Defining new classes

Kawa provides various mechanisms for defining new classes. The define-class and define-simple-class forms will usually be the preferred mechanisms. They have basically the same syntax, but have a couple of differences. define-class allows multiple inheritance as well as true nested (first-class) class objects. However, the implementation is more complex: code using it is slightly slower, and the mapping to Java classes is a little less obvious. (Each Scheme class is implemented as a pair of an interface and an implementation class.) A class defined by define-simple-class is slightly more efficient, and it is easier to access it from Java code.

The syntax of define-class are mostly compatible with that in the Guile and Stk dialects of Scheme.

Syntax: define-class class-name (supers ...) (annotation|option-pair)* field-or-method-decl ...

Syntax: define-simple-class class-name (supers ...) (annotation|option-pair)* field-or-method-decl ...

Defines a new class named class-name. If define-simple-class is used, creates a normal Java class named class-name in the current package. (If class-name has the form <xyz> the Java implementation type is named xyz.) For define-class the implementation is unspecified. In most cases, the compiler creates a class pair, consisting of a Java interface and a Java implementation class.

class-name ::= identifier
option-pair ::= option-keyword option-value
field-or-method-decl ::= field-decl | method-decl

General class properties

The class inherits from the classes and interfaces listed in supers. This is a list of names of classes that are in scope (perhaps imported using require), or names for existing classes or interfaces optionally surrounded by <>, such as <gnu.lists.Sequence>. If define-simple-class is used, at most one of these may be the name of a normal Java class or classes defined using define-simple-class; the rest must be interfaces or classes defined using define-class. If define-class is used, all of the classes listed in supers should be interfaces or classes defined using define-class.

interface: make-interface

Specifies whether Kawa generates a Java class, interface, or both. If make-interface is #t, then a Java interface is generated. In that case all the supertypes must be interfaces, and all the declared methods must be abstract. If make-interface is #f, then a Java class is generated. If interface: is unspecified, the default is #f for define-simple-class. For define-class the default is to generate an interface, and in addition (if needed) a helper class that implements the interface. (In that case any non-abstract methods are compiled to static methods. The methods that implement the interface are just wrapper methods that call the real static methods. This allows Kawa to implement true multiple inheritance.)

access: kind

Specifies the Java access permission on the class. Can be one of 'public (which is the default in Kawa), 'package (which the default "unnamed" permission in Java code), 'protected, 'private, 'volatile, or 'transient. Can also be used to specify final, abstract, or enum, as in Java. (You don’t need to explicitly specify the class is abstract if any method-body is #!abstract, or you specify interface: #t.) The kind can also be a list, as for example:

access: '(protected volatile)
class-name: "cname"

Specifies the Java name of the created class. The name specified after define-class or define-simple-class is the Scheme name, i.e. the name of a Scheme variable that is bound to the class. The Java name is by default derived from the Scheme name, but you can override the default with a class-name: specifier. If the cname has no periods, then it is a name in the package of the main (module) class. If the cname starts with a period, then you get a class nested within the module class. In this case the actual class name is moduleClass$rname, where rname is cname without the initial period. To force a class in the top-level (unnamed) package (something not recommended) write a period at the end of the cname.

Declaring fields

field-decl ::= (field-name (annotation | opt-type-specifier | field-option)*)
field-name ::= identifier
field-option ::= keyword expression

As a matter of style the following order is suggested, though this not enforced:

Each field-decl declares a instance "slot" (field) with the given field-name. By default it is publicly visible, but you can specify a different visiblity with the access: specifier. The following field-option keywords are implemented:

type: type

Specifies that type is the type of (the values of) the field. Equivalent to ‘:: type’.

allocation: kind

If kind is 'class or 'static a single slot is shared between all instances of the class (and its sub-classes). Not yet implemented for define-class, only for define-simple-class. In Java terms this is a static field.

If kind is 'instance then each instance has a separate value "slot", and they are not shared. In Java terms, this is a non-static field. This is the default.

access: kind

Specifies the Java access permission on the field. Can be one of 'private, 'protected, 'public (which is the default in Kawa), or 'package (which the default "unnamed" permission in Java code). Can also be used to specify volatile, transient, enum, or final, as in Java, or a quoted list with these symbols.

init: expr

An expression used to initialize the slot. The expression is evaluated in a scope that includes the field and method names of the current class.

init-form: expr

An expression used to initialize the slot. The lexical environment of the expr is that of the define-class; it does not include the field and method names of the current class. or define-simple-class.

init-value: value

A value expression used to initialize the slot. For now this is synonymous with init-form:, but that may change (depending on what other implementation do), so to be safe only use init-value: with a literal.

init-keyword: name:

A keyword that that can be used to initialize instance in make calls. For now, this is ignored, and name should be the same as the field’s field-name.

