Only allow inheritance within the same package - scala

I have a trait which will have a finite number of subclasses. At first I used the sealed modifier and defined the trait and all its subclasses in the same file. After the classes grew, I decided I wanted to refactor them into separate files, however once I did this I could no longer use the sealed modifier on the trait due to the constraint that all subclasses of a sealed trait must be in the same file.
In Scala, Is there a similar way to have a finite number of subclasses for a trait within the same package, across separate files while still gaining the compile-time advantages of sealed traits when doing exhaustive pattern matching?

Sealing is the only way to ask for exhaustiveness checking. However, you could define package-private traits AGuts, BGuts, CGuts, etc. in separate files and then in one file create sealed subclasses A, B, C, etc. as you did before, but mixing in those guts defined elsewhere.
sealed abstract class Thing
class A extends Thing with AGuts
class B extends Thing with BGuts
...

Related

Abstract class and trait use case [duplicate]

What is the advantage of using an abstract class instead of a trait (apart from performance)? It seems like abstract classes can be replaced by traits in most cases.
I can think of two differences
Abstract classes can have constructor parameters as well as type parameters. Traits can have only type parameters. There was some discussion that in future even traits can have constructor parameters
Abstract classes are fully interoperable with Java. You can call them from Java code without any wrappers. Traits are fully interoperable only if they do not contain any implementation code
There's a section in Programming in Scala called "To trait, or not to trait?" which addresses this question. Since the 1st ed is available online, I'm hoping it's OK to quote the whole thing here. (Any serious Scala programmer should buy the book):
Whenever you implement a reusable collection of behavior, you will
have to decide whether you want to use a trait or an abstract class.
There is no firm rule, but this section contains a few guidelines to
consider.
If the behavior will not be reused, then make it a concrete class. It
is not reusable behavior after all.
If it might be reused in multiple, unrelated classes, make it a trait.
Only traits can be mixed into different parts of the class hierarchy.
If you want to inherit from it in Java code, use an abstract class.
Since traits with code do not have a close Java analog, it tends to be
awkward to inherit from a trait in a Java class. Inheriting from a
Scala class, meanwhile, is exactly like inheriting from a Java class.
As one exception, a Scala trait with only abstract members translates
directly to a Java interface, so you should feel free to define such
traits even if you expect Java code to inherit from it. See Chapter 29
for more information on working with Java and Scala together.
If you plan to distribute it in compiled form, and you expect outside
groups to write classes inheriting from it, you might lean towards
using an abstract class. The issue is that when a trait gains or loses
a member, any classes that inherit from it must be recompiled, even if
they have not changed. If outside clients will only call into the
behavior, instead of inheriting from it, then using a trait is fine.
If efficiency is very important, lean towards using a class. Most Java
runtimes make a virtual method invocation of a class member a faster
operation than an interface method invocation. Traits get compiled to
interfaces and therefore may pay a slight performance overhead.
However, you should make this choice only if you know that the trait
in question constitutes a performance bottleneck and have evidence
that using a class instead actually solves the problem.
If you still do not know, after considering the above, then start by
making it as a trait. You can always change it later, and in general
using a trait keeps more options open.
As #Mushtaq Ahmed mentioned, a trait cannot have any parameters passed to the primary constructor of a class.
Another difference is the treatment of super.
The other difference between classes and traits is that whereas in classes, super calls are statically bound, in traits, they are dynamically bound. If you write super.toString in a class, you know exactly which method implementation will be invoked. When you write the same thing in a trait, however, the method implementation to invoke for the super call is undefined when you define the trait.
See the rest of Chapter 12 for more details.
Edit 1 (2013):
There is a subtle difference in the way abstract classes behaves compared to traits. One of the linearization rules is that it preserves the inheritance hierarchy of the classes, which tends to push abstract classes later in the chain while traits can happily be mixed in. In certain circumstances, it's actually preferable to be in latter position of the class linearization, so abstract classes could be used for that. See constraining class linearization (mixin order) in Scala.
Edit 2 (2018):
As of Scala 2.12, trait's binary compatibility behavior has changed. Prior to 2.12, adding or removing a member to the trait required recompilation of all classes that inherit the trait, even if the classes have not changed. This is due to the way traits were encoded in JVM.
As of Scala 2.12, traits compile to Java interfaces, so the requirement has relaxed a bit. If the trait does any of the following, its subclasses still require recompilation:
defining fields (val or var, but a constant is ok – final val without result type)
calling super
initializer statements in the body
extending a class
relying on linearization to find implementations in the right supertrait
But if the trait does not, you can now update it without breaking binary compatibility.
