Writing Apache Clownfish classes

Parcels

Every Clownfish class belongs to a Clownfish parcel. Parcels are used for namespacing and versioning. Information about parcels is stored in .cfp files which contain a JSON hash with the following keys:

• name: The parcel’s name. It must contain only letters.

• nickname: A short nickname. It must contain only letters. This nickname, followed by an underscore, is used to prefix generated C symbols and macros. Depending on the kind of symbol, a lowercase or uppercase prefix will be used.

• version: A version specifier of the following form (without whitespace):

  version-specifier = "v" version-number
  version-number = digit+ | digit+ "." version-number
  

• prerequisites: A hash containing the prerequisite parcels. The hash keys are the parcel names. The values contain the minimum required version.

An example .cfp file might look like:

{
    "name": "Pathfinder",
    "nickname": "Pfind",
    "version": "v2.3.8",
    "prerequisites": {
        "Clownfish": "v0.6.0"
    }
}

A parcel specifier of the following form is used in Clownfish header files:

parcel-specifier = "parcel" parcel-name ";"
parcel-name = identifier

For example:

parcel Pathfinder;

Every .cfh file starts with a parcel specifier containing the name of the parcel for all classes in the header file.

Initialization

Every Clownfish parcel must be initialized before it is used. The initialization function is named {parcel_nick}_bootstrap_parcel and takes no arguments.

Example call:

pfind_bootstrap_parcel();

The generated host language bindings call the bootstrap function automatically. C projects must call the function manually.

Short names

If a macro with the uppercase name {PARCEL_NICK}_USE_SHORT_NAMES is defined before including a generated C header, additional macros without the parcel prefix will be defined for most symbols.

Example:

#define PFIND_USE_SHORT_NAMES
#include <Pathfinder/Graph.h>
#include <Pathfinder/Path.h>

/* Prefixes can be omitted. */
Path *path = Graph_Find_Shortest_Path(graph);

/* Without PFIND_USE_SHORT_NAMES, one would have to write: */
pfind_Path *path = PFIND_Graph_Find_Shortest_Path(graph);

For object types in Clownfish header files, prefixes of class structs can also be omitted unless multiple parcels declare classes with the same last component of the class name.

The “Clownfish” parcel

The Clownfish runtime classes live in a parcel named Clownfish with nickname Cfish. Consequently, the short name macro is named CFISH_USE_SHORT_NAMES.

Declaring classes

Classes are declared in Clownfish header files using a declaration of the following form:

class-declaration = class-exposure-specifier?
                    class-modifier*
                    "class" class-name
                    ("nickname" class-nickname)?
                    ("inherits" class-name)?
                    "{" class-contents "}"
class-exposure-specifier = "public"
class-modifier = "inert" | "final"
class-name = identifier | identifier "::" class-name
class-nickname = identifier
class-contents = (variable-declaration | function-declaration)*

Class name components must start with an uppercase letter and must not contain underscores. The last component must contain at least one lowercase letter and must be unique for every class in a parcel.

For every class, a type with the name {parcel_nick}_{Class_Last_Comp} is defined in the generated C header. This is an opaque typedef used to ensure type safety.

For every class, a global variable with the uppercase name {PARCEL_NICK}_{CLASS_LAST_COMP} is defined. This variable is a pointer to a Clownfish::Class object which is initialized when bootstrapping the parcel.

Non-inert classes inherit from Clownfish::Obj by default.

Example of a class declaration:

parcel Pathfinder;

public class Pathfinder::Graph::VisibilityGraph nickname VisGraph
    extends Clownfish::Obj {
    /* Variables and functions */
}

This will generate:

typedef struct pfind_VisibilityGraph pfind_VisibilityGraph;
extern cfish_Class *PFIND_VISIBILITYGRAPH;

Class exposure

API documentation will only be generated for classes with public exposure.

