I noticed that Scala provide lazy vals. But I don't get what they do.
scala> val x = 15
x: Int = 15
scala> lazy val y = 13
y: Int = <lazy>
scala> x
res0: Int = 15
scala> y
res1: Int = 13
The REPL shows that y is a lazy val, but how is it different from a normal val?
The difference between them is, that a val is executed when it is defined whereas a lazy val is executed when it is accessed the first time.
scala> val x = { println("x"); 15 }
x
x: Int = 15
scala> lazy val y = { println("y"); 13 }
y: Int = <lazy>
scala> x
res2: Int = 15
scala> y
y
res3: Int = 13
scala> y
res4: Int = 13
In contrast to a method (defined with def) a lazy val is executed once and then never again. This can be useful when an operation takes long time to complete and when it is not sure if it is later used.
scala> class X { val x = { Thread.sleep(2000); 15 } }
defined class X
scala> class Y { lazy val y = { Thread.sleep(2000); 13 } }
defined class Y
scala> new X
res5: X = X#262505b7 // we have to wait two seconds to the result
scala> new Y
res6: Y = Y#1555bd22 // this appears immediately
Here, when the values x and y are never used, only x unnecessarily wasting resources. If we suppose that y has no side effects and that we do not know how often it is accessed (never, once, thousands of times) it is useless to declare it as def since we don't want to execute it several times.
If you want to know how lazy vals are implemented, see this question.
This feature helps not only delaying expensive calculations, but is also useful to construct mutual dependent or cyclic structures. E.g. this leads to an stack overflow:
trait Foo { val foo: Foo }
case class Fee extends Foo { val foo = Faa() }
case class Faa extends Foo { val foo = Fee() }
println(Fee().foo)
//StackOverflowException
But with lazy vals it works fine
trait Foo { val foo: Foo }
case class Fee extends Foo { lazy val foo = Faa() }
case class Faa extends Foo { lazy val foo = Fee() }
println(Fee().foo)
//Faa()
I understand that the answer is given but I wrote a simple example to make it easy to understand for beginners like me:
var x = { println("x"); 15 }
lazy val y = { println("y"); x + 1 }
println("-----")
x = 17
println("y is: " + y)
Output of above code is:
x
-----
y
y is: 18
As it can be seen, x is printed when it's initialized, but y is not printed when it's initialized in same way (I have taken x as var intentionally here - to explain when y gets initialized). Next when y is called, it's initialized as well as value of last 'x' is taken into consideration but not the old one.
Hope this helps.
A lazy val is most easily understood as a "memoized (no-arg) def".
Like a def, a lazy val is not evaluated until it is invoked. But the result is saved so that subsequent invocations return the saved value. The memoized result takes up space in your data structure, like a val.
As others have mentioned, the use cases for a lazy val are to defer expensive computations until they are needed and store their results, and to solve certain circular dependencies between values.
Lazy vals are in fact implemented more or less as memoized defs. You can read about the details of their implementation here:
http://docs.scala-lang.org/sips/pending/improved-lazy-val-initialization.html
Also lazy is useful without cyclic dependencies, as in the following code:
abstract class X {
val x: String
println ("x is "+x.length)
}
object Y extends X { val x = "Hello" }
Y
Accessing Y will now throw null pointer exception, because x is not yet initialized.
The following, however, works fine:
abstract class X {
val x: String
println ("x is "+x.length)
}
object Y extends X { lazy val x = "Hello" }
Y
EDIT: the following will also work:
object Y extends { val x = "Hello" } with X
This is called an "early initializer". See this SO question for more details.
A demonstration of lazy - as defined above - execution when defined vs execution when accessed: (using 2.12.7 scala shell)
// compiler says this is ok when it is lazy
scala> lazy val t: Int = t
t: Int = <lazy>
//however when executed, t recursively calls itself, and causes a StackOverflowError
scala> t
java.lang.StackOverflowError
...
// when the t is initialized to itself un-lazily, the compiler warns you of the recursive call
scala> val t: Int = t
<console>:12: warning: value t does nothing other than call itself recursively
val t: Int = t
scala> lazy val lazyEight = {
| println("I am lazy !")
| 8
| }
lazyEight: Int = <lazy>
scala> lazyEight
I am lazy !
res1: Int = 8
All vals are initialized during object construction
Use lazy keyword to defer initialization until first usage
Attention: Lazy vals are not final and therefore might show performance drawbacks
Related
I noticed that Scala provide lazy vals. But I don't get what they do.
scala> val x = 15
x: Int = 15
scala> lazy val y = 13
y: Int = <lazy>
scala> x
res0: Int = 15
scala> y
res1: Int = 13
The REPL shows that y is a lazy val, but how is it different from a normal val?