For whatever it is worth, Odersky et al's Programming in Scala recommends that, when you doubt, you use traits. You can always change them into abstract classes later on if needed.
Other than the fact that you cannot directly extend multiple abstract classes, but you can mixin multiple traits into a class, it's worth mentioning that traits are stackable, since super calls in a trait are dynamically bound (it is referring a class or trait mixed before current one).
From Thomas's answer in Difference between Abstract Class and Trait:
trait A{
def a = 1
}
trait X extends A{
override def a = {
println("X")
super.a
}
}
trait Y extends A{
override def a = {
println("Y")
super.a
}
}
scala> val xy = new AnyRef with X with Y
xy: java.lang.Object with X with Y = $anon$1#6e9b6a
scala> xy.a
Y
X
res0: Int = 1
scala> val yx = new AnyRef with Y with X
yx: java.lang.Object with Y with X = $anon$1#188c838
scala> yx.a
X
Y
res1: Int = 1
When extending an abstract class, this shows that the subclass is of a similar kind. This is not neccessarily the case when using traits, I think.
In Programming Scala the authors say that abstract classes make a classical object oriented "is-a" relationship while traits are a scala-way of composition.
Abstract classes can contain behaviour - They can parameterized with constructor args (which traits can't) and represent a working entity. Traits instead just represent a single feature, an interface of one functionality.
A class can inherit from multiple traits but only one abstract class.
Abstract classes can have constructor parameters as well as type parameters. Traits can have only type parameters. For example, you can’t say trait t(i: Int) { }; the i parameter is illegal.
Abstract classes are fully interoperable with Java. You can call them from Java code without any wrappers. Traits are fully interoperable only if they do not contain any implementation code.

Why the first base class in parent list must be non-trait class?

In the Scala spec, it's said that in a class template sc extends mt1, mt2, ..., mtn
Each trait reference mti must denote a trait. By contrast, the
superclass constructor sc normally refers to a class which is not a
trait. It is possible to write a list of parents that starts with a
trait reference, e.g. mt1 with …… with mtn. In that case the
list of parents is implicitly extended to include the supertype of
mt1 as first parent type. The new supertype must have at least one
constructor that does not take parameters. In the following, we will
always assume that this implicit extension has been performed, so that
the first parent class of a template is a regular superclass
constructor, not a trait reference.
If I understand it correctly, I think it means:
trait Base1 {}
trait Base2 {}
class Sub extends Base1 with Base2 {}
Will be implicitly extended to:
trait Base1 {}
trait Base2 {}
class Sub extends Object with Base1 with Base2 {}
My questions are:
Is my understanding correct?
Does this requirement (the first subclass in the parent list must be non-trait class) and the implicit extension only applies to class template (e.g. class Sub extends Mt1, Mt2) or also trait template (e.g. trait Sub extends Mt1, Mt2)?
Why this requirement and the implicit extension is necessary?
Disclaimer: I'm not and never was a member of the "Scala design committee" or anything like that, so the answer on the "why?" question is mostly speculation but I think a useful one.
Disclaimer #2: I've written this post over several hours and in several takes so it is probably not very consistent
Disclaimer #3 (a shameful self-promotion for the future readers): If you find this quite long answer useful, you might also take a look at my another long answer to another question by Lifu Huang on a similar topic.
Short answers
This is one of those complicated things for which I don't think there is a good short answer unless you already know what the answer is. Although my real answer will be long, here are my best short answers:
Why the first base class in parent list must be non-trait class?
Because there has to be only one non-trait base class and it makes thing easier if it is always the first
Is my understanding correct?
Yes, your implicit example is what will happen. However I'm not sure that it shows full understanding of the topic.
Does this requirement (the first subclass in the parent list must be non-trait class) and the implicit extension only applies to class template (e.g. class Sub extends Mt1, Mt2) or also trait template (e.g. trait Sub extends Mt1, Mt2)?
No, implicit extensions happens for traits as well. Actually how else you could expect Mt1 to have its own "supertype" to be promoted down to the class that extends it?
Actually here are two IMHO non-obvious examples proving this is true:
Example #1
trait TAny extends Any
trait TNo
// works
class CGood(val value: Int) extends AnyVal with TAny
// fails
// illegal inheritance; superclass AnyVal is not a subclass of the superclass Object
class CBad(val value: Int) extends AnyVal with TNo
This example fails because the spec says
The extends clause extends scsc with mt1mt1 with …… with mtnmtn can be omitted, in which case extends scala.AnyRef is assumed.
so TNo actually extends AnyRef which is incompatible with AnyVal.
