Fork me on GitHub


Java on Coffeescript.


Doppio is an implementation of the Java Virtual Machine that can run in just about any reasonable JavaScript engine. The goal is to be able to run JVM programs in the browser without any plug-ins, which means that any Java programs running in it are safely sandboxed by the JavaScript engine. Doppio is written in CoffeeScript, which is a nice terse language that maps directly into JavaScript.

Doppio is also the Italian word for 'double', and is another name for a double espresso.

Doppio started out as the mid-term project for a Graduate Systems Seminar. It has since taken on a life of its own, and is complete enough to run the following programs:

The code has been tested on the latest versions of Chrome, Firefox, Safari, Opera, IE 9, IE 10, and Node, but should run in just about any browser.

Check out the demo!

Notes on Architecture

Library Support

Doppio was built with the goal of supporting as much of the Java Class Library as possible. As the official Java Class Library is mostly implemented in Java, Doppio uses the Java 6 JCL with reimplementations of any needed "native methods", which were originally implemented in C, in CoffeeScript.

Doppio is designed to run both on the console and the browser. In the console implementation, system calls are handled by Node.JS. In the browser, we emulate a simple LocalStorage-backed filesystem with an API very similar to Node's, so the same code can operate in both environments. If the browser does not support LocalStorage, then the filesystem exists purely in memory and is cleared when the user leaves the page.

Since LocalStorage has a storage limit of 5MB, Doppio's web frontend will only store files created or modified by the user or by programs run by the user. We do this to avoid storing massive amounts of JCL classes in the limited storage available to us.


Emulating primitives was slightly tricky, since JavaScript only exposes the 64-bit double as its sole numeric primitive. Technically, you can also coax out 32-bit signed integers, which trivially map into JavaScript doubles.

Floats are slightly tricky; while the common case is handled nicely by a double, there are some edge cases that need to be specifically addressed (mainly, +/- infinity and underflow, which Doppio handles).

64-bit longs are tough. They cannot fit into the 52 bits of precision provided by a double. Fortunately, this problem has already been tackled in the Google Closure library.

Objects (Heap Management)

JVM objects are mapped to JS objects with the same field names, bundled inside a larger object that contains some metadata. Instead of simulating an actual heap, we pass JS object references around. Thus, garbage collection is automatically handled by the JavaScript engine's GC. However, since Java methods like hashCode require each Object to have a unique ID, we store an auto-incremented ref field in each object's metadata that acts as an imaginary heap address.


We currently have a basic thread implementation in Doppio that should be mostly spec-conformant. Since JavaScript is essentially single threaded, only one thread runs at a time. At certain yield points, the running thread will pause and will allow other threads to execute.

At the moment, these yield points occur only at the following times:

We hope to improve thread support in the next release to expand the amount of programs that are compatible with Doppio.


While developing Doppio, we had to balance two concerns: Java code performance, and yielding the JavaScript thread often enough so the JavaScript engine does not get angry at us!

In addition, since the browser DOM is largely asynchronous, we had to emulate some blocking Java operations, such as standard input, in an asynchronous fashion. And if we didn't yield the JavaScript thread after updating the DOM, we quickly discovered that the user would not see the change until the next time Doppio yielded the JavaScript thread.

We solved this by implementing a 'yield' construct: upon encountering a blocking function, we throw a YieldException to pause the VM. This exception would also contain the asynchronous function that we are waiting on, which eventually calls the VM and resumes the program. We used this for standard input operations, among other things.

In the next version of Doppio, we hope to isolate the JVM logic inside of a WebWorker to prevent Doppio from monopolizing the main JavaScript thread. We are already making great progress on this front, and look forward to when it is complete!

Interpreter Design

We have a class for each type of opcode in the JVM. Each bytecode instruction in a method is translated into an object and placed into an array of instructions. Running a method is a matter of jumping to the correct opcode, calling its run method, and repeating until the method completes.

We had experimented with doing a giant switch statement, but we could not get it to perform as well as the one-object-per-instruction design due to the large number of JVM opcodes. We hope to perform more experiments in the future to see if we can further increase the performance of the main interpreter loop.


There is much more to do! We are currently trying to refine the JVM internals before trying to bolt on a compilation engine.

Up next on our roadmap is:

Once we deal with those issues, we'll start looking into per-method JIT compilation and control-flow reconstruction (since we're going from low level bytecode to high level JavaScript). There is already a large body of work on Java decompilation that we can leverage for this task, as well as things like Emscripten's 'Relooper' algorithm. Contributions are definitely welcome!


Doppio uses the jQuery and Underscore.js libraries. Editing is provided by the Ace editor, and the console is a fork based off Chris Done's jquery-console. Layout is based off Twitter's Bootstrap. The font for 'Doppio' is Bitter from Google Web Fonts, and the coffee icon is by Maximilian Becker, from The Noun Project collection.

Doppio itself is the work of CJ Carey, Jez Ng, John Vilk, and Jonny Leahey, and is MIT Licensed.