nilscript v2.0.1

oj

oj is a superset of the JavaScript language inspired by the latest versions of Objective-C. It features a fast, simple runtime without a dynamic messaging overhead.

oj is designed to ease the pain of syncing class interfaces (not necessarily implementations) between Objective-C projects and their web counterparts.

In our case, we use it to sync Tenuto with the musictheory.net exercises, and Theory Lessons with the musictheory.net lessons.

Installation

npm install ojc

Main Features

• Classes
  • Basic Syntax
  • Behind the Scenes
  • Forward Declarations
• The Built-in Base Class
• Methods
  • Falsy Messaging
  • Behind the Scenes
• Properties and Instance Variables
  • Synthesis
  • Using
  • Property Attributes
  • Initialization
  • Behind the Scenes
• Property Observers
• Callbacks
• Selectors
• Protocols
• Boolean/null aliases
• @enum and @const
• @global
• Runtime
• Restrictions
• Hinting
• Type Checking
• API
• Compiling Projects
• Squeezing and Symbolication
• Acknowledgements
• License

Differences from Objective-J

In contrast to Objective-J:

• oj always uses consistent property names. This allows the resulting JavaScript code to be optimized using Closure Compiler's ADVANCED_OPTIMIZATIONS.
• oj uses the native JavaScript runtime to call methods rather than imitating the Objective-C runtime (see below).
• oj focuses on being a language, not a framework. The only requirement at runtime is the runtime.js file.
• oj has full support of @property and the default synthesis of ivars/getters/setters.
• oj includes a built-in obfuscator which hides method and class names in compiled code.

Classes

While Objective-C uses @interface to define a class interface and @implementation for its implementation, oj only uses @implementation (due to the lack of header files in JavaScript). Information that would normally appear in the @interface block, such as @property declarations or the inherited superclass instead appear in @implementation.

Basic syntax

The syntax to create an empty oj class looks like this:

@implementation TheClass
@end

To inherit from a superclass, use a colon followed by the superclass name:

@implementation TheSubClass : TheSuperClass 
@end

Additional instance variables can be added by using a block after class name (or superclass name):

@implementation TheClass {
    String _myStringInstanceVariable;    
}
@end

@implementation TheSubClass : TheSuperClass {
    String _myStringInstanceVariable;    
}
@end

Behind the scenes (Class)

Behind the scenes, the oj compiler changes the @implementation/@end block into a JavaScript function block which is invoked at runtime. Private functions and variables may be declared inside of an @implementation without polluting the global namespace.

@implementation TheClass
let sPrivateStaticVariable = "Private";
function sPrivate() { }
@end

becomes equivalent to:

oj_private_function(…, function() {
    let sPrivateStaticVariable = "Private";
    function sPrivate() { }
});

To prevent undefined behavior, variable declarations must be initialized to a literal or function expression (or left uninitialized).

Note: Only @property, @synthesize, @dynamic, @observe, instance variable declarations, method declarations, variable declarations, or function declarations may be used inside of an @implementation block.

Forward Declarations

In older versions of oj (0.x), the compiler would compile each file separately. This led to situations where a forward declaration of a class was needed:

@forward TheFirstClass;

@implementation TheSecondClass
    
- (void) foo {
    // Without the forward declaration, oj 0.x didn't know if TheFirstClass
    // was a JS identifier or an oj class.
    [TheFirstClass doSomething];
}

@end

oj 1.x+ uses a multi-pass compiler which eliminates the need for forward declarations. In general, the need to use @forward indicates an underlying issue with the dependency tree, which will cause issues if you need to use @const/@enum inlining or the squeezer. For more information, read Compiling Projects.

The Built-in Base Class

Unlike Objective-C, all oj classes inherit from a private root base class. There is no way to specify your own root class (how often do you not inherit from NSObject in your code?).

The root base class provides the following methods:

+ (id) alloc
+ (Class) class
+ (Class) superclass
+ (String) className
+ (BOOL) isSubclassOfClass:(Class)cls

+ (BOOL) instancesRespondToSelector:(SEL)aSelector

- (id) init
- (id) copy

- (Class) class
- (Class) superclass
- (String) className 
- (BOOL) isKindOfClass:(Class)cls
- (BOOL) isMemberOfClass:(Class)cls

- (String) description 

- (BOOL) respondsToSelector:(SEL)aSelector
- (id) performSelector:(SEL)aSelector
- (id) performSelector:(SEL)aSelector withObject:(id)object
- (id) performSelector:(SEL)aSelector withObject:(id)object withObject:(id)object2

- (BOOL) isEqual:(id)anotherObject

While oj 0.x supported +load and +initialize, this feature was removed in oj 1.x to optimize runtime performance. Note: +className and -className are intended for debugging purposes only. When --squeeze is passed into the compiler, class names will be obfuscated/shortened.

