Like in many languages, Kermeta provides two ways to add comments in your models:

Kermeta provides a way to define named and unnamed annotations, that have to be defined just above any model element among Package, ClassDefinition, Property, Operation. Such annotations correspond to MOF tags, and are linked to the elements which immediately follows.

To define a named annotation, you have to use a special symbol "@", whereas an anonymous annotation has to be written between /** and */

Example 1: you can define an annotation to describe the role of a property

@usage "count the number of ..."
reference myCounter : Integer 

Example 2: you can document your classes and operation using /** ... */

/** * This is a documentation tag for the class "myClass" */
class MyClass {
    /** This is a documentation tag for myOperation */
    operation myOperation() is do
        // Unlinked comment
    end
    @desc "This is a named annotation for thisOperation"
    operation thisOperation() is do
        /* This is an unlinked comment */
    end
}

Kermeta provides a block notion to manage scope of variable. Instruction of the block have to be inserted between do and end keywords. Theses two keywords may be omitted for the conditional and for the loop structures.

A variable could only be accessed in the block where it was defined and in its sub blocks:

do
    var v1 : Integer init 3
    var v2 : Integer init 2
    do
        var v3 : Integer
        v3 := v1 + v2

        var v2 : Integer // error : v2 is already declared in the upper block
    end
    var v4 : Integer init v3 // error : v3 is unknown here
end

Kermeta's conditional statement is composed of at least two elements : a boolean expression and a block that is executed if the boolean is evaluated to true. You can add a third element, with the else keyword, that is executed if the boolean expression is evaluated to false.

Example 1: if..then..else block

var v1 : Integer init 2
var v2 : String init "blah"

if v1 > 5 then v1 := v1-5

if v1 == 2 then
    v2 := v1
    v1 := v2 + v1
else
    v1 := 0
end

The if statement is an expression (see Chapter 3, Kermeta Metamodel). As any expression in Kermeta, it can return a value. The return type of the if statement must be a super type of the values "returned" by both then and else blocks (otherwise the type checker will send an error). The values considered as the result of the evaluation (the "returned" values) of the if statement are the last evaluated statement inside then or else block

Example 2: conditional is an expression

var s : String
    s := if false then "a" else "b" end

Example 3: a more complex conditional

var x : String
    var y : Integer init 5 
    x := if y < 3 then 
         stdio.writeln("hello")
         "a" 
          else 
         "b"
         "c" 
          end // The String "c" will be the value of x

Here is a sample of a typical loop in Kermeta.

var v1 : Integer init 3
var v2 : Integer init 6

from var i : Integer init 0
until i == 10
loop
    i := i + 1
end

	Note
	Unlike Java, there is no exit, break or continue function in Kermeta.

See Section 2.14, “Collections” for functions offering iterator-like scanning.

When you need to refer explicitly another entity defined in another file, you have to use the require primitive. It allows loading definitions of an other Kermeta file when file is processed. In the following example, we define a class C which inherits from the A class previously defined.

// file : MyPackage-part1.kmt
package subPackage1;

class A 
{
// ...
}

// file : MyPackage-part2.kmt
package MyPackage;
require "MyPackage-part1.kmt"

class C inherits subPackage1::A 
{
// ...
}

	Note
	In most case, the order of the declaration in not important. The only exception is a very rare situation related to aspects (see Section 2.20, “Weaving kermeta code” ).

	Note
	You can create cyclic dependencies of files, the environment will deal with that.

You can require different kind of entity

Obvioulsy, you can require Kermeta textual syntax, ie. *.kmt files

You can require Kermeta models, ie. *.km files.

You can require Ecore models, ie. *.ecore files. These models can then be used as if they were written in Kermeta. In order to add behavior to the classes defined in .ecore files you may : use the weaving to dynamically weave behavior to the ecore, or roundtrip using the button action in the workbench (ecore->kmt->ecore) to statically add the behavior (as EAnnotations) into the ecore file.

A variant of requiring an ecore file is to require registered EPackage. When Eclipse deploys an ecore model plugin, it also registers the EPackage using a unique identifier (nsuri) in order to retreive it quickly. In Kermeta you can also use this nsuri into the require statement. This approach is useful becauseyou can be sure that you require the very same model as eclispe use to load and save models of that sort. For example, instead of requiring ecore.ecore you may use require "http://www.eclipse.org/emf/2002/Ecore". This also works for all other registered metamodels (Your own registered metamodel, UML metamodel, etc ...). Kermeta user interface provides a view that display all registered EPackages and their nsuri.

