline, because it only has one argument (it is unsaturated). In the call (maximum . f), it will inline to \x -> maximum (f x), and in (maximum . f) 1, the (.) will inline first to a lambda abstraction (producing (\x -> maximum (f x)) 1), which will beta-reduce to maximum (f 1). If (.) were implemented
(.) f g x = f (g x)
(.) would have arity 3, which means it would inline less often (specifically the f . g case, which is a very common argument to higher order functions), likely reducing performance, which is exactly what the comment on it says:
Make sure it has TWO args only on the left, so that it inlines
when applied to two functions, even if there is no final argument
Final answer: the two forms should be equivalent, according to the language's semantics, but in GHC the two forms have different characteristics when it comes to optimization, even if they always give the same result.
When talking about type signatures, there is no such thing as a "multi-parameter function". All functions are single-parameter, period. Haskell doesn't need to somehow "translate" multi-parameter functions into single-parameter ones, because the former doesn't exist at all.
All function type signatures look like a -> b, where a is argument type and b is return type. Sometimes b may just happen to contain more arrows ->, in which case we, humans (but not the compiler), may say that the function has multiple parameters.
When talking about the syntax for implementations, i.e. f x y = z - that is merely syntactic sugar, which gets desugared (i.e. mechanically transformed) into f = \x -> \y -> z during compilation.
In Haskell, some lists are cyclic:
ones = 1 : ones
Others are not:
nums = [1..]
And then there are things like this:
more_ones = f 1 where f x = x : f x
This denotes the same value as ones, and certainly that value is a repeating sequence. But whether it's represented in memory as a cyclic data structure is doubtful. (An implementation could do so, but this answer explains that "it's unlikely that this will happen in practice".)
Suppose we take a Haskell implementation and hack into it a built-in function isCycle :: [a] -> Bool that examines the structure of the in-memory representation of the argument. It returns True if the list is physically cyclic and False if the argument is of finite length. Otherwise, it will fail to terminate. (I imagine "hacking it in" because it's impossible to write that function in Haskell.)
Would the existence of this function break any interesting properties of the language?
Would the existence of this function break any interesting properties of the language?
Yes it would. It would break referential transparency (see also the Wikipedia article). A Haskell expression can be always replaced by its value. In other words, it depends only on the passed arguments and nothing else. If we had
isCycle :: [a] -> Bool
as you propose, expressions using it would not satisfy this property any more. They could depend on the internal memory representation of values. In consequence, other laws would be violated. For example the identity law for Functor
fmap id === id
would not hold any more: You'd be able to distinguish between ones and fmap id ones, as the latter would be acyclic. And compiler optimizations such as applying the above law would not longer preserve program properties.
However another question would be having function
isCycleIO :: [a] -> IO Bool
as IO actions are allowed to examine and change anything.
A pure solution could be to have a data type that internally distinguishes the two:
import qualified Data.Foldable as F
data SmartList a = Cyclic [a] | Acyclic [a]
instance Functor SmartList where
fmap f (Cyclic xs) = Cyclic (map f xs)
fmap f (Acyclic xs) = Acyclic (map f xs)
instance F.Foldable SmartList where
foldr f z (Acyclic xs) = F.foldr f z xs
foldr f _ (Cyclic xs) = let r = F.foldr f r xs in r
Of course it wouldn't be able to recognize if a generic list is cyclic or not, but for many operations it'd be possible to preserve the knowledge of having Cyclic values.
In the general case, no you can't identify a cyclic list. However if the list is being generated by an unfold operation then you can. Data.List contains this:
unfoldr :: (b -> Maybe (a, b)) -> b -> [a]
The first argument is a function that takes a "state" argument of type "b" and may return an element of the list and a new state. The second argument is the initial state. "Nothing" means the list ends.
If the state ever recurs then the list will repeat from the point of the last state. So if we instead use a different unfold function that returns a list of (a, b) pairs we can inspect the state corresponding to each element. If the same state is seen twice then the list is cyclic. Of course this assumes that the state is an instance of Eq or something.
While writing some lambda functions in Haskell, I was originally writing the functions like:
tru = \t f -> t
fls = \t f -> f
However, I soon noticed from the examples online that such functions are frequently written like:
tru = \t -> \f -> t
fls = \t -> \f -> f
Specifically, each of the items passed to the function have their own \ and -> as opposed to above. When checking the types of these they appear to be the same. My question is, are they equivalent or do they actually differ in some way? And not only for these two functions, but does it make a difference for functions in general? Thank you much!
They're the same, Haskell automatically curries things to keep things syntax nice. The following are all equivalent**
foo a b = (a, b)
foo a = \b -> (a, b)
foo = \a b -> (a, b)
foo = \a -> \b -> (a, b)
-- Or we can simply eta convert leaving
foo = (,)
If you want to be idiomatic, prefer either the first or the last. Introducing unnecessary lambdas is good for teaching currying, but in real code just adds syntactic clutter.
However in raw lambda calculus (not Haskell) most manually curry with
\a -> \b -> a b
Because people don't write a lot of lambda calculus by hand and when they do they tend to stick unsugared lambda calculus to keep things simple.