Methods

Methods are defined in an @implementation block and use standard Objective-C syntax:

@implementation TheClass
    
- (String) doSomethingWithString:(String)string andNumber:(Number)number
{
    return string + "-" + number;    
}

// Returns "Foo-5"
- (String) anotherMethod
{
    return [self doSomethingWithString:"Foo" andNumber:5];
}
    
@end

Old-school bare method declarations may also be used:

@implementation TheClass
    
- doSomethingWithString:string andNumber:number
{
    return string + "-" + number;    
}
    
@end

Falsy Messaging

Just as Objective-C supports messaging nil, oj supports the concept of "Falsy Messaging".

Any message to a falsy JavaScript value (false / undefined / null / 0 / "" / NaN ) will return that value.

let foo = null;
let result = [foo doSomething];  // result is null

Behind the Scenes (Methods)

Behind the scenes, oj methods are simply renamed JavaScript functions. Each colon (:) in a method name is replaced by an underscore and a prefix is added to the start of the method name.

Hence:

- (String) doSomethingWithString:(String)string andNumber:(Number)number
{
    return string + "-" + number;    
}

becomes the equivalent of:

TheClass.prototype.$oj_f_doSomethingWithString_andNumber_ = function(string, number)
{
    return string + "-" + number;    
}

Messages to an object are simply JavaScript function calls wrapped in a falsey check. Hence:

 let result = [anObject doSomethingWithString:"Hello" andNumber:0];
 

becomes the equivalent of:

 let result = anObject && anObject.doSomethingWithString_andNumber_("Hello", 0);
 

The compiler will produce slightly different output depending on:

• if the return value is needed
• if the message receiver is a JavaScript expression.
• if the message receiver is known to be non-falsey
• if the message receiver is self
• if the message receiver is super

Sometimes the compiler will choose to use oj.msgSend() rather than a direct function call.

Properties and Instance Variables

oj uses the Objective-C 2.0 @property syntax which originally appeared in Mac OS X 10.5 Leopard. It also supports the concept of default property synthesis added in Xcode 4.4.

In addition, oj allows storage for additional instance variables (ivars) to be defined on a class.

A class that uses a property, private ivar, and accesses them in a method may look like this:

@implementation TheClass {
    Number _privateNumberIvar;
}
    
@property Number publicNumberProperty; // Generates publicNumberProperty ivar

- (Number) addPublicAndPrivateNumbers
{
    return _privateNumberIvar + _publicNumberIvar;
}
    
@end

Synthesis

Properties are defined using the @property keyword in an @implementation block:

@implementation TheClass
@property String myStringProperty;
@end

In the above example, the compiler will automatically synthesize a backing instance variable _myStringProperty for myStringProperty. It will also create an accessor method pair: -setMyStringProperty: and -myStringProperty.

If a different backing instance variable is desired, the @synthesize directive is used:

@implementation TheClass
@property String myStringProperty;
    
// Maps myStringProperty property to m_myStringProperty instance variable
@synthesize myStringProperty=m_MyStringProperty;
@end

As in Objective-C, @synthesize without an = results in the same name being used for the backing instance variable:

@implementation TheClass
@property String myStringProperty;
    
// Maps myStringProperty property to myStringProperty instance variable
@synthesize myStringProperty;
@end

The @dynamic directive suppresses the generation of both the backing instance variable and the setter/getter pair.

@implementation TheClass
@property String myStringProperty;
@dynamic myStringProperty; // No instance variable, getter, nor setter is synthesized
@end

In addition, multiple properties may be specified in @synthesize and @dynamic:

@synthesize prop1, prop2, prop3=m_prop3;
@dynamic dynamic1,dynamic2;

Using

To access any instance variable, simply use its name. No this. or self. prefix is needed:

- (void) logSheepCount
{
    console.log(_numberOfSheep);
}

Property Attributes

All valid Objective-C attributes may be used on a declared property:

@property (nontomic,copy,getter=myStringGetter) String myString;

However, some are ignored due to differences between JavaScript and Objective-C.

copy uses oj.makeCopy in the setter.

struct uses oj.makeCopy in both the setter and the getter. It is intended to assist the porting of C structs, which are pass-by-value rather than pass-by-reference.