Warning

A special attention must be put when requiring resources. It is not allowed to get several versions of the same class that comes from several required files. Kermeta cannot know which version it must use, so you'll get an error. A typical error, is to require both an ecore file and the registered version of it. They represent two distinct resources in memory. A more clever error, is when the two versions are hidden inside an erroneous ecore file which uses several ways to access some metaclass and actually is not consistent... (ex: you find references to both http://www.eclipse.org/emf/2002/Ecore and platform:/plugin/org.eclipse.emf.ecore/model/Ecore.ecore in the ecore file or one of the dependent files)

In order to use the seamless java import (still prototype in v0.4.2), you can require a jar file. It will automatically convert the java into a Kermeta representation in order to be able to call java code as if it was written in Kermeta. see Section 2.22, “Using existing java code in Kermeta” for more details.

Kermeta lets you organise your code the way you prefer. There is no predefined organisation of your code, but the language and the workbench proposes various mecanisms and tools to structure it according to your needs and the style you wish for your code. Typically if you use kermeta internal weaver (see Section 2.20, “Weaving kermeta code”), manually transform your ecore into Kermeta, eventually weaving the ecore using the merge button available on the worbench.

For example, you can put all your classes into a single file (like in Ecore, every classes is in the .ecore file) or you may create a file for each of them and then use the require to indicates that they must know each other You can also use an organisation like java, with a file per class and a directory per package...

Kermeta implements a few primitive types. By primitive types, we mean types that are basic to any language, i.e integer, and that have a "literal" value. See below table.

// Simple datatypes examples
var myVar1 : Integer init 10
var myVar2 : Integer
var myVar4 : String init "a new string value"
var myVar5 : boolean

Kermeta also supports some other primitive types like Float and Character, but they currently don't have a surface syntax for their literals. The way to get them is to use the convertion methods available on String (for both of them) or Integer (for Float).

For example:

var c : Character init "f".elementAt(0)
var r : Real init "4.5".toReal

You can define enumerations using the following syntax. Note that each enumeration literal must end with a ";".

enumeration Size { small; normal; big; huge; } 

You can manipulate enumerated variables and literals with classical operators such as in the following example.

var mySize : Size
if ( mySize == Size.small ) then 
    stdio.writeln("This is small !") 
end

	Note
	Enumeration is a concept of the same level as Class, they must both be defined in a package.

In Kermeta, you can define your own datatype based on existing types without a hard dependency like inheritance.

This is done using the alias syntax.

Ex:

alias MyString : kermeta::standard::String;

In this sample, MyString has all the behavior of kermeta::standard::String but is declared locally. This means that you don't have any dependency to the framework, even to the String in the framework.

Obviously you can reuse existing type names :

alias String : kermeta::standard::String;

This will create a new type String in your package with the behavior of String in the framework.

The definition of an alias is different from the use of "using" statement (as defined in Section 2.13.3, “Syntaxic sugars”), when you write

using kermeta::standard

you simply defined a syntactical shortcut allowing you to access any definition in this package from within this file.

Wheras defining an alias allows you to access this new definition from another package if needed.

	Tip
	It is interesting to redefine your own datatype for all the standard type you use in your metamodel, so when you convert the file into ecore in order to have serialisation, you won't have any dependency to framework.ecore (which is the ecore version of the framework where kermeta standard type) This allow a lazy coupling of the type definitions.

As introduced in the "Hello world" example (see Section 1.2.1, “First Program”), Kermeta is an object-oriented language. Kermeta provides all MOF concepts like properties, attributes, package. In addition, it provides a body to operations and derived properties.

Classes and abstract classes can be defined in a Java-like way. Class definition must be placed into brackets as it is in the following example.

// an empty simple class
class myFirstClass 
{

}

// a simple abstract class
abstract class MyAbstractClass
{

}

Additionally, for better code robustness, and more user-friendly programming, classes can use the genericity mechanisms (see Section 2.10, “Genericity” for more information).

// This is a parametric class
class A<G> {
}

// This is the type variable binding : G is binded with Integer
var a : A<Integer>
a := A<Integer>.new

There are some limitations in regards to Java. For example, you cannot define nested classes in Kermeta. Kermeta offers the same structural concepts than MOF language.