** modulo the monomorphism restriction, which won't impact you anyways with a type signature.
Though, as jozefg said, they are themselves equivalent, they may lead to different execution behaviour when combined with local variable bindings. Consider
f, f' :: Int -> Int -> Int
with the two definitions
f a x = μ*x
where μ = sum [1..a]
and
f' a = \x -> μ*x
where μ = sum [1..a]
These sure look equivalent, and certainly will always yield the same results.
GHCi, version 7.6.2: http://www.haskell.org/ghc/ :? for help
...
[1 of 1] Compiling Main ( def0.hs, interpreted )
Ok, modules loaded: Main.
*Main> sum $ map (f 10000) [1..10000]
2500500025000000
*Main> sum $ map (f' 10000) [1..10000]
2500500025000000
However, if you try this yourself, you'll notice that with f takes quite a lot of time whereas with f' you get the result immediately. The reason is that f' is written in a form that prompts GHC to compile it so that actually f' 10000 is evaluated before starting to map it over the list. In that step, the value μ is calculated and stored in the closure of (f' 10000). On the other hand, f is treated simply as "one function of two variables"; (f 10000) is merely stored as a closure containing the parameter 10000 and μ is not calculated at all at first. Only when map applies (f 10000) to each element in the list, the whole sum [1..a] is calculated, which takes some time for each element in [1..10000]. With f', this was not necessary because μ was pre-calculated.
In principle, common-subexpression elimination is an optimisation that GHC is able to do itself, so you might at times get good performance even with a definition like f. But you can't really count on it.
My application multiplies vectors after a (costly) conversion using an FFT. As a result, when I write
f :: (Num a) => a -> [a] -> [a]
f c xs = map (c*) xs
I only want to compute the FFT of c once, rather than for every element of xs. There really isn't any need to store the FFT of c for the entire program, just in the local scope.
I attempted to define my Num instance like:
data Foo = Scalar c
| Vec Bool v -- the bool indicates which domain v is in
instance Num Foo where
(*) (Scalar c) = \x -> case x of
Scalar d -> Scalar (c*d)
Vec b v-> Vec b $ map (c*) v
(*) v1 = let Vec True v = fft v1
in \x -> case x of
Scalar d -> Vec True $ map (c*) v
v2 -> Vec True $ zipWith (*) v (fft v2)
Then, in an application, I call a function similar to f (which works on arbitrary Nums) where c=Vec False v, and I expected that this would be just as fast as if I hack f to:
g :: Foo -> [Foo] -> [Foo]
g c xs = let c' = fft c
in map (c'*) xs
The function g makes the memoization of fft c occur, and is much faster than calling f (no matter how I define (*)). I don't understand what is going wrong with f. Is it my definition of (*) in the Num instance? Does it have something to do with f working over all Nums, and GHC therefore being unable to figure out how to partially compute (*)?
Note: I checked the core output for my Num instance, and (*) is indeed represented as nested lambdas with the FFT conversion in the top level lambda. So it looks like this is at least capable of being memoized. I have also tried both judicious and reckless use of bang patterns to attempt to force evaluation to no effect.
As a side note, even if I can figure out how to make (*) memoize its first argument, there is still another problem with how it is defined: A programmer wanting to use the Foo data type has to know about this memoization capability. If she wrote
map (*c) xs
no memoization would occur. (It must be written as (map (c*) xs)) Now that I think about it, I'm not entirely sure how GHC would rewrite the (*c) version since I have curried (*). But I did a quick test to verify that both (*c) and (c*) work as expected: (c*) makes c the first arg to *, while (*c) makes c the second arg to *. So the problem is that it is not obvious how one should write the multiplication to ensure memoization. Is this just an inherent downside to the infix notation (and the implicit assumption that the arguments to * are symmetric)?
The second, less pressing issue is that the case where we map (v*) onto a list of scalars. In this case, (hopefully) the fft of v would be computed and stored, even though it is unnecessary since the other multiplicand is a scalar. Is there any way around this?
Thanks
I believe stable-memo package could solve your problem. It memoizes values not using equality but by reference identity:
Whereas most memo combinators memoize based on equality, stable-memo does it based on whether the exact same argument has been passed to the function before (that is, is the same argument in memory).
And it automatically drops memoized values when their keys are garbage collected:
stable-memo doesn't retain the keys it has seen so far, which allows them to be garbage collected if they will no longer be used. Finalizers are put in place to remove the corresponding entries from the memo table if this happens.
So if you define something like
fft = memo fft'
where fft' = ... -- your old definition
you'll get pretty much what you need: Calling map (c *) xs will memoize the computation of fft inside the first call to (*) and it gets reused on subsequent calls to (c *). And if c is garbage collected, so is fft' c.
See also this answer to How to add fields that only cache something to ADT?
I can see two problems that might prevent memoization:
First, f has an overloaded type and works for all Num instances. So f cannot use memoization unless it is either specialized (which usually requires a SPECIALIZE pragma) or inlined (which may happen automatically, but is more reliable with an INLINE pragma).
Second, the definition of (*) for Foo performs pattern matching on the first argument, but f multiplies with an unknown c. So within f, even if specialized, no memoization can occur. Once again, it very much depends on f being inlined, and a concrete argument for c to be supplied, so that inlining can actually appear.
So I think it'd help to see how exactly you're calling f. Note that if f is defined using two arguments, it has to be given two arguments, otherwise it cannot be inlined. It would furthermore help to see the actual definition of Foo, as the one you are giving mentions c and v which aren't in scope.