The most important part of a function is of course its implementation, which in Xtend is either a single block expression or a rich string expression.
The block expression allows to have imperative code sequences. It consists of a sequence of expressions, and returns the value of the last expression. The return type of a block is also the type of the last expression. Empty blocks return null. Variable declarations are only allowed within blocks and cannot be used as a block's last expression.
A block expression is surrounded by curly braces and contains at least one expression. It can optionally be terminated by a semicolon.
{ doSideEffect("foo") result }
{ var x = greeting(); if (x.equals("Hello ")) { x+"World!"; } else { x; } }
XBlockExpression: '{' (XExpressionInsideBlock ';'?)* '}';
A literal denotes a fixed unchangeable value. Literals for string, integers, booleans, null and Java types are supported.
A string literal as syntactically defined in section Xbase_Syntax_StringLiteral is a valid expression and returns an instance of java.lang.String of the given value.
"Hello World !"
An integer literal as defined in section Xbase_Syntax_IntegerLiteral creates an int. There is no signed int. If you put a minus operator in front of an int literal it is taken as a UnaryOperator with one argument (the positive int literal).
There are two boolean literals, true and false which correspond to their Java counterpart of type boolean.
The null pointer literal is, like in Java, null. It is a member of any reference type.
Type literals are specified using the keyword typeof :
There are a couple of common predefined infix operators. In contrast to Java, the operators are not limited to operations on certain types. Instead an operator-to-method mapping allows users to redefine the operators for any type just by implementing the corresponding method signature. The following defines the operators and the corresponding Java method signatures / expressions.
e1 += e2 | e1._operator_add(e2) |
e1 || e2 | e1._operator_or(e2) |
e1 && e2 | e1._operator_and(e2) |
e1 == e2 | e1._operator_equals(e2) |
e1 != e2 | e1._operator_notEquals(e2) |
e1 < e2 | e1._operator_lessThan(e2) |
e1 > e2 | e1._operator_greaterThan(e2) |
e1 <= e2 | e1._operator_lessEqualsThan(e2) |
e1 >= e2 | e1._operator_greaterEqualsThan(e2) |
e1 -> e2 | e1._operator_mappedTo(e2) |
e1 .. e2 | e1._operator_upTo(e2) |
e1 + e2 | e1._operator_plus(e2) |
e1 - e2 | e1._operator_minus(e2) |
e1 * e2 | e1._operator_multiply(e2) |
e1 / e2 | e1._operator_divide(e2) |
e1 % e2 | e1._operator_modulo(e2) |
e1 ** e2 | e1._operator_power(e2) |
! e1 | e1._operator_not() |
- e1 | e1._operator_minus() |
If the operators || and && are used in a context where the left hand operand is of type boolean, the operation is evaluated in short circuit mode, which means that the right hand operand might not be evaluated at all in the following cases:
Local variables can be reassigned using the = operator. Also properties can be set using that operator: Given the expression
myObj.myProperty = "foo"
The compiler first looks up whether there is an accessible Java Field called myProperty on the type of myObj. If there is one it translates to the following Java expression :
myObj.myProperty = "foo";
Remember in Xbase everything is an expression and has to return something. In the case of simple assignments the return value is the value returned from the corresponding Java expression, which is the assigned value.
If there is no accessible field on the left operand's type, a method called setMyProperty(OneArg) (JavaBeans setter method) is looked up. It has to take one argument of the type (or a super type) of the right hand operand. The return value will be whatever the setter method returns (which usually is null). As a result the compiler translates to :
myObj.setMyProperty("foo")
A feature call is used to invoke members of objects, such as fields and methods, but also can refer to local variables and parameters, which are made available for the current expression's scope.
The following snippet is a simplification of the real Xtext rules, which cover more than the concrete syntax.