The difference between them is, that a val is executed when it is defined whereas a lazy val is executed when it is accessed the first time.
scala> val x = { println("x"); 15 }
x
x: Int = 15
scala> lazy val y = { println("y"); 13 }
y: Int = <lazy>
scala> x
res2: Int = 15
scala> y
y
res3: Int = 13
scala> y
res4: Int = 13
In contrast to a method (defined with def) a lazy val is executed once and then never again. This can be useful when an operation takes long time to complete and when it is not sure if it is later used.
scala> class X { val x = { Thread.sleep(2000); 15 } }
defined class X
scala> class Y { lazy val y = { Thread.sleep(2000); 13 } }
defined class Y
scala> new X
res5: X = X#262505b7 // we have to wait two seconds to the result
scala> new Y
res6: Y = Y#1555bd22 // this appears immediately
Here, when the values x and y are never used, only x unnecessarily wasting resources. If we suppose that y has no side effects and that we do not know how often it is accessed (never, once, thousands of times) it is useless to declare it as def since we don't want to execute it several times.
If you want to know how lazy vals are implemented, see this question.
This feature helps not only delaying expensive calculations, but is also useful to construct mutual dependent or cyclic structures. E.g. this leads to an stack overflow:
trait Foo { val foo: Foo }
case class Fee extends Foo { val foo = Faa() }
case class Faa extends Foo { val foo = Fee() }
println(Fee().foo)
//StackOverflowException
But with lazy vals it works fine
trait Foo { val foo: Foo }
case class Fee extends Foo { lazy val foo = Faa() }
case class Faa extends Foo { lazy val foo = Fee() }
println(Fee().foo)
//Faa()
I understand that the answer is given but I wrote a simple example to make it easy to understand for beginners like me:
var x = { println("x"); 15 }
lazy val y = { println("y"); x + 1 }
println("-----")
x = 17
println("y is: " + y)
Output of above code is:
x
-----
y
y is: 18
As it can be seen, x is printed when it's initialized, but y is not printed when it's initialized in same way (I have taken x as var intentionally here - to explain when y gets initialized). Next when y is called, it's initialized as well as value of last 'x' is taken into consideration but not the old one.
Hope this helps.
A lazy val is most easily understood as a "memoized (no-arg) def".
Like a def, a lazy val is not evaluated until it is invoked. But the result is saved so that subsequent invocations return the saved value. The memoized result takes up space in your data structure, like a val.
As others have mentioned, the use cases for a lazy val are to defer expensive computations until they are needed and store their results, and to solve certain circular dependencies between values.
Lazy vals are in fact implemented more or less as memoized defs. You can read about the details of their implementation here:
http://docs.scala-lang.org/sips/pending/improved-lazy-val-initialization.html
Also lazy is useful without cyclic dependencies, as in the following code:
abstract class X {
val x: String
println ("x is "+x.length)
}
object Y extends X { val x = "Hello" }
Y
Accessing Y will now throw null pointer exception, because x is not yet initialized.
The following, however, works fine:
abstract class X {
val x: String
println ("x is "+x.length)
}
object Y extends X { lazy val x = "Hello" }
Y
EDIT: the following will also work:
object Y extends { val x = "Hello" } with X
This is called an "early initializer". See this SO question for more details.
A demonstration of lazy - as defined above - execution when defined vs execution when accessed: (using 2.12.7 scala shell)
// compiler says this is ok when it is lazy
scala> lazy val t: Int = t
t: Int = <lazy>
//however when executed, t recursively calls itself, and causes a StackOverflowError
scala> t
java.lang.StackOverflowError
...
// when the t is initialized to itself un-lazily, the compiler warns you of the recursive call
scala> val t: Int = t
<console>:12: warning: value t does nothing other than call itself recursively
val t: Int = t
scala> lazy val lazyEight = {
| println("I am lazy !")
| 8
| }
lazyEight: Int = <lazy>
scala> lazyEight
I am lazy !
res1: Int = 8
All vals are initialized during object construction
Use lazy keyword to defer initialization until first usage
Attention: Lazy vals are not final and therefore might show performance drawbacks
In scala def is used to define a method and val, var are used for defining variables.
Consider the following code:
scala> def i = 3
i: Int
scala> i.getClass()
res0: Class[Int] = int
scala> val v = 2
v: Int = 2
scala> v.getClass()
res1: Class[Int] = int
scala> println(v)
2
scala> println(i)
3
scala> i+v
res4: Int = 5
scala> def o = () => 2+3
o: () => Int
scala> o.getClass()
res5: Class[_ <: () => Int] = class $$Lambda$1139/1753607449
Why does variable definition work using def? If it is defining a function that returns an Int then why does getClass show Int instead of a function object?