Example #2
class CFirst
class CSecond extends CFirst
// did you know that traits can extend classes as well?
trait TFirst extends CFirst
trait TSecond extends CSecond
// works
class ChildGood extends TSecond with TFirst
// fails
// illegal inheritance; superclass CFirst is not a subclass of the superclass CSecond of the mixin trait TSecond
class ChildBad extends TFirst with TSecond
Again ChildBad fails because TSecond requires CSecond but TFirst only provides CFirst as the base class.
Why this requirement and the implicit extension is necessary?
There are three major reasons:
Compatibility with the main target platform (JVM)
Traits have "mixin" semantics: you have a class and you mix additional behavior in
Completeness, consistency and simplicity of the rest of the spec (e.g. of linearization rules). This might be restated as following: each class must declare 0 or 1 base non-trait classes and after compilation the target platform enforces that there will be exactly 1 non-trait base class. So it makes the rest of the spec easier if you just assume there is always exactly one base class. In such way you have to write this implicit extension rules only once rather than each time when the behavior depends on the base class.
Scala spec goals/intentions
I believe that when one reads a spec there are two different sets of questions:
What exactly is written? What is the meaning of the spec?
Why it is written so? What was the intention?
Actually I think in many cases #2 is more important than #1 but unfortunately specs rarely explicitly contain insights into that area. Anyway I will start with my speculations over #2: what were the intentions/goals/limitations of the classes system in Scala? The main high-level goal was to create a type system richer than the one in Java or .Net (which are quite similar) but that can be:
compiled back to an efficient code in those target platforms
allow reasonable two-way interaction between the Scala code and the "native" code in the target platforms
Side note: Support of the .Net was dropped years ago but it was one of the target platforms for years and this affected the design.
Single base class
Short summary: this section describes some reasons why Scala designers had a strong motivation to have the "exactly one base class" rule in the language.
A major problem with OO design and particularly inheritance is that AFAIK the question: "where exactly is the border between the "good and useful" practices and the "bad" ones?" is open. It means that each language must find out its own trade off between making impossible what is wrong and making possible (and easy) what is useful. Many believe that in C++, which obviously was a major inspiration for Java and .Net, that trade off is shifted too much into "allow everything even if it is potentially harmful" zone. It made many designers of newer languages to seek for more restricting trade off. Particularly both JVM and .Net platform enforce the rule that all types are split into "value types" (aka primitive types), "classes" and "interfaces" and each class, except the root class (java.lang.Object/System.Object), has exactly one "base class" and zero or more "base interfaces". This decision was a reaction to many issues of multiple inheritance including infamous "diamond problem" but actually many others as well.
Sidenote (about memory layout): Another major problem with multiple inheritance is objects layout in memory. Consider following ridiculous (and impossible in current Scala) example inspired by Achilles and the tortoise:
trait Achilles {
def getAchillesPos: Int
def stepAchilles(): Unit
}
class AchillesImpl(var achillesPos: Int) extends Achilles {
def getAchillesPos: Int = achillesPos
def stepAchilles(): Unit = {
achillesPos += 2
}
}
class TortoiseImpl(var tortoisePos: Int) {
def getTortoisePos: Int = tortoisePos
def stepTortoise(): Unit = {
tortoisePos += 1
}
}
class AchillesAndTortoise(handicap: Int) extends AchillesImpl(0) with TortoiseImpl(handicap) {
def catchTortoise(): Int = {
var time = 0
while (getAchillesPos < getTortoisePos) {
time += 1
stepAchilles()
stepTortoise()
}
time
}
}
The tricky part here is how to actually lay achillesPos and tortoisePos fields out in the memory (of the object). The issue is that you probably want to have only one compiled copy of all the methods in the memory and you want the code to be efficient. This means that getAchillesPos and stepAchilles should have know some fixed offset of the achillesPos regarding to the this pointer. Similarly getTortoisePos and stepTortoise should have know some fixed offset of the tortoisePos regarding to the this pointer. And all choices you have to achieve this goal don't look nice. For example:
You might decide that achillesPos is always first and tortoisePos is always second. But this means that in the instances of TortoiseImpl tortoisePos should also be the second field but there is nothing to fill the first field with so you waste some memory. Moreover if both AchillesImpl and TortoiseImpl come from pre-compiled libraries, you should have some way to move access to the fields in them as well.
You might try to "fix" this pointer on-the-fly when you call into TortoiseImpl (AFAIK this is the way C++ really works). This becomes especially funny when TortoiseImpl is an abstract class that is aware of the trait Achilles (but not the specific class AchillesImpl) via extends and tries to call back some methods from there via this or pass this to some method that takes Achilles as an argument so this has to be "fixed back". Note that this is not the same as the "diamond problem" because there is only one copy of all fields and implementations.