Kermeta provides a way to add operational (action) semantics to your metamodels. For that, you can define operations with their body, as in a classical programming language. You can also specify abstract operations (they have no body). Kermeta requires that every class that contains an abstract operation must be declared as an abstract class.

In the following example, we define some operations, based on a visitor pattern implementation

class Inventory 
{
    operation visitLibrary(l : Library) is
    do
        writeln("Inventory : ");
        l.books.each(b : Book | b.accept(self))
        writeln("----")
    end

    operation visitBook(b : Book) is
    do
        stdio.write("Book : ", b.title, " by ")
        b.authors.each(a : Author | a.accept(self))
    end

    operation visitAuthor(a : Author) is 
    do
        stdio.write(a.name, " ", a.lastName)
    end
}

class Library {
    // ...
    operation accept(visitor : Inventory) is
    do
        visitor.visitLibrary(self)
    end
}

class Book
{
    // ...
    operation accept(visitor : Inventory) is
    do
        visitor.visitBook(self)
    end
}

class Author
{
    // ...
    operation accept(visitor : Inventory) is
    do
        visitor.visitAuthor(self)
    end
}

In this small example we define an Inventory class which can go over the library structure and print books informations. For that, we apply the visitor GoF pattern's on the library structure defining an accept method in every library structures.

2.8.2.1. Result

The special variable result is used to store the value that will be returned by the operation.

operation getName() : String is
    do
        result := self.name
    end

	Note
	This is different of the return in java, since it doesn't end the block. Other instructions can be used after the result assignment.

class A{
    operation main0() is do
        // do something
    end
    operation main1( arg1 : String) is do
        // do something with 1rst argument
    end
    operation main3( arg1 : String, arg2 : String) is do
        // do something with 1st and 2nd arguments
    end
}
// If you ask to launch main0, kermeta interpereter will create an instance of A and will run main0 on it.

Kermeta doesn't use constructors.

However in some situation, you may need to provide some initialization operation. The best approach is simply to declare an operation which will return self after having done the initialization work.

class A
{
	attribute name : String
	operation initialize(name : String) : A is do
		self.name := name
	end
}        
        

// now you can use it easily in one line
	var aA : A init A.new.initialize("hello")
        

Here are some explanation about some design choice of Kermeta classes.

abstract class Person
{
    attribute name : string
    attribute lastName : string

    reference father : Male#children 
    reference mother : Female#children
}

class Male inherits Person
{
    reference children : oset Person[0..*]#father
    reference wife : Female[0..1]#husband
}

class Female inherits Person
{
    reference children : oset Person[0..*]#mother
    reference husband : Male[0..1]#wife
}

In this example, we define a simple model which represent simples family trees. Here, persons are defined by their name and last name. Each person have a father and mother and respectively might have children. The figure representing this example is Figure 2.1, “A simple family tree model”

Figure 2.1. A simple family tree model

As Kermeta is strongly typed, you cannot assign or pass something of the wrong type.

For example :

class A {
}
class SubA inherits A {
}
class AnotherSubA inherits A {
}

// ...
	var aA    : A init SubA.new
	var aSubA : SubA
	aSubA := aA  // doesn't work because not type safe
        

In this situation you must use one of the following methods : conditionnal assignment or asType

	aSubA ?= aA  // works
	var aAnotherSubA : AnotherSubA
	aAnotherSubA ?= aA  // works too, but has been assigned Void and not the value ...
        	

class B {
}
// ...	
	var aB : B
	aB ?= aA  // doesn't work because not type safe
        	

class C {
	operation needASubA(sa : SubA) : Void	is do
		// do something
	end
}
	// ... typical code using ?= 
	var aC : C init C.new	
	aC.needASubA(aA.asType(SubA))
        	

class Parent
{
     reference children : oset Child[0..*]#parent
}

class Child
{
     reference parent : Parent#children

}

class Male { }

class GrandFather inherits Parent, Male
{
     boolean healthy : Boolean
}

The above example defines a class "GrandFather" that inherits at the same time the class "Parent", and the class "Male". Its graphical representation is shown in below figure.

Figure 2.2. : multiple inheritance

In the following sample, when an operation is declared for the first time (in the parent class Person), it uses the operation keyword. Then, whenever you override it, you will have to use the method keyword.

