'. '

Bck2BrwsrMangling

From APIDesign

Jump to: navigation, search

When translating the ByteCode to JavaScript the Bck2Brwsr project needs to face a common problem. One needs to find Good Name for meaningful objects in the old world (aka. Java) inside the new world (in this case JavaScript). This happens whenever one is translating a typed language (like Java or C++) to untyped (like JavaScript and C) one needs to mange the names, so the new names still allow to express method and field overloading.

Contents

Like JNI

There is a common mangling scheme specified by JNI for C. The Bck2Brwsr projects mimics the specification as closely as possible and extends it only when the JVM compatibility differs from Java source compatibility and requires different treatment. The mangling is based on underscore encoding substitution.

Fully Qualified Names

Fully qualified name uses '_' to separate package names and class name. The global virtual machine object vm has methods to obtain all (once referenced) classes. One can get reference to String as:

var clazz = vm.java_lang_String(false);

Methods

  • There is "__" after name of a method and before its arguments
  • return type is encoded first, parameters follow
  • If there is an '_' in the name or argument segment, it gets replaced by "_1"
  • array signatures start with '[' - such character is replaced by "_3"
  • object signatures end with ';' - that character is replaced by "_2"

As a result to call method String.substring(int, int) - e.g. a method that return string and takes two integers as arguments -it be written as:

var s = "...";
var r = s.substring__Ljava_lang_String_2II(0, 5);

Static Method

When calling a static method, one first needs to obtain the name of a class. The class is made available in a global object called "vm". As such calling String.valueOf(10) is translated to:

var clazz = vm.java_lang_String(false);
var r = clazz.valueOf__Ljava_lang_String_2I(10);

Accessing an Instance Field

To support subclasses defining the same field (like in case of InheritanceA and InheritanceB classes) the Bck2Brwsr needed to create accessor method to access each field in its declaring class. The accessor which prefixes the name of the field with "_" (forming a name that can't clash with mangled method names). The proper way to access field value defined in String class would then be:

var getValue = vm.java_lang_String(true)._value.call(this);
var newValue = "...";
vm.java_lang_String(true)._value.call(this, newValue);

This can often be simplified to:

var getValue = this._value();
var newValue = "...";
this._value(newValue);

which does the same in most cases. Only when there is a subclass defining its own field value, the result would not be correct. Thus this kind of usage is appropriate when one knows the class is final and can't be subclassed (another reason to follow the ClientAPI advice).

Accessing a Static Field

Static fields are wrapped by accessors as well. However accessing them is simpler as they don't need proper this argument. One can use:

var getValue = vm.java_lang_String(true)._CASE_INSENSITIVE_ORDER();
var newValue = "...";
vm.java_lang_String(true)._CASE_INSENSITIVE_ORDER(newValue);

Possible Future Work

There is an experimental branch which replaces well known object types with a single letter:

  • "Ljava_lang_String_2" would become 's'
  • "Ljava_lang_Object_2" would become 'o'

Applying this pattern would shorten the generated JavaScript code. The experiments done so far, however have not yield too convincing results.

Personal tools
buy