I'm trying to write a computational workflow which would allow a computation which can produce side effects like log or sleep and a return value
A usage example would be something like this
let add x y =
compute {
do! log (sprintf "add: x = %d, y= %d" x y)
do! sleep 1000
let r = x + y
do! log (sprintf "add: result= %d" r)
return r
}
...
let result = run (add 100 1000)
and I would like the side effects to be produced when executeComputation is called.
My attempt is
type Effect =
| Log of string
| Sleep of int
type Computation<'t> = Computation of Lazy<'t * Effect list>
let private bind (u : 'u, effs : Effect list)
(f : 'u -> 'v * Effect list)
: ('v * Effect list) =
let v, newEffs = f u
let allEffects = List.append effs newEffs
v, allEffects
type ComputeBuilder() =
member this.Zero() = lazy ((), [])
member this.Return(x) = x, []
member this.ReturnFrom(Computation f) = f.Force()
member this.Bind(x, f) = bind x f
member this.Delay(funcToDelay) = funcToDelay
member this.Run(funcToRun) = Computation (lazy funcToRun())
let compute = new ComputeBuilder()
let log msg = (), [Log msg]
let sleep ms = (), [Sleep ms]
let run (Computation x) = x.Force()
...but the compiler complains about the let! lines in the following code:
let x =
compute {
let! a = add 10 20
let! b = add 11 2000
return a + b
}
Error FS0001: This expression was expected to have type
'a * Effect list
but here has type
Computation<'b> (FS0001)
Any suggestions?
The main thing that is not right with your definition is that some of the members of the computation builder use your Computation<'T> type and some of the other members use directly a pair of value and list of effects.
To make it type check, you need to be consistent. The following version uses Computation<'T> everywhere - have a look at the type signature of Bind, for example:
let private bind (Computation c) f : Computation<_> =
Computation(Lazy.Create(fun () ->
let u, effs = c.Value
let (Computation(c2)) = f u
let v, newEffs = c2.Value
let allEffects = List.append effs newEffs
v, allEffects))
type ComputeBuilder() =
member this.Zero() = Computation(lazy ((), []))
member this.Return(x) = Computation(lazy (x, []))
member this.ReturnFrom(c) = c
member this.Bind(x, f) = bind x f
member this.Delay(funcToDelay:_ -> Computation<_>) =
Computation(Lazy.Create(fun () ->
let (Computation(r)) = funcToDelay()
r.Value))
Related
Consider a generic container like:
type Foo<'t> =
{
Foo : 't
}
This generic function works OK:
module Foo =
let inline emptyFoo () =
{
Foo = LanguagePrimitives.GenericZero
}
But this value does not:
module Foo =
let emptyFoo =
{
Foo = LanguagePrimitives.GenericZero
}
This is because the compiler infers emptyFoo to have type Foo<obj>.
However, the standard library has generic values like List.empty, so how is that achieved there?
You have to make it explicitly generic.
List.empty is implemented like this:
let empty<'T> = ([ ] : 'T list)
You could implement a List by yourself, with an empty. It looks like this.
type L<'a> =
| Null
| Cons of 'a * L<'a>
module L =
let empty = Null
let xs = Cons(1, Cons(2, L.empty))
let ys = Cons(1.0,Cons(2.0,L.empty))
So, why does L.empty in this case works in a generic way? Because the value Null has no special value attached to it. You could say, it is compatible with every other generic.
In your Record on the other hand, you always must produce a value. LanguagePrimitive.GenericZero is not some Generic value. It help so to resolve to a special zero value, and this value is determined by the other code you write.
For example
let x = LanguagePrimitives.GenericZero
is also obj
let x = LanguagePrimitives.GenericZero + 1
will be int. And
let x = LanguagePrimitives.GenericZero + 1.0
will be float. So in some case you can think of GenericZero just as a placeholder for zero for the special type you need, but the code needs to determine at this point, which type you want.