You might agree to have a unique copy of the methods compiled for each specific class that are aware of the specific layout. This is bad for memory usage and performance because it blows CPU caches and forces JIT to make independent optimizations for each.
You might say that no method except for getter and setter can have direct access to the fields and should use getters and setters instead. Or store all the fields in some kind of a dictionary which is effectively the same. This might be bad for performance (but this is the closest to what Scala does with mixin-traits).
In the actual Scala this issue does not exist because trait can't really declare any fields. When you declare val or var in a trait, you actually declare a getter (and a setter) method(s) that will be implemented by particular class that extends the trait and each class has full control over layout of the fields. And actually in terms of performance this most probably would work OK because JVM (JIT) can inline such a virtual call in many real-world scenarios.
End of the Sidenote
Another major point is interoperability with the target platform. Even if Scala somehow supported true multiple-inheritance so you can have a type that inherits from String with Date and that can be passed to both methods that expect String and that expect Date, how this would look like from the Java point of view? Also if the target platform enforces the rule that every class has to be an (indirect) sub-type of the same root class (Object), you can't work this around in your higher level language.
Traits and Mix-ins
Many think that "one class and many interfaces" trade-off that was made in Java and .Net is too restrictive. For example it makes it hard to share common default implementation of some of the interface methods between different classes. Actually over the time Java and .Net designers seem to come to the same conclusion and rolled out they own fixes for this kind of issues: Extension methods in .Net and then Default methods in Java. Scala designers added a feature called Mixins that was known to fare well in many practical cases. However unlike many other dynamic languages that has similar feature, Scala still had to meet the "exactly one base class" rule and other limitations of the target platform.
It is important to note that there are important scenarios when mixins are used in practice is to implement a variation of the Decorator or Adapter patterns both of which relies on the fact that you can restrict your base type to something more specific than Any or AnyRef. Prime example of such usage is the scala.collection package.
Scala syntax
So now you have following goals/restrictions:
Exactly one base class for each class
Ability to add logic to classes from mixins
Support of mixins with restricted base type
Classes from the target platform (Java) when seen from Scala are mapped to the Scala classes (because what else they can be mapped to?) and they come pre-compiled and we don't want to mess with their implementation
Other good qualities such as simplicity, type safety, determinism, etc.
If you want some kind of multiple inheritance support in your language, you need to develop conflict resolution rules: what happens when several base types provide some logic that would fit the same "slot" in your class. After prohibition of fields in traits we are left with the following "slots":
Base class in terms of the target platform
Constructors
Methods with the same name and signature
And possible conflict resolution strategies are:
Prohibit (fail compilation)
Decide which one wins and wipes others
Somehow chain them
Somehow preserve all with renaming. This is not really possible in JVM. For example in .Net see Explicit Interface Implementation
In a sense Scala uses all available (i.e. first 3) strategies but the high-level goal is: let's try to preserve as many logic as we can.
The most important part for this discussion is conflicts resolution for constructors and methods.
We want the rules to be the same for different slots because otherwise it is not clear how to achieve safety (if traits A and B both override methods foo and bar but resolution rules for foo and bar are different, invariants for A and B might easily be broken). Scala's approach is based on the class linearization. In short these is the way to "flatten" hierarchy of the base classes into a simple linear structure in some predictive way that is based on the idea that the lefter type in the with chain - the more "base" (higher in the inheritance) it is. After you do this, conflict resolution rule for methods becomes simple: you go through the list of the base types and chain behavior via super calls; if super is not called, you stop chaining. This produce quite predictable semantics that people can reason about.
Now assume you allow non-trait class to be not first. Consider following example:
class CBase {
def getValue = 2
}
trait TFirst extends CBase {
override def getValue = super.getValue + 1
}
trait TSecond extends CFirst {
override def getValue = super.getValue * 2
}
class CThird extends CBase with TSecond {
override def getValue = 100 - super.getValue
}
class Child extends TFirst with TSecond with CThird
In which order TFirst.getValue and TSecond.getValue should be called? Obviously CThird is already compiled and you can't change what the super for it is, so it has to be moved to the first position and there is already TSecond.getValue call inside it. But on the other hand this breaks the rule that everything on the left is base and everything on the right is child. The simplest way to not introduce such confusion is to enforce the rule that non-trait classes must go first.
The same logic applies if you just extend the previous example by substituting class CThird with a trait that extends it:
trait TFourth extends CThird
class AnotherChild extends TFirst with TSecond with TFourth
Again, the only non-trait class AnotherChild can extend is CThird and this again makes conflict resolution rules quite hard to reason about.
