Learning Scala and functional programming in general. In the following tail-recursive factorial implementation:
def factorialTailRec(n: Int) : Int = {
#tailrec
def factorialRec(n: Int, f: => Int): Int = {
if (n == 0) f else factorialRec(n - 1, n * f)
}
factorialRec(n, 1)
}
I wonder whether there is any benefit to having the second parameter called by value vs called by name (as I have done). In the first case, every stack frame is burdened with a product. In the second case, if my understanding is correct, the entire chain of products will be carried over to the case if ( n== 0) at the nth stack frame, so we will still have to perform the same number of multiplications. Unfortunately, this is not a product of form a^n, which can be calculated in log_2n steps through repeated squaring, but a product of terms that differ by 1 every time. So I can't see any possible way of optimizing the final product: it will still require the multiplication of O(n) terms.
Is this correct? Is call by value equivalent to call by name here, in terms of complexity?
Let me just expand a little bit what you've already been told in comments.
That's how by-name parameters are desugared by the compiler:
#tailrec
def factorialTailRec(n: Int, f: => Int): Int = {
if (n == 0) {
val fEvaluated = f
fEvaluated
} else {
val fEvaluated = f // <-- here we are going deeper into stack.
factorialTailRec(n - 1, n * fEvaluated)
}
}
Through experimentation I found out that with the call by name formalism, the method becomes... non-tail recursive! I made this example code to compare factorial tail-recursively, and factorial non-tail-recursively:
package example
import scala.annotation.tailrec
object Factorial extends App {
val ITERS = 100000
def factorialTailRec(n: Int) : Int = {
#tailrec
def factorialTailRec(n: Int, f: => Int): Int = {
if (n == 0) f else factorialTailRec(n - 1, n * f)
}
factorialTailRec(n, 1)
}
for(i <-1 to ITERS) println("factorialTailRec(" + i + ") = " + factorialTailRec(i))
def factorial(n:Int) : Int = {
if(n == 0) 1 else n * factorial(n-1)
}
for(i <-1 to ITERS) println("factorial(" + i + ") = " + factorial(i))
}
Observe that the inner tailRec function calls the second argument by name. for which the #tailRec annotation still does NOT throw a compile-time error!
I've been playing around with different values for the ITERS variable, and for a value of 100,000, I receive a... StackOverflowError!
(The result of zero is there because of overflow of Int.)
So I went ahead and changed the signature of factorialTailRec/2, to:
def factorialTailRec(n: Int, f: Int): Int
i.e call by value for the argument f. This time, the portion of main that runs factorialTailRec finishes absolutely fine, whereas, of course, factorial/1 crashes at the exact same integer.
Very, very interesting. It seems as if call by name in this situation maintains the stack frames because of the need of computation of the products themselves all the way back to the call chain.
As I understand it, in Scala, a function may be called either
by-value or
by-name
For example, given the following declarations, do we know how the function will be called?
Declaration:
def f (x:Int, y:Int) = x;
Call
f (1,2)
f (23+55,5)
f (12+3, 44*11)
What are the rules please?
The example you have given only uses call-by-value, so I will give a new, simpler, example that shows the difference.
First, let's assume we have a function with a side-effect. This function prints something out and then returns an Int.
def something() = {
println("calling something")
1 // return value
}
Now we are going to define two function that accept Int arguments that are exactly the same except that one takes the argument in a call-by-value style (x: Int) and the other in a call-by-name style (x: => Int).
def callByValue(x: Int) = {
println("x1=" + x)
println("x2=" + x)
}
def callByName(x: => Int) = {
println("x1=" + x)
println("x2=" + x)
}
Now what happens when we call them with our side-effecting function?
scala> callByValue(something())
calling something
x1=1
x2=1
scala> callByName(something())
calling something
x1=1
calling something
x2=1
So you can see that in the call-by-value version, the side-effect of the passed-in function call (something()) only happened once. However, in the call-by-name version, the side-effect happened twice.