Inert classes

Inert classes must contain only inert variables or inert functions, that is, neither instance variables nor methods. They must not inherit from another class nor be inherited from. They’re essentially nothing more than a namespace for functions and global variables.

Final classes

For final classes, every method is made final, regardless of the method modifier. Final classes must not be inherited from.

Variables

Variables are declared with a declaration of the following form:

variable-declaration = variable-modifier*
                       type variable-name ";"
variable-modifier = "inert"
variable-name = identifier

Inert variables

Inert variables are global class variables of which only a single copy exists. They are declared in the generated C header with the name {parcel_nick}_{Class_Nick}_{Variable_Name} and must be defined in a C source file.

Example:

public class Pathfinder::Path {
    public inert int max_path_length;
}

This will generate:

extern int pfind_Path_max_path_length;

The C source file defining the variable will typically use short names. So the definition will look like:

int Path_max_path_length = 5000;

Instance variables

Non-inert variables are instance variables and added to the class’s ivars struct.

Example:

public class Pathfinder::Path {
    int num_nodes;

    public int
    Get_Num_Nodes(Path *self);
}

This will add a num_nodes member to the ivars struct of Path.

The ivars struct

To access instance variables, the macro C_{PARCEL_NICK}_{CLASS_LAST_COMP} must be defined before including the generated header file. This will make a struct named {parcel_nick}_{Class_Name}IVARS with a corresponding typedef and short name available that contains all instance variables of the class and all superclasses from the same parcel. Instance variables defined in other parcels are not accessible. This is by design to guarantee ABI stability if the instance variable layout of a superclass from another parcel changes in a different version. If you need to access an instance variable from another parcel, add accessor methods.

A pointer to the ivars struct can be obtained by calling an inline function named {parcel_nick}_{Class_Name}_IVARS. This function takes the object of the class (typically self) as argument.

Example using short names:

#define C_PFIND_PATH
#define PFIND_USE_SHORT_NAMES
#include "Pathfinder/Path.h"

int
Path_get_num_nodes(Path *self) {
    PathIVARS *ivars = Path_IVARS(self);
    return ivars->num_nodes;
}

Functions

function-declaration = function-exposure-specifier?
                       function-modifier*
                       return-type function-name
                       "(" param-list? ")" ";"
function-exposure-specifier = "public"
function-modifier = "inert" | "inline" | "abstract" | "final"
return-type = return-type-qualifier* type
return-type-qualifier = "incremented" | "nullable"
function-name = identifier
param-list = param | param "," param-list
param = param-qualifier* type param-name ("=" scalar-constant)?
param-name = identifier
param-qualifier = "decremented"

Function exposure

API documentation will only be generated for functions with public exposure.

Inert functions

Inert functions are dispatched statically. They are declared in the generated C header with the name {parcel_nick}_{Class_Nick}_{Function_Name} and must be defined in a C source file. They must be neither abstract nor final.

Example:

public class Pathfinder::Graph::VisibilityGraph nickname VisGraph
    extends Clownfish::Obj {

    public inert incremented VisibilityGraph*
    new(int node_capacity);
}

This will generate:

pfind_VisibilityGraph*
pfind_VisGraph_new(int node_capacity);

The C source file implementing the inert function will typically use short names. So the implementation will look like:

#define PFIND_USE_SHORT_NAMES
#include "Pathfinder/Graph/VisibilityGraph.h"

VisibilityGraph*
VisGraph_new(int node_capacity) {
    /* Implementation */
}

Inline functions

Inert functions can be inline. They should be defined as static inline functions in a C block in the Clownfish header file. The macro CFISH_INLINE expands to the C compiler’s inline keyword and should be used for portability.

Methods

Non-inert functions are dynamically dispatched methods. Their name must start with an uppercase letter and every underscore must be followed by an uppercase letter. Methods must not be declared inline.

The first parameter of a method must be a pointer to an object of the method’s class which receives the object on which the method was invoked. By convention, this parameter is named self.