That's why Scala makes a rule much simpler: whatever provides the base class must come from the first position. And then it makes sense to extend the same rule upon the traits as well so if the first position is occupied by some trait - it also defines the base class.
1) Basically yes, your understanding is correct. Like in Java, every class inherits from java.lang.Object (AnyRef in Scala). So, since you are defining a concrete class, you will implicitly inherits from Object. If you check with the REPL, you got:
scala> trait Base1 {}
defined trait Base1
scala> trait Base2 {}
defined trait Base2
scala> class Sub extends Base1 with Base2 {}
defined class Sub
scala> classOf[Sub].getSuperclass
res0: Class[_ >: Sub] = class java.lang.Object
2) Yes, from the "Traits" paragraph in the specs, this applies also to them. In "Templates" paragraph we have:
The new supertype must have at least one constructor that does not take parameters
And then in "Traits" paragraph:
Unlike normal classes, traits cannot have constructor parameters. Furthermore, no constructor arguments are passed to the superclass of the trait. This is not necessary as traits are initialized after the superclass is initialized.
Assume a trait D defines some aspect of an instance x of type C (i.e. D is a base class of C). Then the actual supertype of D in x is the compound type consisting of all the base classes in L(C) that succeed D.
This is needed to define the base constructor with no-parameters.
3) As per answer (2), it's needed to define the base constructor

Why do we need traits in scala?

So, I was trying to make a finagle server, talk to sentry (not important), and stumbled upon a case, where I needed to inherit from two classes (not traits) at the same time, let's call them class SentryHandler extends Handler and class TwitterHandler extends Handler, and assume, that I need to create MyHandler, that inherits from both of them.
After a moment of stupidity, when I thought it was impossible without using a dreaded "delegation pattern", I found a solution:
trait SentryTrait extends SentryHandler
class MyHandler extends TwitterHandler with SentryTrait
Now, this got me thinking: what is the purpose of having the notion of "trait" to being with? If the idea was to enforce that you can inherit from multiple traits but only a single class, it seems awfully easy to get around. It kinda sounds like class is supposed to be the "main" line of inheritance (that you "extend a class with traits", but that isn't true either: you can extend a trait with (or without) a bunch of other traits, and no class at all.
You cannot instantiate a trait, but the same holds for an abstract class ...
The only real difference I can think of is that a trait cannot have constructor parameters. But what is the significance of that?
I mean, why not? What would the problem with something like this?
class Foo(bar: String, baz: String) extends Bar(bar) with Baz(baz)
Your solution (if I understood correctly) - doesn't work. You cannot multiinherit classes in scala:
scala> class Handler
defined class Handler
scala> class SentryHandler extends Handler
defined class SentryHandler
scala> class TwitterHandler extends Handler
defined class TwitterHandler
scala> trait SentryTrait extends SentryHandler
defined trait SentryTrait
scala> class MyHandler extends TwitterHandler with SentryTrait
<console>:11: error: illegal inheritance; superclass TwitterHandler
is not a subclass of the superclass SentryHandler
of the mixin trait SentryTrait
class MyHandler extends TwitterHandler with SentryTrait
As for the question - why traits, as I see it, this is because traits are stackable in order to solve the famous diamond problem
trait Base { def x: Unit = () }
trait A extends Base { override def x: Unit = { println("A"); super.x}}
trait B extends Base { override def x: Unit = { println("B"); super.x}}
class T1 extends A with B {}
class T2 extends B with A {}
(new T1).x // Outputs B then A
(new T2).x // Outputs A then B
Even though trait A super is Base (for T1) it calls B implementation rather then Base. This is due to trait linearization
So for classes if you extend something - you can be sure that this base will be called next. But this is not true for traits. And that's probably why you do not have trait constructor parameters
The question should rather be: why do we need classes in Scala? Martin Odersky has said that Scala could get by with just traits. We would need to add constructors to traits, so that instances of traits can be constructed. That's okay, Odersky has said that he has worked out a linearization algorithm for trait constructors.
The real purpose is platform interoperability.
Several of the platforms Scala intends to integrate with (currently Java, formerly .NET, maybe in the future Cocoa/Core Foundation/Swift/Objective-C) have a distinct notion of classes, and it is not always easy to have a 1:1 mapping between Scala traits and platform classes. This is different, for example, from interfaces: there is a trivial mapping between platform interfaces and Scala traits – a trait with only abstract members is isomorphic to an interface.
Classes, packages, and null are some examples of Scala features whose main purpose is platform integration.