This is because call-by-value functions compute the passed-in expression's value before calling the function, thus the same value is accessed every time. Instead, call-by-name functions recompute the passed-in expression's value every time it is accessed.
Here is an example from Martin Odersky:
def test (x:Int, y: Int)= x*x
We want to examine the evaluation strategy and determine which one is faster (less steps) in these conditions:
test (2,3)
call by value: test(2,3) -> 2*2 -> 4
call by name: test(2,3) -> 2*2 -> 4
Here the result is reached with the same number of steps.
test (3+4,8)
call by value: test (7,8) -> 7*7 -> 49
call by name: (3+4) (3+4) -> 7(3+4)-> 7*7 ->49
Here call by value is faster.
test (7,2*4)
call by value: test(7,8) -> 7*7 -> 49
call by name: 7 * 7 -> 49
Here call by name is faster
test (3+4, 2*4)
call by value: test(7,2*4) -> test(7, 8) -> 7*7 -> 49
call by name: (3+4)(3+4) -> 7(3+4) -> 7*7 -> 49
The result is reached within the same steps.
In the case of your example all the parameters will be evaluated before it's called in the function , as you're only defining them by value.
If you want to define your parameters by name you should pass a code block:
def f(x: => Int, y:Int) = x
This way the parameter x will not be evaluated until it's called in the function.
This little post here explains this nicely too.
To iteratate #Ben's point in the above comments, I think it's best to think of "call-by-name" as just syntactic sugar. The parser just wraps the expressions in anonymous functions, so that they can be called at a later point, when they are used.
In effect, instead of defining
def callByName(x: => Int) = {
println("x1=" + x)
println("x2=" + x)
}
and running:
scala> callByName(something())
calling something
x1=1
calling something
x2=1
You could also write:
def callAlsoByName(x: () => Int) = {
println("x1=" + x())
println("x2=" + x())
}
And run it as follows for the same effect:
callAlsoByName(() => {something()})
calling something
x1=1
calling something
x2=1
I will try to explain by a simple use case rather than by just providing an example
Imagine you want to build a "nagger app" that will Nag you every time since time last you got nagged.
Examine the following implementations:
object main {
def main(args: Array[String]) {
def onTime(time: Long) {
while(time != time) println("Time to Nag!")
println("no nags for you!")
}
def onRealtime(time: => Long) {
while(time != time) println("Realtime Nagging executed!")
}
onTime(System.nanoTime())
onRealtime(System.nanoTime())
}
}
In the above implementation the nagger will work only when passing by name
the reason is that, when passing by value it will re-used and therefore the value will not be re-evaluated while when passing by name the value will be re-evaluated every time the variables is accessed
Typically, parameters to functions are by-value parameters; that is, the value of the parameter is determined before it is passed to the function. But what if we need to write a function that accepts as a parameter an expression that we don't want evaluated until it's called within our function? For this circumstance, Scala offers call-by-name parameters.
A call-by-name mechanism passes a code block to the callee and each time the callee accesses the parameter, the code block is executed and the value is calculated.
object Test {
def main(args: Array[String]) {
delayed(time());
}
def time() = {
println("Getting time in nano seconds")
System.nanoTime
}
def delayed( t: => Long ) = {
println("In delayed method")
println("Param: " + t)
t
}
}
1. C:/>scalac Test.scala
2. scala Test
3. In delayed method
4. Getting time in nano seconds
5. Param: 81303808765843
6. Getting time in nano seconds
As i assume, the call-by-value function as discuss above pass just the values to the function. According to Martin Odersky It is a Evaluation strategy follow by a Scala that play the important role in function evaluation. But, Make it simple to call-by-name. its like a pass the function as a argument to the method also know as Higher-Order-Functions. When the method access the value of passed parameter, it call the implementation of passed functions. as Below:
According to #dhg example, create the method first as:
def something() = {
println("calling something")
1 // return value
}
This function contain one println statement and return an integer value. Create the function, who have arguments as a call-by-name:
def callByName(x: => Int) = {
println("x1=" + x)
println("x2=" + x)
}
This function parameter, is define an anonymous function who have return one integer value. In this x contain an definition of function who have 0 passed arguments but return int value and our something function contain same signature. When we call the function, we pass the function as a argument to callByName. But in the case of call-by-value its only pass the integer value to the function. We call the function as below:
scala> callByName(something())
calling something
x1=1
calling something
x2=1
In this our something method called twice, because when we access the value of x in callByName method, its call to the defintion of something method.