You could change your type, with an option to provide a real empty.
type Foo<'t> = {
Foo: 't option
}
module Foo =
let empty = { Foo = None }
let x = Foo.empty // Foo<'a>
let y = { x with Foo = Some 1 } // Foo<int>
let z = { x with Foo = Some 1.0 } // Foo<float>
Zero Member
Maybe you want a Zero Member on some types. For example
type Vector3 = {X:float; Y:float; Z:float} with
static member create x y z = {X=x; Y=y; Z=z}
static member (+) (a,b) = Vector3.create (a.X + b.X) (a.Y + b.Y) (a.Z + b.Z)
static member Zero = Vector3.create 0.0 0.0 0.0
static member DivideByInt(a,b) =
Vector3.create
(LanguagePrimitives.DivideByInt a.X b)
(LanguagePrimitives.DivideByInt a.Y b)
(LanguagePrimitives.DivideByInt a.Z b)
then you can write
let xs = [1;2;3]
let ys = [Vector3.create 1.0 1.0 1.0; Vector3.create 1.0 1.0 1.0]
let inline sum xs =
List.fold (fun a b -> a + b) LanguagePrimitives.GenericZero xs
let sumx = sum xs // int: 6
let sumy = sum ys // Vector3: 2.0 2.0 2.0
type Identity<'T> = Identity of 'T
type IdentityBuilder() =
member __.Bind (Identity x) (k : 'a -> Identity<'b>) = k x
member __.Return x = Identity x
let identity = new IdentityBuilder()
let three = Identity 3
let four = Identity 4
let twelve =
identity.Bind three <| fun t ->
identity.Bind four <| fun f ->
identity.Return (t * f)
let twelve2 = identity {
let! t = three
let! f = four
return t * f
}
twelve does not introduce any problem, but twelve2 gives
FS0001: This expression was expected to have type
'Identity<'a>' but here has type
''b * 'c'
on the line let! t = three.
I thought twelve and twelve2 should be equivalent... Was I mistaken?
As noted in the comment by Szer, you need to use tupled parameters for the Computation Builder methods. However, it is often convenient to use the curried versions for pipelining, as in your example. Therefore, what I usually do is create a module that contains all the functions required for the Computation Builder in curried form, and then use them in the builder itself. That way I can use either the computation expression syntax or the pipelining syntax depending on the scenario.
In your case, that would look something like this:
type Identity<'T> = Identity of 'T
[<CompilationRepresentation(CompilationRepresentationFlags.ModuleSuffix)>]
module Identity =
let bind (f: 'T -> Identity<'U>) (Identity x) = f x
let create x = Identity x
type IdentityBuilder() =
member __.Bind (x, f) = Identity.bind f x
member __.Return x = Identity.create x
let identity = new IdentityBuilder()
let three = Identity 3
let four = Identity 4
let twelve =
three |> Identity.bind (fun t ->
four |> Identity.bind (fun f ->
Identity.create (t * f)))
let twelve2 = identity {
let! t = three
let! f = four
return t * f
}
Another common practice is to define an operator >>= for the bind function so that you can streamline the syntax even more:
let (>>=) f x = Identity.bind x f
let twelve3 = three >>= (fun t -> four >>= (fun f -> Identity.create (t * f)))
The following code sample is from Scott Wlaschin's site F# for fun and profit.
type LoggingBuilder() =
let log p = printfn "expression is %A" p
member this.Bind(x, f) =
log x
f x
member this.Return(x) =
x
let logger = new LoggingBuilder()
let loggedWorkflow =
logger
{
let! x = 42
let! y = 43
let! z = x + y
return z
}
Is there a way to inject a function instead of printfn into the LoggingBuilder()?
You can just add a parameter to the builder type:
type LoggingBuilder(lf: obj -> unit) =
let log p = lf p
member this.Bind(x, f) =
log x
f x
member this.Return(x) =
x
let logger = new LoggingBuilder(printfn "expression is %A")
You could make the builder generic if you want to make the input type more specific than obj e.g.
type LoggingBuilder<'a>(lf: 'a -> unit) =
...
let logger = new LoggingBuilder<int>(printfn "Got %i")
If your intention is to replace printfn with a logger such as NLog, you can use Printf.ksprintf.