Unlike val or var declaration, def i = 3 is not variable declaration. You are defining a method/function which returns a constant 3 and i does not take any parameters.
declaration using val and var get evaluated immediately but in case of lazy val and def evaluation happens when called explicitly.
i is a not argument function. In order to get rid of confusion you could declare it using empty parenthesis as well
def i() = 3
Difference between lazy val and def is
lazy val is lazily evaluated and the result is cached. That means further
def declaration is evaluated every time you call method name.
Example using Scala REPL
scala> lazy val a = { println("a evaluated"); 1}
a: Int = <lazy>
scala> def i = { println("i function evaluated"); 2}
i: Int
scala> a
a evaluated
res0: Int = 1
scala> a
res1: Int = 1
scala> a
res2: Int = 1
scala> i
i function evaluated
res3: Int = 2
scala> i
i function evaluated
res4: Int = 2
scala> i
i function evaluated
res5: Int = 2
Notice that a is evaluated only once and further invocations of a return the cached result i.e lazy val is evaluated once when it is called and the result is stored forever. So you see println output once
Notice function is evaluated every time it is invoked. In this case you see println output every time you invoke the function
General Convention
There's a convention of using an empty parameter list when the method has side effects and leaving them off when its pure.
edited
scala> def i = 1
i: Int
scala> :type i
Int
scala> :type i _
() => Int
EDIT: My answer addresses the question of revision #3.
It is quite useful to look on the code in the middle of the compilation process where you can look on what your code is actually translated to. The following simple program:
object TestApp {
def definedVal = 3
val valVal = 3
lazy val lazyValVal = 3
def main(args: Array[String]) {
println(definedVal)
println(valVal)
println(lazyValVal)
}
}
is translated to the following (using -Xprint:mixin compiler option):
[[syntax trees at end of mixin]] // test.scala
package <empty> {
object TestApp extends Object {
#volatile private[this] var bitmap$0: Boolean = false;
private def lazyValVal$lzycompute(): Int = {
{
TestApp.this.synchronized({
if (TestApp.this.bitmap$0.unary_!())
{
TestApp.this.lazyValVal = 3;
TestApp.this.bitmap$0 = true;
()
};
scala.runtime.BoxedUnit.UNIT
});
()
};
TestApp.this.lazyValVal
};
def definedVal(): Int = 3;
private[this] val valVal: Int = _;
<stable> <accessor> def valVal(): Int = TestApp.this.valVal;
lazy private[this] var lazyValVal: Int = _;
<stable> <accessor> lazy def lazyValVal(): Int = if (TestApp.this.bitmap$0.unary_!())
TestApp.this.lazyValVal$lzycompute()
else
TestApp.this.lazyValVal;
def main(args: Array[String]): Unit = {
scala.this.Predef.println(scala.Int.box(TestApp.this.definedVal()));
scala.this.Predef.println(scala.Int.box(TestApp.this.valVal()));
scala.this.Predef.println(scala.Int.box(TestApp.this.lazyValVal()))
};
def <init>(): TestApp.type = {
TestApp.super.<init>();
TestApp.this.valVal = 3;
()
}
}
}
From the output above it is possible to conclude the following:
definedVal is actually a method.
valVal is a field which is initialized in the constructor and has an automatically generated accessor.
For the lazy field lazyValVal compiler generates compute method which is called only once when the field is accessed the first time.
There is few different concept. Call by name, call by value and call by need. All def is essentially calls by name. What do you mean variable definition using def ??
Looks like duplicate to me:
Call by name vs call by value in Scala, clarification needed
More details in wiki: https://en.wikipedia.org/wiki/Evaluation_strategy#Call_by_name
What is the difference between:
def even: Int => Boolean = _ % 2 == 0
and
val even: Int => Boolean = _ % 2 == 0
Both can be called like even(10).
Method def even evaluates on call and creates new function every time (new instance of Function1).
def even: Int => Boolean = _ % 2 == 0
even eq even
//Boolean = false
val even: Int => Boolean = _ % 2 == 0
even eq even
//Boolean = true
With def you can get new function on every call:
val test: () => Int = {
val r = util.Random.nextInt
() => r
}
test()
// Int = -1049057402
test()
// Int = -1049057402 - same result
def test: () => Int = {
val r = util.Random.nextInt
() => r
}
test()
// Int = -240885810
test()
// Int = -1002157461 - new result
val evaluates when defined, def - when called:
scala> val even: Int => Boolean = ???
scala.NotImplementedError: an implementation is missing
scala> def even: Int => Boolean = ???
even: Int => Boolean
scala> even
scala.NotImplementedError: an implementation is missing
Note that there is a third option: lazy val.