	Note
	In the sample,Person and adopt are abstract, but this has no influence on the operation-method keywords rule, it would have been the same even if Person was not abstract or if adopt had a behavior in this class.

abstract class Person
{
    attribute name : String
    reference father : Male#children 
    reference mother : Female#children
    operation adopt(child : Person) is abstract}

class Male inherits Person 
{
    reference wife : Female#husband
    reference children : oset Person[0..*]#father    method adopt(child : Person) is    do        children.add(child)
        if not wife.children.contains(child) then            child.father := self            wife.adopt(child)
        end    end}

class Female inherits Person
{
    reference husband : Male#wife
    reference children : oset Person[0..*]#mother    method adopt(child : Person) is    do        children.add(child)
        if not husband.children.contains(child) then            child.mother := self
            husband.adopt(child)
        end    end}

A MOF class can have operations but MOF does not provide any way to describe the behavior of these operations. Furthermore MOF does not provide any semantics neither for operation call nor for operation inheritance and redefinition. This section investigates how, while weaving actions into MOF, MOF semantics can be extended to support behavior definition and extension mechanisms provided by the action language. This implies answering several questions concerning redefinition and dispatch.

2.9.5.1. Operation redefinition

MOF does not specify the notion of overriding an operation because from a structural point of view it does not make any sense. To stick to MOF structure one can argue that redefinition should be forbidden in an executable MOF. This is the simplest solution as it also solves the problem of the dynamic dispatch since a simple static binding policy can be used.

However, operation redefinition is one of the key features of Object-Oriented (OO) languages. The OO paradigm has demonstrated that operation redefinition is a useful and powerful mechanism to define the behavior of objects and allow for variability. This would be very convenient to properly model dynamic semantic variation points existing in e.g. UML state-charts. For this reason we believe that an important feature of an executable MOF is to provide a precise behavior redefinition mechanism. The choice of the operation overriding mechanism must take into account the usual problem of redefinition such as method specialization and conflicting redefinitions related to multiple inheritance.

class A {     operation m1() is do         // Some behavior     end     // operation m2 is abstract     operation m2() is abstract } class B inherits A {         // method m1 inherits operation m1 from A     method m1() is do         // Behavior redefinition     end     method m2() is do // Implementation of the abstract method     end }

Table 2.1. Operation redefinition in Kermeta

	Note
	Notice in that sample that method redefinition uses the method keyword instead of operation.

class A
{
    method m(i : Integer) is do
        // [...]
    end 
    method m(s : String) is do // this is not allowed in Kermeta !!!
        // [...]
    end
}

2.9.5.4. Conflicts related to multiple inheritance

This is also a classical problem that has been solved in several OO languages. There are mainly two kinds of conflicts when a class inherits features from several super-classes:

• Several features with the same name might be inherited from different super classes causing a name clash.

• Several implementations of a single operation could be inherited from different super classes.

There are two kinds of solutions to resolve these conflicts. The first one is to have an implicit resolution mechanism which chooses the method to inherit according to an arbitrary policy. The second one is to include in the language constructions that allow the programmer to explicitly resolve conflicts. In Eiffel, for instance, the programmer can rename features in order to avoid name clashes and can select the method to inherit if several redefinition of an operation are inherited from parent classes.

In the current version of Kermeta, we have chosen to include a minimal selection mechanism that allows the user to explicitly select the inherited method to override if several implementations of an operation are inherited. This mechanism does not allow resolving some name clashes and thus reject some ambiguous programs. For the future version of Kermeta we plan to include a more general mechanism such as traits proposed by Schärli et al. In any case we believe the conflict resolution mechanism should be explicit for the programmer.

class O {     operation m() is abstract } class A inherits O {     method m() is do         // [...]     end } class B inherits O {     method m() is do         // [...]     end } class C inherits A, B { // "from" : an explicit selection of the//implementation to inherit     method m() from A is do         // [...]     end }

Table 2.2. Explicit selection of super operation in Kermeta

In Kermeta classes can have a set of type parameters. These type variables can be used in the implementation of the class as any other type. By default a type variable can take as value any type; but a type variable can also be constrained by a type: in that case, this type variable can only be substituted by a sub-type of this type. The following code demonstrates how to create generic classes.

