QuickJS Javascript Engine

Table of Contents

1 Introduction

QuickJS is a small and embeddable Javascript engine. It supports the ES2019 specification 1 including modules, asynchronous generators and proxies.

It optionally supports mathematical extensions such as big integers (BigInt), big floating point numbers (BigFloat) and operator overloading.

1.1 Main Features

2 Usage

2.1 Installation

A Makefile is provided to compile the engine on Linux or MacOS/X. A preliminary Windows support is available thru cross compilation on a Linux host with the MingGW tools.

Edit the top of the Makefile if you wish to select specific options then run make.

You can type make install as root if you wish to install the binaries and support files to /usr/local (this is not necessary to use QuickJS).

2.2 Quick start

qjs is the command line interpreter (Read-Eval-Print Loop). You can pass Javascript files and/or expressions as arguments to execute them: 

./qjs examples/hello.js

qjsc is the command line compiler: 

./qjsc -o hello examples/hello.js
./hello

generates a hello executable with no external dependency.

qjsbn and qjscbn are the corresponding interpreter and compiler with the mathematical extensions: 

./qjsbn examples/pi.js 1000

displays 1000 digits of PI. 

./qjsbnc -o pi examples/pi.js
./pi 1000

compiles and executes the PI program.

2.3 Command line options

2.3.1 qjs interpreter

usage: qjs [options] [files]

Options are:

-h
--help

List options.

-e EXPR
--eval EXPR

Evaluate EXPR.

-i
--interactive

Go to interactive mode (it is not the default when files are provided on the command line).

-m
--module

Load as ES6 module (default=autodetect).

--script

Load as ES6 script (default=autodetect).

Advanced options are:

-d
--dump

Dump the memory usage stats.

-q
--quit

just instantiate the interpreter and quit.

2.3.2 qjsc compiler

usage: qjsc [options] [files]

Options are:

-c

Only output bytecode in a C file. The default is to output an executable file.

-e

Output main() and bytecode in a C file. The default is to output an executable file.

-o output

Set the output filename (default = out.c or a.out).

-N cname

Set the C name of the generated data.

-m

Compile as Javascript module (default=autodetect).

-M module_name[,cname]

Add initialization code for an external C module. See the c_module example.

-x

Byte swapped output (only used for cross compilation).

-flto

Use link time optimization. The compilation is slower but the executable is smaller and faster. This option is automatically set when the -fno-x options are used.

-fno-[eval|string-normalize|regexp|json|proxy|map|typedarray|promise]

Disable selected language features to produce a smaller executable file.

2.4 qjscalc application

The qjscalc application is a superset of the qjsbn command line interpreter implementing a Javascript calculator with arbitrarily large integer and floating point numbers, fractions, complex numbers, polynomials and matrices. The source code is in qjscalc.js. More documentation and a web version are available at http://numcalc.com.

2.5 Built-in tests

Run make test to run the few built-in tests included in the QuickJS archive.

2.6 Test262 (ECMAScript Test Suite)

A test262 runner is included in the QuickJS archive.

For reference, the full test262 tests are provided in the archive qjs-tests-yyyy-mm-dd.tar.xz. You just need to untar it into the QuickJS source code directory.

Alternatively, the test262 tests can be installed with: 

git clone https://github.com/tc39/test262.git test262
cd test262
patch -p1 < ../tests/test262.patch
cd ..

The patch adds the implementation specific harness functions and optimizes the inefficient RegExp character classes and Unicode property escapes tests (the tests themselves are not modified, only a slow string initialization function is optimized).

The tests can be run with 

make test2

The configuration files test262.conf (resp test262bn.conf for the bignum version, test262o.conf for the old ES5.1 tests4)) contain the options to run the various tests. Tests can be excluded based on features or filename.

The file test262_errors.txt contains the current list of errors. The runner displays a message when a new error appears or when an existing error is corrected or modified. Use the -u option to update the current list of errors (or make test2-update).

The file test262_report.txt contains the logs of all the tests. It is useful to have a clearer analysis of a particular error. In case of crash, the last line corresponds to the failing test.

Use the syntax ./run-test262 -c test262.conf -f filename.js to run a single test. Use the syntax ./run-test262 -c test262.conf N to start testing at test number N.

For more information, run ./run-test262 to see the command line options of the test262 runner.

run-test262 accepts the -N option to be invoked from test262-harness5 thru eshost. Unless you want to compare QuickJS with other engines under the same conditions, we do not recommend to run the tests this way as it is much slower (typically half an hour instead of about 100 seconds).

3 Specifications

3.1 Language support

3.1.1 ES2019 support

The ES2019 specification is almost fully supported including the Annex B (legacy web compatibility) and the Unicode related features.

The following features are not supported yet:

3.1.2 JSON

The JSON parser is currently more tolerant than the specification.

3.1.3 ECMA402

ECMA402 (Internationalization API) is not supported.

3.1.4 Extensions

3.1.5 Mathematical extensions

The mathematical extensions are available in the qjsbn version and are fully backward compatible with standard Javascript. See jsbignum.pdf for more information.

3.2 Modules

ES6 modules are fully supported. The default name resolution is the following:

3.3 Standard library

The standard library is included by default in the command line interpreter. It contains the two modules std and os and a few global objects.

3.3.1 Global objects

scriptArgs

Provides the command line arguments. The first argument is the script name.

print(...args)

Print the arguments separated by spaces and a trailing newline.

console.log(...args)

Same as print().