Call by value is general use case as explained by many answers here..
Call-by-name passes a code block to the caller and each time the
caller accesses the parameter, the code block is executed and the
value is calculated.
I will try to demonstrate call by name more simple way with use cases below
Example 1:
Simple example/use case of call by name is below function, which takes function as parameter and gives the time elapsed.
/**
* Executes some code block and prints to stdout the
time taken to execute the block
for interactive testing and debugging.
*/
def time[T](f: => T): T = {
val start = System.nanoTime()
val ret = f
val end = System.nanoTime()
println(s"Time taken: ${(end - start) / 1000 / 1000} ms")
ret
}
Example 2:
apache spark (with scala) uses logging using call by name way see Logging trait
in which its lazily evaluates whether log.isInfoEnabled or not from the below method.
protected def logInfo(msg: => String) {
if (log.isInfoEnabled) log.info(msg)
}
In a Call by Value, the value of the expression is pre-computed at the time of the function call and that particular value is passed as the parameter to the corresponding function. The same value will be used all throughout the function.
Whereas in a Call by Name, the expression itself is passed as a parameter to the function and it is only computed inside the function, whenever that particular parameter is called.
The difference between Call by Name and Call by Value in Scala could be better understood with the below example:
Code Snippet
object CallbyExample extends App {
// function definition of call by value
def CallbyValue(x: Long): Unit = {
println("The current system time via CBV: " + x);
println("The current system time via CBV " + x);
}
// function definition of call by name
def CallbyName(x: => Long): Unit = {
println("The current system time via CBN: " + x);
println("The current system time via CBN: " + x);
}
// function call
CallbyValue(System.nanoTime());
println("\n")
CallbyName(System.nanoTime());
}
Output
The current system time via CBV: 1153969332591521
The current system time via CBV 1153969332591521
The current system time via CBN: 1153969336749571
The current system time via CBN: 1153969336856589
In the above code snippet, for the function call CallbyValue(System.nanoTime()), the system nano time is pre-calculated and that pre-calculated value has been passed a parameter to the function call.
But in the CallbyName(System.nanoTime()) function call, the expression "System.nanoTime())" itself is passed as a parameter to the function call and the value of that expression is calculated when that parameter is used inside the function.
Notice the function definition of the CallbyName function, where there is a => symbol separating the parameter x and its datatype. That particular symbol there indicates the function is of call by name type.
In other words, the call by value function arguments are evaluated once before entering the function, but the call by name function arguments are evaluated inside the function only when they are needed.
Hope this helps!
Here is a quick example I coded to help a colleague of mine who is currently taking the Scala course. What I thought was interesting is that Martin didn't use the && question answer presented earlier in the lecture as an example. In any event I hope this helps.
val start = Instant.now().toEpochMilli
val calc = (x: Boolean) => {
Thread.sleep(3000)
x
}
def callByValue(x: Boolean, y: Boolean): Boolean = {
if (!x) x else y
}
def callByName(x: Boolean, y: => Boolean): Boolean = {
if (!x) x else y
}
new Thread(() => {
println("========================")
println("Call by Value " + callByValue(false, calc(true)))
println("Time " + (Instant.now().toEpochMilli - start) + "ms")
println("========================")
}).start()
new Thread(() => {
println("========================")
println("Call by Name " + callByName(false, calc(true)))
println("Time " + (Instant.now().toEpochMilli - start) + "ms")
println("========================")
}).start()
Thread.sleep(5000)
The output of the code will be the following:
========================
Call by Name false
Time 64ms
========================
Call by Value false
Time 3068ms
========================
Parameters are usually pass by value, which means that they'll be evaluated before being substituted in the function body.