open NLog
open NLog.Config
open NLog.Targets
let private logger =
let config = new LoggingConfiguration()
let consoleTarget = new ColoredConsoleTarget()
config.AddTarget("console", consoleTarget)
consoleTarget.Layout <- Layouts.SimpleLayout.FromString
#"${longdate}|${level:uppercase=true}|${logger}|${message}"
let rule = new LoggingRule("*", LogLevel.Debug, consoleTarget)
config.LoggingRules.Add rule
LogManager.Configuration <- config
LogManager.GetLogger "MyLogger"
type LoggingBuilder(lf: string -> unit) =
let log format =
let doAfter (s: string) = lf s
Printf.ksprintf doAfter format
member this.Bind(x, f) =
log "%A" x
f x
member this.Return(x) =
x
let logger = new LoggingBuilder(logger.Info)
(Having failed to 'grok' FParsec, I followed the advice I read somewhere and started trying to write a little parser myself. Somehow I spotted what looked like a chance to try and monadify it, and now I have N problems...)
This is my 'Result' type (simplified)
type Result<'a> =
| Success of 'a
| Failure of string
Here's the computation expression builder
type ResultBuilder() =
member m.Return a = Success(a)
member m.Bind(r,fn) =
match r with
| Success(a) -> fn a
| Failure(m) -> Failure(m)
In this first example, everything works (compiles) as expected:
module Parser =
let res = ResultBuilder()
let Combine p1 p2 fn =
fun a -> res { let! x = p1 a
let! y = p2 a
return fn(x,y) }
My problem is here: I'd like to be able to catch any failure in the 'combining' function and return a failure, but it says that I should define a 'Zero'.
let Combine2 p1 p2 fn =
fun a -> res { let! x = p1 a
let! y = p2 a
try
return fn(x,y)
with
| ex -> Failure(ex.Message) }
Having no idea what I should return in a Zero, I just threw in member m.Zero() = Failure("hello world"), and it now says I need TryWith.
So:
member m.TryWith(r,fn) =
try
r()
with
| ex -> fn ex
And now it wants Delay, so member m.Delay f = (fun () -> f()).
At which point it says (on the ex -> Failure), This expression should have type 'unit', but has type 'Result<'a>', and I throw up my arms and turn to you guys...
Link for playing: http://dotnetfiddle.net/Ho1sGS
The with block should also return a result from the computation expression. Since you want to return Result.Failure you need to define the member m.ReturnFrom a = a and use it to return the Failure from the with block. In the try block you should also specify that fn returns Success if it doesn't throw.
let Combine2 p1 p2 fn =
fun a -> res { let! x = p1 a
let! y = p2 a
return!
try
Success(fn(x,y))
with
| ex -> Failure(ex.Message)
}
Update:
The original implementation was showing a warning, not an error. The expression in the with block was not used since you returned from the try block, so you could simply add |> ignore. In that case if fn throws then the return value is m.Zero() and the only difference is that you would get "hello world" instead of ex.Message. Illustrated with an example below. Full script here: http://dotnetfiddle.net/mFbeZg
Original implementation with |> ignore to mute the warning:
let Combine3 p1 p2 fn =
fun a -> res { let! x = p1 a
let! y = p2 a
try
return fn(x,y)
with
| ex -> Failure(ex.Message) |> ignore // no warning
}
Run it:
let comb2 a =
let p1' x = Success(x)
let p2' y = Success(y)
let fn' (x,y) = 1/0 // div by zero
let func = Parser.Combine2 p1' p2' fn' a
func()
let comb3 a =
let p1' x = Success(x)
let p2' y = Success(y)
let fn' (x,y) = 1/0 // div by zero
let func = Parser.Combine3 p1' p2' fn' a
func()
let test2 = comb2 1
let test3 = comb3 1
Result:
val test2 : Result<int> = Failure "Attempted to divide by zero."
val test3 : Result<int> = Failure "hello world"
If you want to support try ... with inside a computation builder, you need to add TryWith (as you tried) as well as a few other members including Delay and Run (depending on how you want to implement Delay). In order to be able to return failure, you also need to support return! by adding ReturnFrom:
type ResultBuilder() =
member m.Return a = Success(a)
member m.Bind(r,fn) =
match r with
| Success(a) -> fn a
| Failure(m) -> Failure(m)
member m.TryWith(r,fn) =
try r() with ex -> fn ex
member m.Delay(f) = f
member m.Run(f) = f()
member m.ReturnFrom(r) = r
Now you can do the following:
let Combine2 p1 p2 fn = fun a -> res {
let! x = p1 a
let! y = p2 a
try
return fn(x,y)
with ex ->
return! Failure(ex.Message) }
The trick is that the normal branch uses just return (representing success), but the exception handler uses return! to return an explicitly created result using Failure.