FeatureCall : ID | Expression ('.' ID ('(' Expression (',' Expression)* ')')?)*
Feature calls are directly translated to their Java equivalent with the exception, that for calls to properties an equivalent rule as described in section Xbase_Expressions_PropertyAssignment applies. That is, for the following expression
myObj.myProperty
the compiler first looks for an accessible field in the type of myObj. If no such field exists it looks for a method called myProperty() before it looks for the getter methods getMyProperty(). If none of these members can be found the expression is unbound and a compiliation error is thrown.
If the current scope contains a variable named this, the compiler will make all its members available to the scope. That is if
this.myProperty
myProperty
Checking for null references can make code very unreadable. In many situations it is ok for an expression to return null if a receiver was null. Xbase supports the safe navigation operator ?. to make such code more readable.
Instead of writing
if ( myRef != null ) myRef.doStuff()
one can write
myRef?.doStuff()
Construction of objects is done by invoking Java constructors. The syntax is exactly as in Java.
new String()
new java.util.ArrayList<java.math.BigDecimal>()
XConstructorCall: 'new' QualifiedName ('<' JvmTypeArgument (',' JvmTypeArgument)* '>')? ('('(XExpression (',' XExpression)*)?')')?;
An if expression is used to choose two different values based on a predicate. While it has the syntax of Java's if statement it behaves like Java's ternary operator (predicate ? thenPart : elsePart), i.e. it is an expression that returns a value. Consequently, you can use if expressions deeply nested within expressions.
XIfExpression: 'if' '(' XExpression ')' XExpression ('else' XExpression)?;
An expression if (p) e1 else e2 results in either the value e1 or e2 depending on whether the predicate p evaluates to true or false. The else part is optional which is a shorthand for else null. That means
if (foo) x
is the a short hand for
if (foo) x else null
The type of an if expression is calculated by the return types T1 and T2 of the two expression e1 and e2. It uses the rules defined in section Xbase_Types_CommonSuperType.
The for loop for (T1 variable : iterableOfT1) expression is used to execute a certain expression for each element of an array of an instance of java.lang.Iterable. The local variable is final, hence canot be updated.
The return type of a for loop is void. The type of the local variable can be left out. In that case it is inferred from the type of the array or java.lang.Iterable returned by the iterable expression.
for (String s : myStrings) { doSideEffect(s); }
for (s : myStrings) doSideEffect(s)
XForExpression: 'for' '(' JvmFormalParameter ':' XExpression ')' XExpression ;
A while loop while (predicate) expression is used to execute a certain expression unless the predicate is evaluated to false. The return type of a while loop is void.
XWhileExpression: 'while' '(' predicate=XExpression ')' body=XExpression;
while (true) { doSideEffect("foo"); }
while ( ( i = i + 1 ) < max ) doSideEffect( "foo" )
A do-while loop do expression while (predicate) is used to execute a certain expression unless the predicate is evaluated to false. The difference to the while loop is that the execution starts by executing the block once before evaluating the predicate for the first time. The return type of a do-while loop is void.
XDoWhileExpression: 'do' body=XExpression 'while' '(' predicate=XExpression ')';
do { doSideEffect("foo"); } while (true)
do doSideEffect("foo") while ((i=i+1)<max)
Variable declarations are only allowed within blocks. They are visible in any subsequent expressions in the block. Although overriding or shadowing variables from outer scopes is allowed, it is usually only used to overload the variable name 'this', in order to subsequently access an object's features in an unqualified manner.
A variable declaration starting with the keyword val denotes a so called value, which is essentially a final (i.e. unsettable) variable. In rare cases, one needs to update the value of a reference. In such situations the variable needs to be declared with the keyword var, which stands for 'variable'.
A typical example for using var is a counter in a loop.
{ val max = 100 var i = 0 while (i > max) { println("Hi there!") i = i +1 } }
Variables declared outside a closure using the var keyword are not accessible from within a closure.