// A class with a type variable G that can be bound with any type

class Queue<G>
{
     reference elements : oset G[*]

     operation enqueue(e : G) : Void is do 
         elements.add(e)
     end
     
     operation dequeue() : G is do
         result := elements.first
         elements.removeAt(0)
     end
}

// A class with a type variable C that can be bound with any sub-type of Comparable

class SortedQueue<C : Comparable> inherits Queue<C> 
{
     method enqueue(e : C) : Void is do
         var i : Integer
         from i := 0
         until i == elements.size or e > elements.elementAt(i)
         loop
             i := i + 1
         end
         elements.addAt(i, e)
     end
}

Kermeta operations can contain type parameters. Like type variables for classes these type parameters can be constrained by a super type. However, unlike for classes, for which the bindings to these type parameters are explicit, for operations the actual type to bind to the variable is statically inferred for each call according to the type of the actual parameters.

class Utils {
    operation max<T : Comparable>(a : T, b : T) : T is do
        result := if a > b then a else b end
    end
}

	Note
	Notice in that sample that even the "if" is an expression that can return a value that is assigned here to the special variable "result".

Actually, all types can be used as a parameter of a generic class or generic operation. More, the ModelType, is really useful when combined with generics. see Section 2.21, “Model type”

The user will refer to eclipse documentation for the creation of an EMF model from its ECore meta-model. We suggest to use the Wizard samples to create, at the first hand, an Ecore meta-model, and then, at the second hand, instances of this Ecore meta-model, using the generated EMF reflexive editors.

Once you created the ECore meta-model, please check that you correctly filled the property "Ns URI" of the root package of the Ecore meta-model, otherwise the resource load will fail. This NsURI must equal the real path of your metamodel. (You can modify this property through the Properties View of your meta-model)

In the current version of the EMF resource loader, you have to prepare your EMF Resource following these rules :

@mainClass "root::TestCSLoading"
@mainOperation "main"

package root;

require kermeta
require "cs.ecore"

using kermeta::standard
using kermeta::persistence

class TestCSLoading
{
     operation initialize(uri : String, mm_uri : String) : Set<Object> is do
         /* Initialize the EMF repository */
         var repository     : EMFRepository init EMFRepository.new
         /* Create an EMF Resource */
         var resource : Resource init repository.createResource(uri, mm_uri)
         /* Load the resource */
         resource.load
         /* Get the loaded __root__ instances (a Set<Object>) */
         result := resource  // a resource is a collection of objects contained
     end
 

operation main() is do 
   var instances := self.initialize("./test.cs", "./cs.ecore") 
   from var it : Iterator<Object> init instances.iterator 
   until it.isOff 
   loop
      var o : Object o := it.next 
      if (o == void) then stdio.writeln("Void object!") 
      else 
         stdio.writeln("---------------------------------") 
         stdio.writeln("Objet : " + o.getMetaClass.classDefinition.qualifiedName 
            + " ( " + o.getMetaClass.typeDefinition.ownedAttribute.size.toString+ "attr.)" ) 
      end
      var template : cs::Template // Print instances which type is cs::Template 
      if (cs::Template.isInstance(o)) 
      then 
         template ?= o 
         stdio.writeln(" name : " + template.name) 
         stdio.writeln(" decision : " + template.decision.toString) 
         stdio.writeln(" content : " + template.content) stdio.writeln(" referer : " + template.referer.toString) 
      end
      // Print instances which type is cs::Decision 
      if (cs::Decision.isInstance(o)) 
      then
         decision ?= o 
         stdio.writeln(" name : " + decision.name) 
      end
   end
}

If your resource is dependent of other resources and that EMF succed to load it, the Repository that was used to load your resource will automatically load all these dependent resources.

TODO : give an example.

Kermeta provides package to structure models. So you can define packages and sub-package as deep as you want. There are two main ways to do this, as shown in examples 2 and 3 (below). If you want to define some classes in a package you may define them in a specific file and start this file with a package naming directive (followed by a semi-colon) like in the following example^[2].

Example 1: 1 file.kmt == 1 package naming directive

// My file
package MyNewPackage;

class A { 
// ... 
}

class B {
// ... 
}

Here, classes A and B are defined in a package called "MyNewPackage". All classes defined in this file are under the scope of this package.