@property (copy) Foo foo;
@property (struct) Bar bar;
@property Baz baz;

// Synthesized methods:

- (void) setFoo:(Foo)foo { _foo = oj.makeCopy(foo); }
- (Foo) foo { return _foo; }

- (void) setBar:(Bar)bar { _bar = oj.makeCopy(bar); }
- (Bar) bar { return oj.makeCopy(_bar); }

- (void) setBaz:(Bar)bar { _baz = baz; }
- (Baz) baz { return _baz; }

Initialization

During +alloc, oj initializes all instance variables to one of the following values based on its type:

Boolean         -> false
Number          -> 0
everything else -> null

This allows Number instance variables to be used in math operations without the fear of undefined being converted to NaN by the JavaScript engine.

Behind the Scenes (Properties/ivars)

Unlike other parts of the oj runtime, properties and instance variables aren't intended to be accessed from non-oj JavaScript (they should be private to the subclass which defines them). However, they may need to be accessed in the debugger.

The compiler currently uses a JavaScript property on the instance with the follow name:

$oj_i_{{CLASS NAME}}_{{IVAR NAME}}

Hence, the following oj code:

@interface TheClass

@property (Number) counter;

- (void) incrementCounter
{
    _counter++;
}
    
@end

would compile into:

oj.makeClass(…, function(…) {
    
… // Compiler generates -setCounter: and -counter here

….incrementCounter = function() {
    this.$oj_i_TheClass__counter++;
}

});

Property Observers

In our internal UI frameworks, it's very common to call -setNeedsDisplay or -setNeedsLayout in response to a property change. For example, our Button class has a custom corner radius property:

@implementation Button : ClickableControl

…

@property Number cornerRadius;

…

- (void) setCornerRadius:(Number)cornerRadius
{
    if (_cornerRadius != cornerRadius) {
        _cornerRadius = cornerRadius;
        [self setNeedsDisplay];
    }
}

@end

Often, every property in these classes needs a custom setter, resulting in a lot of boilerplate code. Property observers simplify this:

@property String backgroundColor;
@property Number cornerRadius;
@property String title;

@observe (change, after=setNeedsDisplay) backgroundColor, cornerRadius, title;

This example will call [self setNeedsDisplay] after the backgroundColor, colorRadius, or title changes. change is a default attribute and may be omitted.

before= observer methods are passed the new value as an optional parameter. after= observer methods are passed the old value as an optional parameter.

For example:

@property Number foo;
@observe (change, before=_handleFooWillChange:, after=_handleFooDidChange:) foo;
@observe (set,    before=_handleFooWillSet:,    after=_handleFooDidSet:)    foo;

- (void) _handleFooWillChange:(Number)newFoo { … }
- (void) _handleFooDidChange:(Number)oldFoo  { … }
- (void) _handleFooWillSet:(Number)newFoo { … }
- (void) _handleFooDidSet:(Number)oldFoo  { … }

Will generate the following setter:

- (void) setFoo:(Number)newFoo
{
    var oldFoo = _foo;

    [self _handleFooWillSet:newFoo];

    if (oldFoo !== newFoo) {
        [self _handleFooWillChange:newFoo];
        _foo = newFoo;
        [self _handleFooDidChange:oldFoo];
    }

    [self _handleFooDidSet:oldFoo];
}

Callbacks

Javascript frequently requires .bind(this) on callbacks. For example:

Counter.prototype.incrementAfterDelay = function(delay) {
    setTimeout(function() {
        this.count++;
        this.updateDisplay();
    }.bind(this), delay);       // Bind needed for 'this' to work
}

oj handles the binding for you. No additional code is needed to access ivars or self:

- (void) incrementAfterDelay:(Number)delay
{
    setTimeout(function() {
        _count++;
        [self updateDisplay];
    }, delay);
}

Selectors

In order to support consistent property names, selectors are not encoded as strings (as in Objective-C and Objective-J). Instead, they use an object literal syntax:

@selector(foo:bar:baz:) -> { $oj_f_foo_bar_baz_: 1 }

Thus, a call such as:

[object foo:7 bar:8 baz:9]

May (depending on optimizations) be turned into:

oj.msg_send(object, { $oj_f_foo_bar_baz_: 1 }, 7, 8, 9)