The Scala designers try very hard to keep the language small, simple, and orthogonal. But Scala is also explicitly intended to integrate well with existing platforms. In fact, even though Scala is a fine language in itself, it was specifically designed as a replacement for the major platform languages (Java on the Java platform, C# on the .NET platform). And in order to do that, some compromises have to be made:
Scala has classes, even though they are redundant with traits (assuming we add constructors to traits), because it's easy to map Scala classes to platform classes and almost impossible to map traits to platform classes. Just look at the hoops Scala has to jump through to compile traits to efficient JVM bytecode. (For every trait there is an interface which contains the API and a static class which contains the methods. For every class the trait is mixed into, a forwarder class is generated that forwards the method calls to trait methods to the static class belonging to that trait.)
Scala has packages, even though they are redundant with objects. Scala packages can be trivially mapped to Java packages and .NET namespaces. Objects can't.
Package Objects are a way to overcome some of the limitations of packages, if we didn't have packages, we wouldn't need package objects.
Type Erasure. It is perfectly possible to keep generic types around when compiling to the JVM, e.g. you could store them in annotations. But third-party Java libraries will have their types erased anyway, and other languages won't understand the annotations and treat Scala types as erased, too, so you have to deal with Type Erasure anyway, and if you have to do it anyway, then why do both?
null, of course. It is just not possible to automatically map between null and Option in any sane way, when interoperating with real-world Java code. You have to have null in Scala, even though we rather wished it weren't there.
The problem with having constructors and state in a trait (which then makes it a class) is with multiple inheritance. While this is technically possible in a hypothetical language, it is terrible for language definition and for understanding the program code. The diamond problem, mentioned in other responses to this question), causes the highest level base class constructor to be called twice (the constructor of A in the example below).
Consider this code in a Scala-like language that allows multiple inheritance:
Class A(val x: Int)
class B extends A(1)
class C extends A(2)
class D extends B, C
If state is included, then you have to have two copies of the value x in class A. So you have two copies of class A (or one copy and the diamond problem - so called due to the diamond shape of the UML inheritance diagram).
Diamond Multiple Inheritance
The early versions of the C++ compiler (called C-Front) had lots of bugs with this and the compiler or the compiled code often crashed handling them. Issues include if you have a reference to B or C, how do you (the compiler, actually) determine the start of the object? The compiler needs to know that in order to cast the object from the Base type (in the image below, or A in the image above) to the Descendant type (D in the image above).
Multiple Inheritance Memory Layout
But, does this apply to traits? The way I understand it, Traits are an easy way to implement composition using the Delegation Pattern (I assume you all know the GoF patterns). When we implement Delegation in any other language (Java, C++, C#), we keep a reference to the other object and delegate a message to it by calling the method in its class. If traits are implemented in Scala internally by simply keeping a reference and calling its method, then traits do exactly the same thing as Delegation. So, why can't it have a constructor? I think it should be able to have one without violating its intent.
The only real difference I can think of is that a trait cannot have constructor parameters. But what is the significance of that? I mean, why not?
Consider
trait A(val x: Int)
trait B extends A(1)
trait C extends A(2)
class D extends B with C
What should (new D {}).x be? Note: there are plans to add trait parameters in Scala 3, but still with restrictions, so that the above is not allowed.

What is a sealed trait?

Sealed classes are described in 'Programming in Scala', but sealed traits are not.
Where can I find more information about a sealed trait?
I would like to know, if a sealed trait is the same as a sealed class?
Or, if not, what are the differences?
When is it a good idea to use a sealed trait (and when not)?
A sealed trait can be extended only in the same file as its declaration.
They are often used to provide an alternative to enums. Since they can be only extended in a single file, the compiler knows every possible subtypes and can reason about it.
For instance with the declaration:
sealed trait Answer
case object Yes extends Answer
case object No extends Answer
The compiler will emit a warning if a match is not exhaustive:
scala> val x: Answer = Yes
x: Answer = Yes
scala> x match {
| case No => println("No")
| }
<console>:12: warning: match is not exhaustive!
missing combination Yes
So you should use sealed traits (or sealed abstract class) if the number of possible subtypes is finite and known in advance. For more examples you can have a look at list and option implementations.
a sealed trait is the same as a sealed class ?
As far as sealed goes, yes. They share the normal differences between trait and class, of course.
Or, if not, what are the differences ?
Moot.
When is it a good idea to use a sealed trait (and when not) ?
If you have a sealed class X, then you have to check for X as well as any subclasses. The same is not true of sealed abstract class X or sealed trait X. So you could do sealed abstract class X, but that's way more verbose than just trait and for little advantage.