You can force a parameter to be call by name by using the double arrow when defining the function.
// first parameter will be call by value, second call by name, using `=>`
def returnOne(x: Int, y: => Int): Int = 1
// to demonstrate the benefits of call by name, create an infinite recursion
def loop(x: Int): Int = loop(x)
// will return one, since `loop(2)` is passed by name so no evaluated
returnOne(2, loop(2))
// will not terminate, since loop(2) will evaluate.
returnOne(loop(2), 2) // -> returnOne(loop(2), 2) -> returnOne(loop(2), 2) -> ...
There are already lots of fantastic answers for this question in Internet. I will write a compilation of several explanations and examples I have gathered about the topic, just in case someone may find it helpful
INTRODUCTION
call-by-value (CBV)
Typically, parameters to functions are call-by-value parameters; that is, the parameters are evaluated left to right to determine their value before the function itself is evaluated
def first(a: Int, b: Int): Int = a
first(3 + 4, 5 + 6) // will be reduced to first(7, 5 + 6), then first(7, 11), and then 7
call-by-name (CBN)
But what if we need to write a function that accepts as a parameter an expression that we don't to evaluate until it's called within our function? For this circumstance, Scala offers call-by-name parameters. Meaning the parameter is passed into the function as it is, and its valuation takes place after substitution
def first1(a: Int, b: => Int): Int = a
first1(3 + 4, 5 + 6) // will be reduced to (3 + 4) and then to 7
A call-by-name mechanism passes a code block to the call and each time the call accesses the parameter, the code block is executed and the value is calculated. In the following example, delayed prints a message demonstrating that the method has been entered. Next, delayed prints a message with its value. Finally, delayed returns ‘t’:
object Demo {
def main(args: Array[String]) {
delayed(time());
}
def time() = {
println("Getting time in nano seconds")
System.nanoTime
}
def delayed( t: => Long ) = {
println("In delayed method")
println("Param: " + t)
}
}
In delayed method
Getting time in nano seconds
Param: 2027245119786400
PROS AND CONS FOR EACH CASE
CBN:
+Terminates more often * check below above termination *
+ Has the advantage that a function argument is not evaluated if the corresponding parameter is unused in the evaluation of the function body
-It is slower, it creates more classes (meaning the program takes longer to load) and it consumes more memory.
CBV:
+ It is often exponentially more efficient than CBN, because it avoids this repeated recomputation of arguments expressions that call by name entails. It evaluates every function argument only once
+ It plays much nicer with imperative effects and side effects, because you tend to know much better when expressions will be evaluated.
-It may lead to a loop during its parameters evaluation * check below above termination *
What if termination is not guaranteed?
-If CBV evaluation of an expression e terminates, then CBN evaluation of e terminates too
-The other direction is not true
Non-termination example
def first(x:Int, y:Int)=x
Consider the expression first(1,loop)
CBN: first(1,loop) → 1
CBV: first(1,loop) → reduce arguments of this expression. Since one is a loop, it reduce arguments infinivly. It doesn’t terminate
DIFFERENCES IN EACH CASE BEHAVIOUR
Let's define a method test that will be
Def test(x:Int, y:Int) = x * x //for call-by-value
Def test(x: => Int, y: => Int) = x * x //for call-by-name
Case1 test(2,3)
test(2,3) → 2*2 → 4
Since we start with already evaluated arguments it will be the same amount of steps for call-by-value and call-by-name
Case2 test(3+4,8)
call-by-value: test(3+4,8) → test(7,8) → 7 * 7 → 49
call-by-name: (3+4)*(3+4) → 7 * (3+4) → 7 * 7 → 49
In this case call-by-value performs less steps
Case3 test(7, 2*4)
call-by-value: test(7, 2*4) → test(7,8) → 7 * 7 → 49
call-by-name: (7)*(7) → 49
We avoid the unnecessary computation of the second argument
Case4 test(3+4, 2*4)
call-by-value: test(7, 2*4) → test(7,8) → 7 * 7 → 49
call-by-name: (3+4)*(3+4) → 7*(3+4) → 7*7 → 49
Different approach
First, let's assume we have a function with a side-effect. This function prints something out and then returns an Int.