Boolean/null aliases

The oj compiler adds the following keywords for Boolean/null values and replaces them to their JavaScript equivalent:

BOOL    ->  Boolean
YES     ->  true
NO      ->  false

nil     ->  null
Nil     ->  null
NULL    ->  null

Hence:

let nope = NO;
let yep  = YES;
let anObject = nil;

becomes:

let nope = false;
let yep  = true;
let anObject = null;
  

@enum and @const

oj supports C-style enumerations via the @enum keyword and constants via the @const keyword:

@enum OptionalEnumName {
    zero = 0,
    one,
    two,
    three = 3,
    four
}

@const TheConstant = "Hello World";

someFunction(zero, one, two, three, four, TheConstant);

By default, oj compiles the above to:

var zero  = 0;
var one   = 1;
var two   = 2;
var three = 3;
var four  = 4;

var TheConstant = "Hello World";

someFunction(zero, one, two, three, four, TheConstant);

However, when the --inline-enum option is passed into the oj compiler, oj inlines enum values:

someFunction(0, 1, 2, 3, 4, TheConstant);

The --inline-const option inlines TheConstant as well:

someFunction(0, 1, 2, 3, 4, "Hello World");

Note: Inlining causes the enum or const to be lifted to the global scope. Inlining affects all occurrences of that identifier in all files for the current compilation. Inlined enums/consts are persisted via --output-state and --input-state.

@global

To mimic C APIs such as CoreGraphics, oj has the ability to declare global functions and variables with @global.

@global function CGRectMake(x: Number, y: Number, width: Number, height: Number): void {
    return { origin: { x, y }, size: { width, height } };
}
    
@global CGRectZero = CGRectMake(0, 0, 0, 0);
@global CGRectNull = CGRectMake(Infinity, Infinity, 0, 0);

Which transforms into the equivalent of:

$oj_oj._g.CGRectMake = function(x, y, width, height) {
    return { origin: { x, y }, size: { width, height } };
}
    
$oj_oj._g.CGRectZero = $oj_oj._g.CGRectMake(0, 0, 0, 0);
$oj_oj._g.CGRectNull = $oj_oj._g.CGRectMake(Infinity, Infinity, 0, 0);

Unlike inlined enums and consts, globals are assigned at runtime. Hence, in the above code example, care must be given that CGRectMake() isn't used for initializing CGRectZero until after the @global function CGRectMake line. This limitation should not affect globals used from within oj methods (as the global will already be declared by that time).

Protocols

Like Objective-C, oj includes support for protocols. Both @required and @optional methods may be specified:

@protocol ControllerDelegate
@required
- (void) controller:(Controller)controller didPerformAction:(String)action;
@optional
- (BOOL) controller:(Controller)controller shouldPerformAction:(String)action;
@end

@implementation Controller
@property id<ControllerDelegate> delegate
…
@end

@implementation TheClass <ControllerDelegate, TabBarDelegate>
- (void) controller:(Controller)controller didPerformAction:(String)action { … }
…
@end

Unlike Objective-C, there is no NSObject protocol. Instead, all protocols extend a built-in base protocol, which has identical methods to the built-in base class.

Protocol conformance is enforced by the typechecker.

Runtime

oj.noConflict()
Restores the oj global variable to its previous value.

oj.getClassList()
Returns an array of all known oj Class objects.

oj.class_getSuperclass(cls) / oj.getSuperclass(cls)
Returns the superclass of the specified cls.

oj.getSubclassesOfClass(cls)
Returns an array of all subclasses of the specified cls.

oj.isObject(object)
Returns true if object is an oj instance or Class, false otherwise.

oj.sel_isEqual(aSelector, bSelector)
Returns true if two selectors are equal to each other.

oj.class_isSubclassOf(cls, superclass)
Returns true if superclass is the direct superclass of cls, false otherwise.

oj.class_respondsToSelector(cls, aSelector)
Returns true if instances of cls respond to the selector aSelector, false otherwise.

oj.object_getClass(object)
Returns the Class of object.

oj.msgSend(receiver, aSelector, ...)
If receiver is non-falsy, invokes aSelector on it.

oj.sel_getName(aSelector)
oj.class_getName(cls)
-BaseObject className
Returns a human-readable string of a class or selector. Note that this is for debug purposes only! When --squeeze is passed into the compiler, the resulting class/selector names will be obfuscated/shortened.

oj.makeCopy(object)
If object is an oj instance, invokes -copy. If object is a JavaScript array, returns a shallow clone (via slice(0)). If object is a JavaScript primitive, returns object. Else, returns a clone of each key/value pair (via Object.keys) on object.