XVariableDeclaration: ('val' | 'var') JvmTypeReference? ID '=' XExpression;
The return type of a variable declaration expression is always void. The type of the variable itself can either be explicitly declared or be inferred from the right hand side expression. Here is an example for an explicitly declared type:
var List<String> msg = new ArrayList<String>();
Alternatively the type can be left out and will be inferred from the initialization expression:
var msg = new ArrayList<String>(); // -> type ArrayList<String>
A closure is a literal that defines an anonymous function. A closure also captures the current scope, so that any final variables and parameters visible at construction time can be referred to in the closure's expression.
XClosure: '[' ( JvmFormalParameter (',' JvmFormalParameter)* )? '|' XExpression ']';
The surrounding square brackets are optional if the closure is the single argument of a method invocation. That is you can write
myList.find(e|e.name==null)
instead of
myList.find([e|e.name==null])
But in all other cases the square brackets are mandatory:
val func = [String s| s.length>3]
Closures are expressions which produce function objects. The type is a function type, consisting of the types of the parameters as well as the return type. The return type is never specified explicitly but is always inferred from the expression. The parameter types can be inferred if the closure is used in a context where this is possible.
For instance, given the following Java method signature:
public T <T>getFirst(List<T> list, Function0<T,Boolean> predicate)
the type of the parameter can be inferred. Which allows users to write:
arrayList( "Foo", "Bar" ).findFirst( e | e == "Bar" )
arrayList( "Foo", "Bar" ).findFirst( String e | e == "Bar" )
An Xbase closure is a Java object of one of the Function interfaces shipped with the runtime library of Xbase. There is an interface for each number of parameters (current maximum is six parameters). The names of the interfaces are
In order to allow seamless integration with existing Java libraries such as the JDK or Google Guava (formerly known as Google Collect) closures are auto coerced to expected types if those types declare only one method (methods from java.lang.Object don't count).
As a result given the method java.util.Collections.sort(List<T>, Comparator<? super T>) is available as an extension method, it can be invoked like this
newArrayList( 'aaa', 'bb', 'c' ).sort( e1, e2 | if ( e1.length > e2.length ) { -1 } else if ( e1.length < e2.length ) { 1 } else { 0 })
The switch expression is a bit different from Java's. First, there is no fall through which means only one case is evaluated at most. Second, the use of switch is not limited to certain values but can be used for any object reference instead. For a switch expression
switch e { case e1 : er1 case e2 : er2 ... case en : ern default : er }
the main expression e is evaluated first and then each case sequentially. If the switch expression contains a variable declaration using the syntax known from section Xbase_Expressions_ForLoop, the value is bound to the given name. Expressions of type java.lang.Boolean or boolean are not allowed in a switch expression.
The guard of each case clause is evaluated until the switch value equals the result of the case's guard expression or if the case's guard expression evaluates to true. Then the right hand expression of the case evaluated and the result is returned.
If none of the guards matches the default expression is evaluated an returned. If no default expression is specified the expression evaluates to null.
Example:
switch myString { case myString.length>5 : 'a long string.' case 'foo' : 'It's a foo.' default : 'It's a short non-foo string.' }
In addition to the case guards one can add a so called Type Guard which is syntactically just a type reference preceding the than optional case keyword. The compiler will use that type for the switch expression in subsequent expressions. Example:
var Object x = ...; switch x { String case x.length()>0 : x.length() List<?> : x.size() default : -1 }
Only if the switch value passes a type guard, i.e. an instanceof operation returns true, the case's guard expression is executed using the same semantics explained in previously. If the switch expression contains an explicit declaration of a local variable or the expression references a local variable, the type guard acts like a cast, that is all references to the switch value will be of the type specified in the type guard.