It evaluates when called the first time:
scala> lazy val even: Int => Boolean = ???
even: Int => Boolean = <lazy>
scala> even
scala.NotImplementedError: an implementation is missing
But returns the same result (in this case same instance of FunctionN) every time:
lazy val even: Int => Boolean = _ % 2 == 0
even eq even
//Boolean = true
lazy val test: () => Int = {
val r = util.Random.nextInt
() => r
}
test()
// Int = -1068569869
test()
// Int = -1068569869 - same result
Performance
val evaluates when defined.
def evaluates on every call, so performance could be worse than val for multiple calls. You'll get the same performance with a single call. And with no calls you'll get no overhead from def, so you can define it even if you will not use it in some branches.
With a lazy val you'll get a lazy evaluation: you can define it even if you will not use it in some branches, and it evaluates once or never, but you'll get a little overhead from double check locking on every access to your lazy val.
As #SargeBorsch noted you could define method, and this is the fastest option:
def even(i: Int): Boolean = i % 2 == 0
But if you need a function (not method) for function composition or for higher order functions (like filter(even)) compiler will generate a function from your method every time you are using it as function, so performance could be slightly worse than with val.
Consider this:
scala> def even: (Int => Boolean) = {
println("def");
(x => x % 2 == 0)
}
even: Int => Boolean
scala> val even2: (Int => Boolean) = {
println("val");
(x => x % 2 == 0)
}
val //gets printed while declaration. line-4
even2: Int => Boolean = <function1>
scala> even(1)
def
res9: Boolean = false
scala> even2(1)
res10: Boolean = false
Do you see the difference? In short:
def: For every call to even, it calls the body of the even method again. But with even2 i.e. val, the function is initialized only once while declaration (and hence it prints val at line 4 and never again) and the same output is used each time it accessed. For example try doing this:
scala> import scala.util.Random
import scala.util.Random
scala> val x = { Random.nextInt }
x: Int = -1307706866
scala> x
res0: Int = -1307706866
scala> x
res1: Int = -1307706866
When x is initialized, the value returned by Random.nextInt is set as the final value of x. Next time x is used again, it will always return the same value.
You can also lazily initialize x. i.e. first time it is used it is initialized and not while declaration. For example:
scala> lazy val y = { Random.nextInt }
y: Int = <lazy>
scala> y
res4: Int = 323930673
scala> y
res5: Int = 323930673
See this:
var x = 2 // using var as I need to change it to 3 later
val sq = x*x // evaluates right now
x = 3 // no effect! sq is already evaluated
println(sq)
Surprisingly, this will print 4 and not 9! val (even var) is evaluated immediately and assigned.
Now change val to def.. it will print 9! Def is a function call.. it will evaluate each time it is called.
val i.e. "sq" is by Scala definition is fixed. It is evaluated right at the time of declaration, you can't change later. In other examples, where even2 also val, but it declared with function signature i.e. "(Int => Boolean)", so it is not Int type. It is a function and it's value is set by following expression
{
println("val");
(x => x % 2 == 0)
}
As per Scala val property, you can't assign another function to even2, same rule as sq.
About why calling eval2 val function not printing "val" again and again ?
Orig code:
val even2: (Int => Boolean) = {
println("val");
(x => x % 2 == 0)
}
We know, in Scala last statement of above kind of expression (inside { .. }) is actually return to the left hand side. So you end up setting even2 to "x => x % 2 == 0" function, which matches with the type you declared for even2 val type i.e. (Int => Boolean), so compiler is happy. Now even2 only points to "(x => x % 2 == 0)" function (not any other statement before i.e. println("val") etc. Invoking event2 with different parameters will actually invoke "(x => x % 2 == 0)" code, as only that is saved with event2.
scala> even2(2)
res7: Boolean = true
scala> even2(3)
res8: Boolean = false
Just to clarify this more, following is different version of the code.
scala> val even2: (Int => Boolean) = {
| println("val");
| (x => {
| println("inside final fn")
| x % 2 == 0
| })
| }
What will happen ? here we see "inside final fn" printed again and again, when you call even2().
scala> even2(3)
inside final fn
res9: Boolean = false
scala> even2(2)
inside final fn
res10: Boolean = true
scala>
Executing a definition such as def x = e will not evaluate the expression e. In- stead e is evaluated whenever x is invoked.