Hinting

oj provides basic code hinting to catch common errors.

When the --warn-unknown-selectors option is specified, oj warns about usage of undefined selectors/methods. This can help catch typos at compile time:

let c = [[TheClass allc] init]; // Warns if no +allc or -allc method exists on any class

When the --warn-unknown-ivars option is specified, oj checks all JavaScript identifiers prefixed with an underscore. A warning is produced when such an identifier is used in a method declaration and the current class lacks a corresponding @property or instance variable declaration.

@implementation TheClass
    
@property String foo;
    
- (void) checkFoo {
    if (_foi) {  // Warns, likely typo
    }    
}
    
@end

When the --warn-unused-ivars option is specified, oj warns about ivar declarations that are unused within an implementation.

@implementation TheClass {
    id _unused; // Warns
}
@end

When the --warn-unknown-selectors option is used, oj checks each selector against all known selectors.

oj integrates with JSHint via the --jshint option; however, this feature is deprecated and will be removed in the future (2.x). Many JSHint warnings are duplicated by the typechecker.

To prevent false positives, the following JSHint options are forced: asi: true, laxbreak: true, laxcomma: true, newcap: false.

expr: true is enabled on a per-method basis when the oj compiler uses certain optimizations.

The --jshint-ignore option may be used to disable specific JSHint warnings.

Type Checking

When the --check-types option is used, oj performs static type checking via TypeScript.

oj uses an Objective-C inspired syntax for types, which is automatically translated to and from TypeScript types:

Most oj method declarations will have type information and should behave exactly as their Objective-C counterparts. However, JavaScript functions need to be annotated via type annotations, similar to ActionScript and TypeScript:

function getStringWithNumber(a : String, b : Number) : String {
    return a + "-" + b;
}

TypeScript infers variables automatically; however, sometimes an explicit annotation is required. This annotation is similar to TypeScript syntax:

function getNumber() { … }

function doSometingWithNumber() : void {
    let num : Number = getNumber(); // Annotation needed since getNumber() is not annotated
    …
}

oj also provides @type to declare basic types. @type does not affect generated code and only provides hints to the typechecker:

@type MyNumericType = Number;
@type MyRect = { x: Number, y: Number, width: Number, height: Number };
@type MyDoneCallback = function(completed: BOOL): void;
@type MyTypedTuple = [ Number, Number, String ];

function makeSquare(length: Number): MyRect { … }
function loadWithCallback(callback: MyDoneCallback): void { … }

Casting is performed via the @cast operator:

let a : String = @cast(String, 3 + 4 + 6);

Sometimes you may wish to disable type checking for a specific variable or expression. While @cast(any, …) accomplishes this, you can also use the @any convinience operator:

let o = @any({ });

For some projects and coding styles, the default TypeScript rules may be too strict. For example, the following is an error in typescript:

function example() {
    let o = { };
    // This is an error in TypeScript, as 'foo' isn't a property on the '{}' type
    o.foo = "Foo";
}

By default, oj mitigates this by casting all objects literals to the any type. However, this may cause issues with function overloading when using external type definitions. Hence, you can revert to the original TypeScript behavior via the --strict-object-literals option.

TypeScript also requires function calls to strictly match the parameters of the definition. The following is allowed in JavaScript but not in TypeScript:

function foo(a, b) {
    …
}
    
foo(1); // Error in TS: parameter b is required
foo(1, 2, 3); // Error in TS

By default, oj mitigates this by rewriting function definitions so that all parameters are optional. You can revert to the original TypeScript behavior via the --strict-functions option.

For performance reasons, we recommend a separate typechecker pass (in parallel with the main build), with --check-types enabled, --output-language set to none, and TypeScript type definitions (such as those found at DefinitelyTyped) specified using the --prepend option.

oj tries to convert TypeScript error messages back into oj syntax. Please report any confusing error messages.

Restrictions

All identifiers that start with $oj_ or $oj$ are classified as Reserved Words.

Inside an oj method declaration, self is added to the list of Reserved Words. Hence, it may not be used as a variable name.