The main advantage of using an abstract class over a trait is that it can receive parameters. That advantage is particularly relevant when using type classes. Let's say you want to build a sorted tree, for instance. You can write this:
sealed abstract class Tree[T : Ordering]
but you cannot do this:
sealed trait Tree[T : Ordering]
since context bounds (and view bounds) are implemented with implicit parameters. Given that traits can't receive parameters, you can't do that.
Personally, I prefer sealed trait and use it unless some particular reason makes me use a sealed abstract class. And I'm not talking about subtle reasons, but in-your-face reasons you cannot ignore, such as using type classes.
From the daily-scala blog:
When a trait is "sealed" all of its subclasses are declared within the
same file and that makes the set of subclasses finite which allows
certain compiler checks.
Also I feel the need to point you to the specifications:
The sealed modifier applies to class definitions. A sealed class may not be directly inherited, except if the inheriting template is defined in the same source
file as the inherited class. However, subclasses of a sealed class can be inherited anywhere.
— M. Odersky. The Scala language specification, version 2.8. online, Sept., 2013.
‌‌Briefly:
Sealed traits can only be extended in the same file
List this lets the compiler easily know all possible subtypes
Use sealed traits when the number of possibly subtypes is finite and known in advance
A way of creating something like enum in Java
Help to define algebraic data types (ADTs)
and for more details
Everything about sealed traits in Scala
Traits can also be defined sealed, and only extended by a fixed set of case classes.
The core difference between normal traits and sealed traits can be summarized as follows:
Normal traits are open, so any number of classes can inherit from the trait as long as they provide
all the required methods, and instances of those classes can be used interchangeably via the trait's
required methods.
A normal trait hierarchy makes it easy to add additional sub-classes: just define your class and
implement the necessary methods. However, it makes it difficult to add new methods: a new
method needs to be added to all existing subclasses, of which there may be many.
Sealed traits are closed: they only allow a fixed set of classes to inherit from them, and all
inheriting classes must be defined together with the trait itself in the same file or REPL command.
A sealed trait hierarchy is the opposite: it is easy to add new methods, since a new method can
simply pattern match on each sub-class and decide what it wants to do for each. However, adding
new sub-classes is difficult, as you need to go to all existing pattern matches and add the case to
handle your new sub-class.
As an example
object SealedTraits extends App{
sealed trait Point
case class Point2D(x: Double, y: Double) extends Point
case class Point3D(x: Double, y: Double, z: Double) extends Point
def hypotenuse(p: Point) = p match {
case Point2D(x, y) => math.sqrt(x x + y y)
case Point3D(x, y, z) => math.sqrt(x x + y y + z z)
}
val points: Array[Point] = Array(Point2D(1, 2), Point3D(4, 5, 6))
for (p <- points) println(hypotenuse(p))
// 2.23606797749979
// 8.774964387392123
In general, sealed traits are good for modelling hierarchies where you expect the number of sub-classes
to change very little or not-at-all. A good example of something that can be modeled using sealed trait is
JSON.
A JSON value can only be JSON null, boolean, number, string, array, or dictionary.
JSON has not changed in 20 years, so it is unlikely that anyone will need to extend our JSON
with additional subclasses.
While the set of sub-classes is fixed, the range of operations we may want to do on a JSON blob is
unbounded: parse it, serialize it, pretty-print it, minify it, sanitize it, etc.
Thus it makes sense to model a JSON data structure as a closed sealed trait
hierarchy rather than a normal open trait hierarchy.
sealed trait Json
case class Null() extends Json
case class Bool(value: Boolean) extends Json
case class Str(value: String) extends Json
case class Num(value: Double) extends Json
case class Arr(value: Seq[Json]) extends Json
case class Dict(value: Map[String, Json]) extends Json

What is the advantage of using abstract classes instead of traits?

What is the advantage of using an abstract class instead of a trait (apart from performance)? It seems like abstract classes can be replaced by traits in most cases.
I can think of two differences
Abstract classes can have constructor parameters as well as type parameters. Traits can have only type parameters. There was some discussion that in future even traits can have constructor parameters
Abstract classes are fully interoperable with Java. You can call them from Java code without any wrappers. Traits are fully interoperable only if they do not contain any implementation code
There's a section in Programming in Scala called "To trait, or not to trait?" which addresses this question. Since the 1st ed is available online, I'm hoping it's OK to quote the whole thing here. (Any serious Scala programmer should buy the book):
Whenever you implement a reusable collection of behavior, you will
have to decide whether you want to use a trait or an abstract class.