The field-name can be left out. That indicates a "dummy slot", which is useful for initialization not tied to a specific field. In Java terms this is an instance or static initializer, i.e., a block of code executed when a new instance is created or the class is loaded.

In this example, x is the only actual field. It is first initialized to 10, but if (some-condition) is true then its value is doubled.

(define-simple-class <my-class> ()
  (allocation: 'class
   init: (perform-actions-when-the-class-is-initizalized))
  (x init: 10)
  (init: (if (some-condition) (set! x (* x 2)))))

Declaring methods

method-decl ::= ((method-name formal-arguments)
    method-option* [deprecated-return-specifiermethod-body)
method-name ::= identifier
method-option ::= annotation | opt-return-type | option-pair
method-body ::= body | #!abstract | #!native
deprecated-return-specifier ::= identifier

Each method-decl declares a method, which is by default public and non-static, and whose name is method-name. (If method-name is not a valid Java method name, it is mapped to something reasonable. For example foo-bar? is mapped to isFooBar.) The types of the method arguments can be specified in the formal-arguments. The return type can be specified by a opt-return-type, deprecated-return-specifier, or is otherwise the type of the body. Currently, the formal-arguments cannot contain optional, rest, or keyword parameters. (The plan is to allow optional parameters, implemented using multiple overloaded methods.)

A method-decl in a define-simple-class can have the following option-keywords:

access: kind

Specifies the Java access permission on the method. Can be one of 'private, 'protected, 'public, or 'package. Can also be 'synchronized, 'final, 'strictfp, or a quoted list.

allocation: kind

If kind is 'class or 'static creates a static method.

throws: ( exception-class-name ... )

Specifies a list of checked exception that the method may throw. Equivalent to a throws specification in Java code. For example:

(define-simple-class T
  ((lookup name) throws: (
   (make (string-append prefix name))))

The scope of the body of a method includes the field-decls and method-decls of the class, including those inherited from superclasses and implemented interfaces.

If the method-body is the special form #!abstract, then the method is abstract. This means the method must be overridden in a subclass, and you’re not allowed to create an instance of the enclosing class.

(define-simple-class Searchable () interface: #t
  ((search value) :: boolean #!abstract))

If the method-body is the special form #!native, then the method is native, implemented using JNI.

The special method-name*init*’ can be used to name a non-default constructor (only if make-interface discussed above is #f). It can be used to initialize a freshly-allocated instance using passed-in parameters. You can call a superclass or a sibling constructor using the invoke-special special function. (This is general but admittedly a bit verbose; a more compact form may be added in the future.) See the example below.


In the following example we define a simple class 2d-vector and a class 3d-vector that extends it. (This is for illustration only - defining 3-dimensional points as an extension of 2-dimensional points does not really make sense.)

(define-simple-class 2d-vector ()
  (x ::double init-keyword: x:)
  ;; Alternative type-specification syntax.
  (y type: double init-keyword: y:)
  (zero-2d :: 2d-vector allocation: 'static
   init-value: (2d-vector 0))
  ;; An object initializer (constructor) method.
  ((*init* (x0 ::double) (y0 ::double))
   (set! x x0)
   (set! y y0))
  ((*init* (xy0 ::double))
   ;; Call above 2-argument constructor.
   (invoke-special 2d-vector (this) '*init* xy0 xy0))
  ;; Need a default constructor as well.
  ((*init*) #!void)
  ((add (other ::2d-vector)) ::2d-vector
   ;; Kawa compiles this using primitive Java types!
     x: (+ x other:x)
     y: (+ y other:y)))
  ((scale (factor ::double)) ::2d-vector
   (2d-vector x: (* factor x) y: (* factor y))))

(define-simple-class 3d-vector (2d-vector)
  (z type: double init-value: 0.0 init-keyword: z:)
  ;; A constructor which calls the superclass constructor.
  ((*init* (x0 ::double) (y0 ::double) (z0 ::double))
   (invoke-special 2d-vector (this) '*init* x0 y0)
   (set! z z0))
  ;; Need a default constructor.
  ((*init*) #!void)
  ((scale (factor ::double)) ::2d-vector
   ;; Note we cannot override the return type to 3d-vector
   ;; because Kawa doesn't yet support covariant return types.
     x: (* factor x)
     y: (* factor (this):y) ;; Alternative syntax.
     z: (* factor z))))

Note we define both explicit non-default constructor methods, and we associate fields with keywords, so they can be named when allocating an object. Using keywords requires a default constructor, and since having non-default constructors suppresses the implicit default constructor we have to explicitly define it. Using both styles of constructors is rather redundant, though.