Alternatively, Scala offers a value definition
val x = e,which does evaluate the right-hand-side as part of the evaluation of the definition.
If x is then used subsequently, it is immediately replaced by the pre-computed value of e, so that the expression need not be evaluated again.
also, Val is a by value evaluation. Which means the right-hand side expression is evaluated during definition. Where Def is by name evaluation. It will not evaluate until it's used.
In addition to the above helpful replies, my findings are:
def test1: Int => Int = {
x => x
}
--test1: test1[] => Int => Int
def test2(): Int => Int = {
x => x+1
}
--test2: test2[]() => Int => Int
def test3(): Int = 4
--test3: test3[]() => Int
The above shows that “def” is a method (with zero argument parameters) that returns another function "Int => Int” when invoked.
The conversion of methods to functions is well explained here: https://tpolecat.github.io/2014/06/09/methods-functions.html
In REPL,
scala> def even: Int => Boolean = { _% 2 == 0 }
even: Int => Boolean
scala> val even: Int => Boolean = { _% 2 == 0 }
even: Int => Boolean = $$Lambda$1157/1017502292#57a0aeb8
def means call-by-name, evaluated on demand
val means call-by-value, evaluated while initialization
Note: There are different types of functions in Scala: abstract, concrete, anonymous, high order, pure, impure etc...
Explaining val function:
A val function in Scala is a complete object. There are traits in Scala to represent functions with various numbers of arguments: Function0, Function1, Function2, etc. As an instance of a class that implements one of these traits, a function object has methods. One of these methods is the apply method, which contains the code that implements the body of the function.
When we create a variable whose value is a function object and we then reference that variable followed by parentheses, that gets converted into a call to the apply method of the function object.
Explaining Method i.e def:
Methods in Scala are not values, but functions are.
A Scala method, as in Java, is a part of a class. It has a name, a signature, optionally some annotations, and some bytecode.
The implementation of a method is an ordered sequence of statements that produces a value that must be compatible with its return type.
In Scala 2.10.4, Given the following class:
scala> class Foo {
| val x = true
| val f = if (x) 100 else 200
| }
defined class Foo
The following two examples make sense to me:
scala> new Foo {}.f
res0: Int = 100
scala> new Foo { override val x = false}.f
res1: Int = 200
But, why doesn't this call return 100?
scala> new Foo { override val x = true }.f
res2: Int = 200
Because vals aren't initialized more than once, x is actually null (or false for a default Boolean) during the initialization of Foo, and then initialized in the anonymous class that is extending Foo in your example.
We can test it more easily with an AnyRef:
class Foo {
val x = ""
val f = if (x == null) "x is null" else "not null"
}
scala> new Foo { override val x = "a" }.f
res10: String = x is null
scala> new Foo {}.f
res11: String = not null
There's a full explanation in the Scala FAQ. Excerpt:
Naturally when a val is overridden, it is not initialized more than once. So though x2 in the above example is seemingly defined at every point, this is not the case: an overridden val will appear to be null during the construction of superclasses, as will an abstract val.
A simple way to avoid this would be to use a lazy val or def, if the val being referenced may be overridden.
Additionally, you can use the -Xcheckinit compiler flag to warn you about potential initialization errors like this.
In response to a comment from another question, I tried putting this code into Scala:
trait Foo
new Foo { self =>
val self.x = 3
}
It doesn't compile, of course, but the error is puzzling me:
recursive value x$1 needs type
val self.x = 3
^
How did this code get past the parser?
The -Xprint:parse is also kind of bizarre:
<synthetic> private[this] val x$1 = 3: #scala.unchecked match {
case self.x => ()
}
Is that a match in the type annotation for 3? edit: Apparently not; that's the syntax for annotations.
Variable definitions in Scala are actually pattern matching. That is, when you write
val x = y
println(x)
That's basically the same as writing
y match {
case x =>
println(x)
This can easily be seen in things like this:
val List(a, b, c) = someList
val RegexPattern(year, month, date) = "2013-10-23"
Another thing that's valid is constant pattern matching:
object X {
val One = 1
}
scala> val X.One = 1
scala> val X.One = 2
scala.MatchError: 2 (of class java.lang.Integer)
And anything with a parenthesis will call an extractor:
object Y {
val ymd = """(\d\d\d\d)-(\d\d)-(\d\d)""".r
}
scala> val Y.ymd(year, month, day) = "2013-10-23"
year: String = 2013
month: String = 10
day: String = 23
So, you see, there's nothing syntactically illegal with what you wrote, just the specifics.