def something() = {
println("calling something")
1 // return value
}
Now we are going to define two function that accept Int arguments that are exactly the same except that one takes the argument in a call-by-value style (x: Int) and the other in a call-by-name style (x: => Int).
def callByValue(x: Int) = {
println("x1=" + x)
println("x2=" + x)
}
def callByName(x: => Int) = {
println("x1=" + x)
println("x2=" + x)
}
Now what happens when we call them with our side-effecting function?
scala> callByValue(something())
calling something
x1=1
x2=1
scala> callByName(something())
calling something
x1=1
calling something
x2=1
So you can see that in the call-by-value version, the side-effect of the passed-in function call (something()) only happened once. However, in the call-by-name version, the side-effect happened twice.
This is because call-by-value functions compute the passed-in expression's value before calling the function, thus the same value is accessed every time. However, call-by-name functions recompute the passed-in expression's value every time it is accessed.
EXAMPLES WHERE IT IS BETTER TO USE CALL-BY-NAME
From: https://stackoverflow.com/a/19036068/1773841
Simple performance example: logging.
Let's imagine an interface like this:
trait Logger {
def info(msg: => String)
def warn(msg: => String)
def error(msg: => String)
}
And then used like this:
logger.info("Time spent on X: " + computeTimeSpent)
If the info method doesn't do anything (because, say, the logging level was configured for higher than that), then computeTimeSpent never gets called, saving time. This happens a lot with loggers, where one often sees string manipulation which can be expensive relative to the tasks being logged.
Correctness example: logic operators.
You have probably seen code like this:
if (ref != null && ref.isSomething)
Imagine you would declare && method like this:
trait Boolean {
def &&(other: Boolean): Boolean
}
then, whenever ref is null, you'll get an error because isSomething will be called on a nullreference before being passed to &&. For this reason, the actual declaration is:
trait Boolean {
def &&(other: => Boolean): Boolean =
if (this) this else other
}
Going through an example should help you better understand the difference.
Let's definie a simple function that returns the current time:
def getTime = System.currentTimeMillis
Now we'll define a function, by name, that prints two times delayed by a second:
def getTimeByName(f: => Long) = { println(f); Thread.sleep(1000); println(f)}
And a one by value:
def getTimeByValue(f: Long) = { println(f); Thread.sleep(1000); println(f)}
Now let's call each:
getTimeByName(getTime)
// prints:
// 1514451008323
// 1514451009325
getTimeByValue(getTime)
// prints:
// 1514451024846
// 1514451024846
The result should explain the difference. The snippet is available here.
CallByName is invoked when used and callByValue is invoked whenever the statement is encountered.
For example:-
I have a infinite loop i.e. if you execute this function we will never get scala prompt.
scala> def loop(x:Int) :Int = loop(x-1)
loop: (x: Int)Int
a callByName function takes above loop method as an argument and it is never used inside its body.
scala> def callByName(x:Int,y: => Int)=x
callByName: (x: Int, y: => Int)Int
On execution of callByName method we don't find any problem ( we get scala prompt back ) as we are no where using the loop function inside callByName function.
scala> callByName(1,loop(10))
res1: Int = 1
scala>
a callByValue function takes above loop method as a parameter as a result inside function or expression is evaluated before executing outer function there by loop function executed recursively and we never get scala prompt back.
scala> def callByValue(x:Int,y:Int) = x
callByValue: (x: Int, y: Int)Int
scala> callByValue(1,loop(1))
See this:
object NameVsVal extends App {
def mul(x: Int, y: => Int) : Int = {
println("mul")
x * y
}
def add(x: Int, y: Int): Int = {
println("add")
x + y
}
println(mul(3, add(2, 1)))
}
y: => Int is call by name. What is passed as call by name is add(2, 1). This will be evaluated lazily. So output on console will be "mul" followed by "add", although add seems to be called first. Call by name acts as kind of passing a function pointer.