For every method, an inline wrapper for dynamic dispatch is defined in the generated C header with the name {PARCEL_NICK}_{Class_Nick}_{Method_Name}. Additionally, an implementing function is declared with the name {PARCEL_NICK}_{Class_Nick}_{Method_Name}_IMP. The Clownfish compiler also generates a typedef for the method’s function pointer type named {PARCEL_NICK}_{Class_Nick}_{Method_Name}_t. Wrappers and typedefs are created for all subclasses whether they override a method or not.

Example:

public class Pathfinder::Graph::VisibilityGraph nickname VisGraph
    extends Clownfish::Obj {

    public void
    Add_Node(VisibilityGraph *self, decremented Node *node);
}

This will generate:

/* Wrapper for dynamic dispatch */
static inline void
PFIND_VisGraph_Add_Node(pfind_VisibilityGraph *self, pfind_Node *node) {
    /* Inline code for wrapper */
}

/* Declaration of implementing function */
void
PFIND_VisGraph_Add_Node_IMP(pfind_VisibilityGraph *self,
                            pfind_Node *node);

/* Declaration of function pointer type */
typedef void
(*PFIND_VisGraph_Add_Node_t)(pfind_VisibilityGraph *self,
                             pfind_Node *node);

The implementing function of non-abstract methods must be defined in a C source file. This file will typically define the short names macro. So the implementation will look like:

#define PFIND_USE_SHORT_NAMES
#include "Pathfinder/Graph/VisibilityGraph.h"

void
VisGraph_Add_Node_IMP(VisibilityGraph *self, Node *node) {
    /* Implementation */
}

Looking up methods

Clownfish defines a macro named CFISH_METHOD_PTR that looks up the pointer to the implementing function of a method. The first parameter of the macro is a pointer to the Clownfish::Class object of the method’s class, the second is the unshortened name of the method wrapper. If short names for the Clownfish parcel are used, the macro is also available under the name METHOD_PTR.

To lookup methods from a superclass, there’s a macro CFISH_SUPER_METHOD_PTR with the same parameters.

Example using short names:

// Note that the name of the method wrapper must not be shortened.
VisGraph_Add_Node_t add_node
    = METHOD_PTR(VISIBILITYGRAPH, Pfind_VisGraph_Add_Node);

VisGraph_Add_Node_t super_add_node
    = SUPER_METHOD_PTR(VISIBILITYGRAPH, Pfind_VisGraph_Add_Node);

Abstract methods

For abstract methods, the Clownfish compiler generates an implementing function which throws an error. They should be overridden in a subclass.

Final methods

Final methods must not be overridden. They must not be abstract.

Nullable return type

If a function has a nullable return type, it must return a pointer. Non-nullable functions must never return the NULL pointer.

Incremented return type

Incremented return types must be pointers to Clownfish objects. The function will either return a new object with an initial reference count of 1 or increment the reference count. The caller must decrement the reference count of the returned object when it’s no longer used.

For returned objects with non-incremented return type, usually no additional handling of reference counts is required. Only if an object is returned from an accessor or a container object and the caller wants to use the object longer than the returning object retains a reference, it must increment the reference count itself and decrement when the object is no longer used.

Decremented parameters

Decremented parameters must be pointers to Clownfish objects. The function will either decrement the reference count of the passed-in object or retain a reference without incrementing the reference count. If the caller wants to use the passed-in object afterwards, it usually must increment its reference count before the call and decrement it when it’s no longer used. If the caller does not make further use of the passed-in object, it must not decrement its reference count after the call.

This is typically used in container classes like Vector:

String *string = String_newf("Hello");
Vec_Push(array, (Obj*)string);
// No need to DECREF the string.

Default parameter values

Default parameter values can be given as integer, float, or string literals. The values true, false, and NULL are also supported. The default values are only used by certain host language bindings. They’re not supported when calling a function from C.