The return type of a switch expression is computed using the rules defined in section Xbase_Types_CommonSuperType. The set of types from which the common super type is computed corresponds to the types of each case's result expression. In case a switch expression's type is computed using the expected type from the context, it is sufficient to return the expected type if all case branches types conform to the expected type.
switch foo { Entity : foo.superType.name Datatype : foo.name default : throw new IllegalStateException }
switch x : foo.bar.complicated('hello',42) { case "hello42" : ... case x.length<2 : ... default : .... }
XSwitchExpression: 'switch' (ID ':')? XExpression '{' XCasePart+ ('default' ':' XExpression))? '}'; XCasePart: JvmTypeReference? ('case' XExpression)? ':' XExpression ); }
Type cast behave like casts in Java, but have a slightly more readable syntax. Type casts bind stronger than any other operator but weaker than feature calls.
The conformance rules for casts are defined in the Java Language Specification.
XCastedExpression: Expression 'as' JvmTypeReference;
Rich Strings allow for readable string concatenation, which is the main thing you do when writing a code generator. Let's have a look at an example of how a typical function with template expressions looks like:
toClass(Entity e) ''' package «e.packageName»; «placeImports» public class «e.name» «IF e.extends!=null»extends «e.extends»«ENDIF» { «FOREACH e.members» «member.toMember» «ENDFOREACH» } '''
If you are familiar with Xpand, you'll notice that it is exactly the same syntax. The difference is, that the template syntax is actually an expression, which means it can occur everywhere where an expression is expected. For instance in conjunction the powerful switch expression from Xbase:
toMember(Member m) { switch m { Field : '''private «m.type» «m.name» ;''' Method case isAbstract : ''' abstract «...''' Method : ''' ..... ''' } }
There is a special IF to be used within rich strings which is identical in syntax and meaning to the old IF from Xpand. Note that you could also use the if expression, but since it has not an explicit terminal token, it is not as readable in that context.
Also the FOR statement is available and can only be used in the context of a rich string. It also supports the SEPARATOR from Xpand. In addition, a BEFORE expression can be defined that is only evaluated if the loop is at least evaluated once before the very first iteration. Consequently AFTER is evaluated after the last iteration if there is any element.
The rich string is translated to an efficient string concatenation and the return type of a rich string is CharSequence which allows room for efficient implementation.
One of the key features of rich strings is the smart handling of white space in the template output. The white space is not written into the output data structure as is but preprocessed. This allows for readable templates as well as nicely formatted output. This can be achieved by applying three simple rules when the rich string is evaluated.
The behavior is best described with a set of examples. The following table assumes a data structure of nested nodes.
class Template { print(Node n) ''' node «n.name» {} ''' } |
node NodeName{} |
The indentation before node «n.name» will be skipped as it is relative to the opening mark of the rich string and thereby not considered to be relevant for the output but only for readability of the template itself.
class Template { print(Node n) ''' node «n.name» { «IF hasChildren» «n.children*.print» «ENDIF» } ''' } |
node Parent{ node FirstChild { } node SecondChild { node Leaf { } } } |
As in the previous example, there is no indentation on the root level for the same reason. The first nesting level has only one indentation level in the output. This is derived from the indentation of the IF hasChildren condition in the template which is nested in the node. The additional nesting of the recursive invocation children*.print is not visible in the output as it is relative the the surrounding control structure. The line with IF and ENDIF contain only control structures thus they are skipped in the output. Note the additional indentation of the node Leaf which happens due to the first rule: Indentation is propagated to called templates.
Although an explicit return is often not necessary, it is supported. In a closure for instance a return expression is always implied if the expression itself is not of type void. Anyway you can make it explicit:
listOfStrings.map(e| { if (e==null) return "NULL" e.toUpperCase })
Like in Java it is possible to throw java.lang.Throwable. The syntax is exactly the same as in Java.
{ ... if (myList.isEmpty) throw new IllegalArgumentException("the list must not be empty") ... }
The try-catch-finally expression is used to handle exceptional situations. You are not forced to declare checked exceptions, if you don't catch checked exceptions they are rethrown in a wrapping runtime exception. Other than that the syntax again is like the one known from Java.
try { throw new RuntimeException() } catch (NullPointerException e) { // handle e } finally { // do stuff }