That said, if you are interested in parsers, then you need to use a different type - what you are describing here is more like the option (or Maybe) monad. To implement parser combinators you need a type that represents a parser rather than result of a parser. See for example this article.
I have written a function that takes an array as input and returns an array of equal size as output. For example:
myFunc [| "apple"; "orange"; "banana" |]
> val it : (string * string) [] =
[|("red", "sphere"); ("orange", "sphere"); ("yellow", "oblong")|]
Now I want to assign the results via a let binding. For example:
let [|
( appleColor, appleShape );
( orangeColor, orangeShape );
( bananaColor, bananaShape )
|] =
myFunc [| "apple"; "orange"; "banana" |]
Which works great...
> val orangeShape : string = "sphere"
> val orangeColor : string = "orange"
> val bananaShape : string = "oblong"
> val bananaColor : string = "yellow"
> val appleShape : string = "sphere"
> val appleColor : string = "red"
...except it produces a warning:
warning FS0025: Incomplete pattern matches on this expression. For example, the value '[|_; _; _; _|]' may indicate a case not covered by the pattern(s).
The source and reason for the warning has already been covered, I'm just looking for a succinct work-around. This function call occurs near the top of my function, and I don't like the idea of putting the entire function body inside a match:
let otherFunc =
match myFunc [| "apple"; "orange"; "banana" |] with
| [|
( appleColor, appleShape );
( orangeColor, orangeShape );
( bananaColor, bananaShape )
|] ->
// ... the rest of my function logic
| _ -> failwith "Something impossible just happened!"
That just smells bad. I don't like the idea of ignoring the warning either - goes against my better judgment. Are there any other options open to me, or do I just need to find a different approach entirely?
One possibility if you expect this kind of calling pattern to be frequent is to make wrappers that act on the sizes of tuples you expect, e.g.
myFunc3 (in1,in2,in3) =
match myFunc [|in1;in2;in3|] with
[|out1;out2;out3|] -> out1, out2, out3
_ -> failwith "Internal error"
etc. But all it does is move the ugly code to a standard place, and writing out the wrappers will be inconvenient.
I don't think there's any better option with this API, because there's no way to tell the compiler that myFunc always returns the same number of elements it is passed.
Another option might be to replace myFunc with an IDisposable class:
type MyClass() =
let expensiveResource = ...
member this.MyFunc(v) = ...calculate something with v using expensiveResource
interface IDisposable with
override this.Dispose() = // cleanup resource
and then use it in a block like
use myClass = new MyClass()
let appleColor, appleShape = myClass.MyFunc(apple)
...
Adapting #Ganesh's answer, here's a primitive way to approach the problem:
let Tuple2Map f (u, v)
= (f u, f v)
let Tuple3Map f (u, v, w)
= (f u, f v, f w)
let Tuple4Map f (u, v, w, x)
= (f u, f v, f w, f x)
Example:
let Square x = x * x
let (a,b) = Tuple2Map Square (4,6)
// Output:
// val b : int = 36
// val a : int = 16
But I guess something even more primitive would be this:
let Square x = x * x
let (a,b) = (Square 4, Square 6)
And if the function name is too long, e.g.
// Really wordy way to assign to (a,b)
let FunctionWithLotsOfInput w x y z = w * x * y * z
let (a,b) =
(FunctionWithLotsOfInput input1 input2 input3 input4A,
FunctionWithLotsOfInput input1 input2 input3 input4B)
We can define temporary function
let FunctionWithLotsOfInput w x y z = w * x * y * z
// Partially applied function, temporary function
let (a,b) =
let f = (FunctionWithLotsOfInput input1 input2 input3)
(f input4A, f input4B)