C blocks

Clownfish headers can contain C blocks which start with a line containing the string __C__ and end on a line containing the string __END_C__. The contents of a C block are copied verbatim to the generated C header.

Example:

__C__

struct pfind_AuxiliaryStruct {
    int a;
    int b;
};

__END_C__

Object life cycle

Object creation

Objects are allocated by invoking the Make_Obj method on a class’s Clownfish::Class object.

Any inert function can be used to construct objects from C. But to support inheritance and object creation from the host language, Clownfish classes need a separate function to initialize objects. The initializer must take a pointer to an object as first argument and return a pointer to the same object. If the parent class has an initializer, it should be called first by the subclass’s initializer.

By convention, the standard constructor is named new and the standard initializer init.

Example:

/* Clownfish header */

class Vehicle {
    double max_speed;

    inert Vehicle*
    init(Vehicle *self, double max_speed);
}

class Train inherits Vehicle {
    double track_gauge;

    inert incremented Train*
    new(double max_speed, double track_gauge);

    inert Train*
    init(Train *self, double max_speed, double track_gauge);
}

/* Implementation */

Train*
Train_new(double max_speed, double track_gauge) {
    Train *self = (Train*)Class_Make_Obj(TRAIN);
    return Train_init(self, max_speed, track_gauge);
}

Train*
Train_init(Train *self, double max_speed, double track_gauge) {
    Vehicle_init((Vehicle*)self, max_speed);
    self->track_gauge = track_gauge;
    return self;
}

Reference counting

Clownfish uses reference counting for memory management. Objects are created with a reference count of 1. There are two macros CFISH_INCREF and CFISH_DECREF to increment and decrement reference counts. If short names for the Clownfish parcel are enabled, the macros can be abbreviated to INCREF and DECREF. Both macros take a pointer to an object as argument. NULL pointers are allowed. CFISH_INCREF returns a pointer to the object. This value might differ from the passed-in pointer in some cases. So if a reference is retained, the pointer returned from CFISH_INCREF should be used. CFISH_DECREF returns the modified reference count.

Examples:

self->value = INCREF(arg);

DECREF(object);

Object destruction

If an object’s reference count reaches 0, its Destroy method is called. This public method takes no arguments besides self and has no return value. It should release the resources held by the object and finally call the Destroy method of the superclass via the CFISH_SUPER_DESTROY macro with short name SUPER_DESTROY. This macro takes the self pointer as first argument and a pointer to the object’s Clownfish::Class as second argument. The Destroy method of the Clownfish::Obj class will eventually free the object struct.

Example:

/* Clownfish header */

class Path {
    Vector *nodes;

    public void
    Destroy(Path *self);
}

/* Implementation */

void
Path_Destroy_IMP(Path *self) {
    DECREF(self->nodes);
    SUPER_DESTROY(self, PATH);
}

Documentation

The Clownfish compiler creates documentation in the host language’s preferred format from so-called DocuComments found in the .cfh files. DocuComments use Markdown (CommonMark flavor) for formatting. DocuComments are multi-line C-style comments that start with /**. They immediately precede the documented class, inert function, or method. A left border consisting of whitespace and asterisks is stripped.

The DocuComment for a class should start with a short description (everything up until the first period .) which may appear in the name section of a man page, for example.

DocuComments for functions and methods may end with a series of @param and @return directives which document the parameters and return values.

Example:

/** Class describing a train.
 *
 * The Train class describes a train. It extends the Vehicle class and
 * adds some useful properties specific to trains.
 */
public class Train inherits Vehicle {
    /** Create a new Train object.
     *
     * @param max_speed The maximum speed in km/h.
     * @param track_gauge The track gauge in mm.
     */
    public inert incremented Train*
    new(double max_speed, double track_gauge);

    /** Accessor for maximum speed.
     *
     * @return the maximum speed in km/h.
     */
    public double
    Get_Max_Speed(Train *self);
}

The Clownfish compiler also looks for standalone Markdown .md files in the source directories which will be included in the documentation.