ECMAScript to C compiler.
Switch branches/tags
Nothing to show
Clone or download


Descripten is an ECMAScript to C compiler. It consists of two major components: a compiler that compiles ECMAScript source code into C, and a run-time library that's linked into the generated C program.

Ultimately the project goal is to compile ECMAScript into the LLVM language. Integration with LLVM will allow Descripten to utilize the various tools for optimization while also providing a compiler from ECMAScript to machine code without invoking an external C compiler.


Descripten implements ECMAScript 5.1 and is continuously evaluated against the test262 test suite. Currently more than 99.6% of all (11000+) tests passes (tested against revision 309).


Building is supported on GNU/Linux and Mac OS X. Windows is untested at this moment. The following 3rd party libraries are required to build Descripten:

You'll need to build the garbage collector from source because it needs to be configured to find tagged pointers. Descripten uses 64-bit NaN-boxing and the garbage collector should use the following pointer mask: 0xffffffffffff.

There is a patch available in patches/gc/pointer_mask.patch, but simply putting the following into gcconfig.h should work:

#define POINTER_MASK ((GC_word)0xffffffffffff)

You'll also need autotools, pkg-config and a C++11 compliant compiler. To build, simply do the following:

make -j8


Simply run the compiler binary to compile ECMAScript source files into C. Then compile the generated C files with your favorite C compiler and link with the common, parser and run-time libraries.

./compiler file1.js file2.js -o out.c

There is also a utility script that will perform all steps for you:

./ program.js

It will produce program.c, program.o and program which is the final compiled binary. As the script name suggests it will also run the program for you.


Below are a few technical highlights that someone might find interesting.


In ECMAScript a value can be of many different types. To represent a value internally Descripten uses a technique called NaN-boxing. By utilizing the fact that double precision floating point (IEEE 754-1985) Not-a-Number values can hide a 52-bit payload, Descripten can essentially represent all ECMAScript value types using only 64-bits.

Property cache

The naive way of looking up a property on an object is to search the object properties at run-time, using a hash map or similar. Property lookups are however rather common which calls for a more efficient approach. Descripten tries to classify objects at run-time, giving them an internal type in order to use a property cache for fast property access.


Since the ECMAScript language contains facilities for evaluating the language itself (the eval function), Descripten also contains an ECMAScript interpreter. This interpreter is very rudimentary and not much effort has been put into making it fast. It evaluates the AST using a standard visitor pattern. It has merely been added for completeness and to be able to run the test262 test suite.


This project is a work in progress. Below you'll find a few ideas that I would like to see implemented.

Use static single assignment form for internal code representation

Descripten compiles ECMAScript into an internal intermediate representation before producing the C code. The intermediate representation does not currently conform to Single Static Assignment form. However, it would be beneficial if it did since it would make program analysis and optimization easier.

Perform static type inference

One benefit of ahead-of-time compilers is that they can afford spending more time on static program analysis. Descripten should try to analyze the program and infer value types. This would result in a great performance boost: values can be unboxed and faster specialized routines can be used when operating on the values.

Implement a generational, moving and precise garbage collector

Descripten currently uses a generic conservative garbage collector designed for C and C++. It's not well tuned for this project. By using a generational and moving garbage collector it should be possible to achieve both faster memory allocations and faster collection cycles.