You can also define explicitly sub-packages using braces (see the following example):

Example 2: Defining subpackages using braces

// file : MyPackage-part1.kmt
package MyPackage;

package subPackage1 
{
    class A 
    {
// ...
    }
}

package subPackage2
{
    class B
    {
// ...
    }
}

In this example, we define a main package called "MyPackage" which contains 2 sub-packages "subPackage1" and "subPackage2". The first one contains the A class and the second one the B class.

TODO : example of use inside code (without the "using" primitive)

If you want, you can declare the same package in several files. You can also directly define subpackages in the package naming directive of your file (see example 1 above). In the following example, we define a new sub-package of "MyPackage" called "subPackage3" directly in the file. All features defined in this file will belong to this sub-package.

Example 3: Defining subpackage using :: syntactic sugar

// file : subPackage3.kmt
package MyPackage::subPackage3;

class C 
{
// ...
}

If a given file, when you want to use a class definition that is not in the scope of the current package inside which you are working, you have to provide the full path of this class definition to be able to use it. However, you can also define shortcuts to make your code clearer: the using primitive provides such a shortcut. Taking the previous example, it becomes:

// file : MyPackage-part2.kmt
package MyPackage;

require "MyPackage-part1.kmt"

using subPackage1 // <- "shortcut"! 

class C inherits A 
{
// ...
}

The using statement is a syntactical shorcut, it has no conterpart in kermeta metamodel.

	Note
	The using statement is different from the definition of a local datatype with an alias. (See Local datatype using "alias" subsection )

Kermeta defines some collection data types handling sets of values. The different available collection types are the result of a combination of the constraints unique and ordered.

Figure 2.4. : The kermeta collections

Another way to define set of objects would have been to use arrays. In fact, Kermeta does not define explicitly the concept of array, but it provides a multiplicity concept which can be used instead. Multiplicities are a way to define a lower and upper bound on a collection. Syntactically, lower and upper bounds are defined between brackets and are separated by two dots. Bounds can be an integer literal value or a star to specify there's no upper bound.

Example 1: how to declare collections

using kermeta::standard		// don't need to specify it all the time
	// This is the simpliest and recommanded way of declaring a collection variable
	var myColA : Set<Integer> // this is equivalent to saying set Integer[0..*]
	var myColB : OrderedSet<Integer> // this is equivalent to saying oset Integer[0..*]
		

	// Collection with multiplicities
	var myCol1 : set Integer[4..6] // At least 4 elements and never more than 6
	var myCol3 : seq String[2..*] // At least two strings in the sequence
	var myCol4 : set String[1..*] // An non empty set
	var myCol5 : String[1..*] // If you don't specify any keyword, it is an ordered set
        

There is currently no way to define a collection by extension like you can do in C or Java. You must initialize your collection either by calling new (Kermeta constructor operation) on your collection type, or initialize by copy.

Example 2: initialize collections

// Example of declaration of variables as Collections. All those syntaxes are valid
var myCol1 : set Integer[0..*]    init kermeta::standard::Set<Integer>.new
// Fill in myCol1
myCol1.add(10)
myCol1.add(50)

var myCol2 : oset String[0..*]    init kermeta::standard::OrderedSet<String>.new
var myCol3 : bag Boolean[0..*]    init kermeta::standard::Bag<Boolean>.new
var myCol4 : seq Integer[0..*]    init kermeta::standard::Sequence<Integer>.new
// if no keyword specified, and multiplicity is set, it is an OrderedSet

var myCol4 : String[0..*]init kermeta::standard::OrderedSet<String>.new

var myCol1a : seq Integer[0..*]    init myCol1
var myCol2a : oset String[0..*]    init myCol2
var myCol3a : kermeta::standard::Bag<Boolean> init myCol3
var myCol3a : kermeta::standard::Sequence<Integer> init myCol4

	Note
	Conclusion : in most cases, you don't need to use this special syntax, you can simply use the generic collection names (Set<Something>, OrderedSet<Something>, etc.) available in Kermeta framework. Moreover, lower and upper bounds aren’t checked yet by Kermeta type checker and interpreter.

The collections in Kermeta implement several functions based on lambda expressions (see Section 2.17, “Lambda Expressions and functions”). These ones are very useful for model navigation.