The oj compiler uses the global variable $oj_oj to access the runtime. You should not use $oj_oj directly or modify it in your source code. In a web browser environment, runtime.js also defines the global variable oj for the runtime. You may use oj.noConflict() to restore the previous value of oj. If you are using a linter or obfuscator, add both $oj_oj and oj as global variable names.

In order to support compiler optimizations, the following method names are reserved and may not be overridden/implemented in subclasses:

alloc
class
className
instancesRespondToSelector:
respondsToSelector:
superclass
isSubclassOfClass:
isKindOfClass:
isMemberOfClass:

API

Traditionally, oj's API consisted of a single compile method:

let ojc = require("ojc");
let options = { … };
    
ojc.compile(options, function(err, results) {
    
});

To allow for fast incremental compiles, oj 2.x adds a Compiler constructor:

let ojc = require("ojc");

// Important: create one compiler per output file.
let compiler = new ojc.Compiler();

let options = { … };

// Call doCompile() each time one of the files specified by options.files changes
function doCompile(callback) {
    compiler.compile(options, function(err, results) {
        callback(err, results);
    });
}

Below is a list of supported properties for options and results. While other properties are available (see bin/ojc), they are not official API.

Valid properties for the options object:

Valid properties for each file or defs object:

Properties for the result object:

The before-compile key specifies a callback which is called prior to the compiler's oj->js stage. This allows you to preprocess files. After this callback is invoked, a file's content must be valid oj or JavaScript.

The after-compile key specifies a callback which is called each time the compiler generates JavaScript code for a file. This allows you to run the generated JavaScript through a linter (such as JSHint or ESLint), or allows further transformations via Babel. When this callback is invoked, a file's content will be valid JavaScript.

// Simple preprocessor example.  Strips out #pragma lines and logs to console
ojOptions["before-compile"] = function(file, callback) {
    let inLines = file.getContents().split("\n");
    let outLines = [ ];

    inLines.forEach(line => {
        if (line.indexOf("#pragma") == 0) {
            console.log("Pragma found in: " + file.getPath());

            // Push an empty line to maintain the same # of lines
            outLines.push("");

        } else {
            outLines.push(line);
        }
    });
    
    file.setContents(outLines.join("\n"));
    
    callback();
}

// ESLint example
ojOptions["after-compile"] = function(file, callback) {
    if (!linter) linter = require("eslint").linter;

    // file.getContents() returns the generated source as a String
    _.each(linter.verify(file.getContents(), linterOptions), function(warning) {
        // file.addWarning(line, message) adds a warning at a specific line
        file.addWarning(warning.line, warning.message);
    });
    
    // linter#verify() is synchronous and doesn't produce errors, so just call callback()
    callback();
};

// Babel example
ojOptions["after-compile"] = function(file, callback) {
    if (!babel) babel = require("babel-core");
    
    // retainLines must be true or oj's output source map will be useless
    babelOptions.retainLines = true;

    try {
        let result = babel.transform(file.getContents(), babelOptions);

        // file.setContents() updates the generated source code with a string.
        // This string must have a 1:1 line mapping to the original string
        file.setContents(result.code);

    } catch (e) {
        file.addWarning(e.loc.line, e.message);
    }

    // Babel's transform API is synchronous
    callback();
};

Note: options.state and result.state are private objects and the format/contents will change between releases. Users are encouraged to use the new Compiler#uses API rather than state. (See below).

--

oj 2.x also adds the symbolicate function as API. This converts an internal oj identifier such as $oj_f_stringWithString_ to a human-readable string ("stringWithString:"). See Squeezing and Symbolication below.

Compiling Projects

The easiest way to use oj is to pass all .oj and .js files in your project into ojc and produce a single .js output file. In general: the more files you compile at the same time, the easier your life will be. However, there are specific situations where a more-complex pipeline is needed.

In our usage, we have two output files: core.js and webapp.js.

core.js contains our model and model-controller classes. It's used by our client-side web app (running in the browser), our server-side backend (running in node/Express), and our iOS applications (running in a JavaScriptCore JSContext).

webapp.js is used exclusively by the client-side web app and contains HTML/CSS view and view-controller classes. In certain cases, webapp.js needs to allocate classes directly from core.js.