There is no firm rule, but this section contains a few guidelines to
consider.
If the behavior will not be reused, then make it a concrete class. It
is not reusable behavior after all.
If it might be reused in multiple, unrelated classes, make it a trait.
Only traits can be mixed into different parts of the class hierarchy.
If you want to inherit from it in Java code, use an abstract class.
Since traits with code do not have a close Java analog, it tends to be
awkward to inherit from a trait in a Java class. Inheriting from a
Scala class, meanwhile, is exactly like inheriting from a Java class.
As one exception, a Scala trait with only abstract members translates
directly to a Java interface, so you should feel free to define such
traits even if you expect Java code to inherit from it. See Chapter 29
for more information on working with Java and Scala together.
If you plan to distribute it in compiled form, and you expect outside
groups to write classes inheriting from it, you might lean towards
using an abstract class. The issue is that when a trait gains or loses
a member, any classes that inherit from it must be recompiled, even if
they have not changed. If outside clients will only call into the
behavior, instead of inheriting from it, then using a trait is fine.
If efficiency is very important, lean towards using a class. Most Java
runtimes make a virtual method invocation of a class member a faster
operation than an interface method invocation. Traits get compiled to
interfaces and therefore may pay a slight performance overhead.
However, you should make this choice only if you know that the trait
in question constitutes a performance bottleneck and have evidence
that using a class instead actually solves the problem.
If you still do not know, after considering the above, then start by
making it as a trait. You can always change it later, and in general
using a trait keeps more options open.
As #Mushtaq Ahmed mentioned, a trait cannot have any parameters passed to the primary constructor of a class.
Another difference is the treatment of super.
The other difference between classes and traits is that whereas in classes, super calls are statically bound, in traits, they are dynamically bound. If you write super.toString in a class, you know exactly which method implementation will be invoked. When you write the same thing in a trait, however, the method implementation to invoke for the super call is undefined when you define the trait.
See the rest of Chapter 12 for more details.
Edit 1 (2013):
There is a subtle difference in the way abstract classes behaves compared to traits. One of the linearization rules is that it preserves the inheritance hierarchy of the classes, which tends to push abstract classes later in the chain while traits can happily be mixed in. In certain circumstances, it's actually preferable to be in latter position of the class linearization, so abstract classes could be used for that. See constraining class linearization (mixin order) in Scala.
Edit 2 (2018):
As of Scala 2.12, trait's binary compatibility behavior has changed. Prior to 2.12, adding or removing a member to the trait required recompilation of all classes that inherit the trait, even if the classes have not changed. This is due to the way traits were encoded in JVM.
As of Scala 2.12, traits compile to Java interfaces, so the requirement has relaxed a bit. If the trait does any of the following, its subclasses still require recompilation:
defining fields (val or var, but a constant is ok – final val without result type)
calling super
initializer statements in the body
extending a class
relying on linearization to find implementations in the right supertrait
But if the trait does not, you can now update it without breaking binary compatibility.
For whatever it is worth, Odersky et al's Programming in Scala recommends that, when you doubt, you use traits. You can always change them into abstract classes later on if needed.
Other than the fact that you cannot directly extend multiple abstract classes, but you can mixin multiple traits into a class, it's worth mentioning that traits are stackable, since super calls in a trait are dynamically bound (it is referring a class or trait mixed before current one).
From Thomas's answer in Difference between Abstract Class and Trait:
trait A{
def a = 1
}
trait X extends A{
override def a = {
println("X")
super.a
}
}
trait Y extends A{
override def a = {
println("Y")
super.a
}
}
scala> val xy = new AnyRef with X with Y
xy: java.lang.Object with X with Y = $anon$1#6e9b6a
scala> xy.a
Y
X
res0: Int = 1
scala> val yx = new AnyRef with Y with X
yx: java.lang.Object with Y with X = $anon$1#188c838
scala> yx.a
X
Y
res1: Int = 1
When extending an abstract class, this shows that the subclass is of a similar kind. This is not neccessarily the case when using traits, I think.
In Programming Scala the authors say that abstract classes make a classical object oriented "is-a" relationship while traits are a scala-way of composition.
Abstract classes can contain behaviour - They can parameterized with constructor args (which traits can't) and represent a working entity. Traits instead just represent a single feature, an interface of one functionality.
A class can inherit from multiple traits but only one abstract class.
Abstract classes can have constructor parameters as well as type parameters. Traits can have only type parameters. For example, you can’t say trait t(i: Int) { }; the i parameter is illegal.
Abstract classes are fully interoperable with Java. You can call them from Java code without any wrappers. Traits are fully interoperable only if they do not contain any implementation code.