Now change from y: => Int to y: Int. Console will show "add" followed by "mul"! Usual way of evaluation.
I don't think all the answers here do the correct justification:
In call by value the arguments are computed just once:
def f(x : Int, y :Int) = x
// following the substitution model
f(12 + 3, 4 * 11)
f(15, 4194304)
15
you can see above that all the arguments are evaluated whether needed are not, normally call-by-value can be fast but not always like in this case.
If the evaluation strategy was call-by-name then the decomposition would have been:
f(12 + 3, 4 * 11)
12 + 3
15
as you can see above we never needed to evaluate 4 * 11 and hence saved a bit of computation which may be beneficial sometimes.
Scala variable evaluation explained here in better https://sudarshankasar.medium.com/evaluation-rules-in-scala-1ed988776ae8
def main(args: Array[String]): Unit = {
//valVarDeclaration 2
println("****starting the app***") // ****starting the app***
val defVarDeclarationCall1 = defVarDeclaration // defVarDeclaration 1
val defVarDeclarationCall2 = defVarDeclaration // defVarDeclaration 1
val valVarDeclarationCall1 = valVarDeclaration //
val valVarDeclarationCall2 = valVarDeclaration //
val lazyValVarDeclarationCall1 = lazyValVarDeclaration // lazyValVarDeclaration 3
val lazyValVarDeclarationCall2 = lazyValVarDeclaration //
callByValue({
println("passing the value "+ 10)
10
}) // passing the value 10
// call by value example
// 10
callByName({
println("passing the value "+ 20)
20
}) // call by name example
// passing the value 20
// 20
}
def defVarDeclaration = {
println("defVarDeclaration " + 1)
1
}
val valVarDeclaration = {
println("valVarDeclaration " + 2)
2
}
lazy val lazyValVarDeclaration = {
println("lazyValVarDeclaration " + 3)
3
}
def callByValue(x: Int): Unit = {
println("call by value example ")
println(x)
}
def callByName(x: => Int): Unit = {
println("call by name example ")
println(x)
}
Note: this is a theoretical question, I am not trying to fix anything, nor am I trying to achieve any effect for a practical purpose
When creating a lambda in Scala using the (arguments)=>expression syntax, can the return type be explicitly provided?
Lambdas are no different than methods on that they both are specified as expressions, but as far as I understand it, the return type of methods is defined easily with the def name(arguments): return type = expression syntax.
Consider this (illustrative) example:
def sequence(start: Int, next: Int=>Int): ()=>Int = {
var x: Int = start
//How can I denote that this function should return an integer?
() => {
var result: Int = x
x = next(x)
result
}
}
You can always declare the type of an expression by appending : and the type. So, for instance:
((x: Int) => x.toString): (Int => String)
This is useful if you, for instance, have a big complicated expression and you don't want to rely upon type inference to get the types straight.
{
if (foo(y)) x => Some(bar(x))
else x => None
}: (Int => Option[Bar])
// Without type ascription, need (x: Int)
But it's probably even clearer if you assign the result to a temporary variable with a specified type:
val fn: Int => Option[Bar] = {
if (foo(y)) x => Some(bar(x))
else _ => None
}
Let say you have this function:
def mulF(a: Int, b: Int): Long = {
a.toLong * b
}
The same function can be written as lambda with defined input and output types:
val mulLambda: (Int, Int) => Long = (x: Int, y: Int) => { x.toLong * y }
x => x:SomeType
Did not know the answer myself as I never had the need for it, but my gut feeling was that this will work. And trying it in a worksheet confirmed it.
Edit: I provided this answer before there was an example above. It is true that this is not needed in the concrete example. But in rare cases where you'd need it, the syntax I showed will work.