In previous versions of oj, this was accomplished via the --output-state and --input-state compiler flags, or the options.state/result.state properties in the compiler API. The state output from core.js would be passed as the state input to webapp.js.

oj 2 introduces a new Compiler API with Compiler#uses and Compiler#compile. This allows both incremental compiles, and allows for more efficient state sharing:

let ojc = require("ojc");
let coreCompiler   = new ojc.Compiler();
let webAppCompiler = new ojc.Compiler();
    
let coreOptions   = { … };
let webAppOptions = { … };

// This tells webAppCompiler to always pull the last state from coreCompiler 
//
// It's your responsibility to watch files for changes and kick off the correct
// doXCompile() functions.
//
// If core.js includes the compiled result of foo.oj, a change to foo.oj 
// needs to call *both* doCoreCompile() and doWebAppCompile()
//
webAppCompiler.uses(coreCompiler);
    
// These functions are called due to file modification events (fs.watch)
function doCoreCompile(callback) {
    coreCompiler.compile(coreOptions, function(err, results) {
        callback(err, results);
    });
}
        
function doWebAppCompile(callback) {
    webAppCompiler.compile(webAppOptions, function(err, results) {
        callback(err, results);
    });
}    

1. All lower-level .js and .oj files are passed into coreCompiler via coreOptions.
2. The compiler products a result object. result.code is saved as core.js.
3. All higher-level .js and .oj files are passed into webAppCompiler. webAppCompiler pulls state from coreCompiler due to the Compiler#uses API.
4. The result.code from this compilation pass is saved as webapp.js.
5. Both core.js and webapp.js are included (in that order) in various HTML files via <script> elements.
6. The oj runtime (runtime.js) is also included in various HTML files. You can obtain its location via the ojc.getRuntimePath API.

--

We've found it best to run a separate typecheck pass in parallel with the core.js/webapp.js build (via a separate node process). This allows one CPU to be dedicated to typechecking while the other performs transpiling. The typecheck pass uses the following options:

• All .js and .oj files (From steps #1 and #3) are passed as INPUT_FILES (or options.files).
• Several .d.ts definitions (for jQuery, underscore, etc.) are specified with the --defs option (or options.defs).
• --output-language is set to none.
• --check-types is enabled

Squeezing and Symbolication

As mentioned in previous sections, oj uses internal identifier names for classes, methods, and ivars. These identifiers are always prefixed with $oj_…:

Since these identifiers can be quite long (and aid in competitor's reverse-engineering efforts), oj features a code minifier/compressor/obfuscator called the squeezer.

When the --squeeze option is passed to the compiler, each $oj_… identifier is replaced with a shortened "squeezed" version (prefixed with $oj$…). For example, all occurrences of $oj_c_Foo might be replaced with $oj$a, all occurrences of $oj_f_initWithFoo_ with $oj$b, etc. @globals are also replaced in this manner.

This is a safe transformation as long as all files are squeezed together (or state is persisted via --output-state and --input-state).

The --squeeze compiler option adds a squeeze property to the compiler results. This is a map of squeezed identifiers to original identifiers:

{
    "$oj$a": "$oj_c_TheClass",
    "$oj$b": "$oj_f_initWithFoo_"
    "$oj$c": "$oj_i_TheClass__firstIvar",
    "$oj$d": "$oj_i_TheClass__secondIvar",
    "$oj$e": "$oj_f_doSomethingWithFoo_bar_baz_",
    …
}

--

Symbolication is the process of transforming an internal identifier (either squeezed or unsqueezed) into a human-readable name. This is frequently used for stack traces in crash reports.

oj 2.x adds ojc.symbolicate(str, squeezeMap) as API. This function replaces all $oj_… identifiers in a string with the human-readable name. If the optional squeezeMap parameter is provided, squeezed $oj$… identifiers are also transformed:

let ojc = require("ojc");

let a = ojc.symbolicate("$oj_c_Foo, $oj_c_Bar");                 // "Foo, Bar"
let a = ojc.symbolicate("$oj_p_TheProtocol");                    // "TheProtocol"
let b = ojc.symbolicate("Exception in $oj_f_stringWithString_"); // "Exception in stringWithString:"
let c = ojc.symbolicate("$oj_i__anIvar");                        // "_anIvar"

// Normally, the 'squeeze' property on the compiler result object would be used for squeezeMap
let squeezeMap = { "$oj$a": "$oj_f_stringWithString_" };
let e = ojc.symbolicate("Exception in $oj$a", squeezeMap); // "Exception in stringWithString:"

Acknowledgements

oj uses a modified version of Esprima for parsing and TypeScript for type checking.

License

runtime.js is public domain.

All other files in this project are licensed under the MIT license.