Example 1: closure definitions for collection manipulation in Kermeta

aCollection.each { e | 
    /* do something with each element e of this collection */
}

aBoolean := aCollection.forAll { e | /* put here a condition */
} // return true if the condition is true for all elements in the collection.

aCollection2 := aCollection.select { e | 
     /* put here a condition that returns true for elements that must  be included in the resulting Collection */
}

aCollection2 := aCollection.reject { e | 
     /* put here a condition that returns true for elements that must  be exclude in the resulting Collection */
}

 // return a new collection which size is the same as in the original 
 // collection, and which element type is the type of the result of the expression.
aCollection2 := aCollection.collect { e | 
     /* put here an expression, for example e.name */
} 

anObject := aCollection.detect { e | /* a condition */} // returns an element (usually the first) that fulfill the condition.

aBoolean := aCollection.exists { e | /* a condition */} // returns true if at least one element fulfill the condition.

We introduce here the 2 first cases, which are relationships between two concrete entities.

class A { attribute x : set X[0..*] } 
class X {}

	Note
	Composition notion also relates to containment notion. So, there are some restriction about valid model. For example, in an association, only one end can be an attribute, otherwise this means that we have an association with a diamond on both end and we cannot say who is the container of the other.

class A { reference x : set X[0..*] } 
class X {}

Attributes, references and properties underlying collections may eventually be specialized. By default, they are represented by an OrderedSet. If you wish to be more precise and for exemple allow to have several item with the same value in your collection, you can use the collection modifiers as defined in Section 2.14, “Collections”

For example :

class A {
	attribute x1 : seq String[0..*]  // allows duplicates, ordered
	attribute x2 : set String[0..*]  // doesn't allow duplicates, not ordered        
	attribute x3 : bag String[0..*]  // allows duplicates, not ordered               
	attribute x4 : oset String[0..*] // (default if no modifier) doesn't allow duplicates, ordered        
}        

The opposite [property] of a property defines an association (bidirectional link) between two entities. An opposite is expressed by a sharp #. In the following example, a is the opposite property of the entity of type B, and b is mutually the opposite property of the entity of type A.

	Caution
	A property whose type is a DataType (i.e String, Boolean, Integer) cannot have an opposite!

Example 1: definition of an attribute, a reference, and an opposite property

This means that a can be accessed from b. Subsequently, if you modify the property b of an instance of A, it will mutually update its opposite property, i.e the property a of the instance of B contained in property b. You can make a test with the following example of code.

Example 2: navigability of opposite properties

var a1 : A init A.new
a1.name := "a1"
var b1 : B init C.new
a1.b := b1
stdio.writeln("b1 opposite : " + b1.a.name) // This prints "a1"!

The following paragraph shows a set of examples of attributes and references.

Figure 2.5. Attributes and references

Example 3: a set of attributes and references, with multiplicities and opposites.

TODO : add an example of code using those classes !! See https://gforge.inria.fr/plugins/scmcvs/cvsweb.php/integration_projects/other/org.openembedd.formations/Kermeta/Basics/docs/FR_formation_Kermeta_1er_niveau.odp?cvsroot=openembedd for such examples (p. 21).

package root;

class A1 {
    attribute b : B1[0..*] 
}
class B1 {}
			

class A2 {}
class B2 {
    reference a : A2 
}
			

class A3 {
    reference b : B3#a
}
class B3 {
    reference a : A3#b
}
			

class A4 {
    reference a : A4[0..*]
}

class A5 {
    attribute ab : A5[0..*]#aa
    reference aa : A5#ab
}

class A6 {
    attribute b : B6#a
}
class B6 {
    reference a : A6[1..1]#b
}

class A7 {
    attribute b : B7[0..*]#a
}
class B7 {
    reference a : A7#b 
}

class A8 {
    attribute b : B8[1..1]#a
}
class B8 {
    reference a : A8#b
}

class A9 {}
class B9 {
    reference a : A9[1..1]
}

	Note
	For every case where the upper bound of a property is upper to 1, the type of the property is OrderedSet . The reader will refer to Section 2.14, “Collections” (except the bag type) to have the available types for a [m..n](n>1) multiplicity property.

In a class definition, a derived property is a property that is derived or calculated, i.e it contains a body, like operations. Usually, such properties are calculated from other properties available from its owning class definition. In practice, you can define the code that you want inside a derived property.

In other words it does not reference to a concrete entity: it is calculated, through the accessor operations getter and setter.

The special parameter value is used in the setter to get the value passed when calling the setter.

Let's take the following class definitions :

// readonly property : it has no setter
class A 
{
    attribute period : Real
    property readonly frequency : Real // property : keyword for derived property
        getter is do
            result := 1/period
        end
}
// modifiable property :
class B 
{
    attribute period : Real
    property frequency : Real
        getter is do
            result result := 1/period
        end
        setter is do
            period := 1/value
        end
}

// a typical use would be (with aB an instance of class B) 
var freq : Real
fred := aB.frequency // to use the getter
aB.frequency := freq + 1 // to use the setter, the period is also updated in the process 

	Note
	To understand to role of the value keyword, you can imagine that in this sample the setter syntax is a shortcut syntax of : setter( value : Real) is do ... (even if actually this syntax isn't supported).

	Warning
	Since derived properties aims to behave like attributes or references, if the multiplicity is greater than 1, it doesn't make sense to define a setter. This is because it means reassigning the internal collection which is in fact calculated.

Properties are accessed or modified as classical properties. See next subsection for examples.

Example 1: let's take the example with A6 and B6 :

class A6 {
     attribute b : B6[0..*]#a
}

class B6 {
    reference a : A6#b 
}

var a6 : A6 init A6.new
var b6 : OrderedSet<B6>
// get the b attribute
// Note that as the attribute as a multiplicity >1 it is an OrderedSet 
b6 := a6.b

var a6 : A6 init A6.new
var b6 : B6 init B6.new
// add b6 to the attribute b of A. 
// Note : you don’t have to initialize b! done through A6.new
a6.b.add(b6)

// OrderedSet owns a method that removes an element given its
     // index in the set. For unordered sets, use "remove" method
a6.b.removeAt(0)
// Also valid : a6.b.remove(b6)

var a6 : A6 init A6.new
var b6 : B6 init B6.new
a6.b.add(b6)
// this assertion is true. Moreover, any instance in a6.b will
// have a6 as their opposite since b is a collection
assert(b6.a == a6)

var a6 : A6 init A6.new
var b6 : B6 init B6.new
// add ab6 to the attribute "b"
a6.b.add(b6)

var a6c : A6 init b6.container()
// this assertion is true
assert(a6c.equals(a6))

It is not different with references that have a [m..1] (m=0 or m=1) multiplicity:

Example 2: the A5 B5 case

class A5 {
    attribute b : B5#a // no multiplicity means [0..1]
}
class B5 {
    reference a : A5[1..1]#b
}

var a5 : A5 init A5.new
var b5 : B5
// get the b attribute
b5 := a5.b

var a5 : A5 init A5.new
var b5 : B5 init B5.new
// set b5 to the attribute b of A.
a5.b := b5

a5.b := void

var a5 : A5 init A5.new
var b5 : B5 init B5.new
a5.b := b5
// this assertion is true.
assert(b5.a == a5)

var a5 : A5 init A5.new
var b5 : B5 init B5.new
// add b5 to the attribute "b"
a5.b := b5

var a5c : A5 init b5.container()
// this assertion is true
assert(a5c.equals(a5))

	Note
	Be careful with attributes or references ref with multiplicity greater than 1, they are automatically initialized with a reflective collection (ie. a collection that is aware of an eventual opposite). So, you cannot assign them using := If you wish to change completely its content with the one of another collection, you must use the clear and addAll operations.

Attribute and reference have one main behavior difference.

Attribute has a notion of containment that reference hasn't.

This has some implication on the behavior of the assignment because an attribute cannot be owned by more than one object at a time.

There is an exception for attributes which type is a primitive type (String, Integer, Boolean, Real, Char) : since those types inherit from ValueType, they are not concerned by the composition, opposite concepts. In this case, the assignment doesn't impact any value but the assigned one.

Example 1: Assignment behavior for attribute

class A { attribute c1 : C }
class B { attribute c2 : C }
class C { }
aA.c1 := C.new
aB.c2 := aA.c1     // now aA.c1 == void !!!

Example 2: Assignment behavior for reference

class A { reference c1 : C }
class B { reference c2 : C }
class C { }

aA.c1 := C.new
aB.c2 := aA.c1     // aB.c2 == aA.c1 and aA.c1 keeps its value !!!

Example 3: Assignment behavior for attribute which type is String

class A { reference c1 : String }class B { reference c2 : String }aA.c1 := "Robert"aB.c2 := aA.c1 // a