3.3.2 std module

The std module provides wrappers to the libc stdlib.h and stdio.h and a few other utilities.

Available exports:

exit(n)

Exit the process.

evalScript(str)

Evaluate the string str as a script (global eval).

loadScript(filename)

Evaluate the file filename as a script (global eval).

Error(errno)

std.Error constructor. Error instances contain the field errno (error code) and message (result of std.Error.strerror(errno)).

The constructor contains the following fields:

EINVAL
EIO
EACCES
EEXIST
ENOSPC
ENOSYS
EBUSY
ENOENT
EPERM
EPIPE

Integer value of common errors (additional error codes may be defined).

strerror(errno)

Return a string that describes the error errno.

open(filename, flags)

Open a file (wrapper to the libc fopen()). Throws std.Error in case of I/O error.

popen(command, flags)

Open a process by creating a pipe (wrapper to the libc popen()). Throws std.Error in case of I/O error.

fdopen(fd, flags)

Open a file from a file handle (wrapper to the libc fdopen()). Throws std.Error in case of I/O error.

tmpfile()

Open a temporary file. Throws std.Error in case of I/O error.

puts(str)

Equivalent to std.out.puts(str).

printf(fmt, ...args)

Equivalent to std.out.printf(fmt, ...args)

sprintf(fmt, ...args)

Equivalent to the libc sprintf().

in
out
err

Wrappers to the libc file stdin, stdout, stderr.

SEEK_SET
SEEK_CUR
SEEK_END

Constants for seek().

gc()

Manually invoke the cycle removal algorithm. The cycle removal algorithm is automatically started when needed, so this function is useful in case of specific memory constraints or for testing.

getenv(name)

Return the value of the environment variable name or undefined if it is not defined.

urlGet(url, options = undefined)

Download url using the curl command line utility. options is an optional object containing the following optional properties:

binary

Boolean (default = false). If true, the response is an ArrayBuffer instead of a string. When a string is returned, the data is assumed to be UTF-8 encoded.

full

Boolean (default = false). If true, return the an object contains the properties response (response content), responseHeaders (headers separated by CRLF), status (status code). If full is false, only the response is returned if the status is between 200 and 299. Otherwise an std.Error exception is raised.

FILE prototype:

close()

Close the file.

puts(str)

Outputs the string with the UTF-8 encoding.

printf(fmt, ...args)

Formatted printf, same formats as the libc printf.

flush()

Flush the buffered file.

seek(offset, whence)

Seek to a give file position (whence is std.SEEK_*). Throws a std.Error in case of I/O error.

tell()

Return the current file position.

eof()

Return true if end of file.

fileno()

Return the associated OS handle.

read(buffer, position, length)

Read length bytes from the file to the ArrayBuffer buffer at byte position position (wrapper to the libc fread).

write(buffer, position, length)

Write length bytes to the file from the ArrayBuffer buffer at byte position position (wrapper to the libc fread).

getline()

Return the next line from the file, assuming UTF-8 encoding, excluding the trailing line feed.

getByte()

Return the next byte from the file.

putByte(c)

Write one byte to the file.

3.3.3 os module

The os module provides Operating System specific functions:

The OS functions usually return 0 if OK or an OS specific negative error code.

Available exports:

open(filename, flags, mode = 0o666)

Open a file. Return a handle or < 0 if error.

O_RDONLY
O_WRONLY
O_RDWR
O_APPEND
O_CREAT
O_EXCL
O_TRUNC

POSIX open flags.

O_TEXT

(Windows specific). Open the file in text mode. The default is binary mode.

close(fd)

Close the file handle fd.

seek(fd, offset, whence)

Seek in the file. Use std.SEEK_* for whence.

read(fd, buffer, offset, length)

Read length bytes from the file handle fd to the ArrayBuffer buffer at byte position offset. Return the number of read bytes or < 0 if error.

write(fd, buffer, offset, length)

Write length bytes to the file handle fd from the ArrayBuffer buffer at byte position offset. Return the number of written bytes or < 0 if error.

isatty(fd)

Return true is fd is a TTY (terminal) handle.

ttyGetWinSize(fd)

Return the TTY size as [width, height] or null if not available.

ttySetRaw(fd)

Set the TTY in raw mode.

remove(filename)

Remove a file. Return 0 if OK or < 0 if error.

rename(oldname, newname)

Rename a file. Return 0 if OK or < 0 if error.

realpath(path)

Return [str, err] where str is the canonicalized absolute pathname of path and err the error code.

getcwd()

Return [str, err] where str is the current working directory and err the error code.

mkdir(path, mode = 0o777)

Create a directory at path. Return the error code.

stat(path)
lstat(path)

Return [obj, err] where obj is an object containing the file status of path. err is the error code. The following fields are defined in obj: dev, ino, mode, nlink, uid, gid, rdev, size, blocks, atime, mtime, ctime. The times are specified in milliseconds since 1970. lstat() is the same as stat() excepts that it returns information about the link itself.

S_IFMT
S_IFIFO
S_IFCHR
S_IFDIR
S_IFBLK
S_IFREG
S_IFSOCK
S_IFLNK
S_ISGID
S_ISUID

Constants to interpret the mode property returned by stat(). They have the same value as in the C system header sys/stat.h.

utimes(path, atime, mtime)

Change the access and modification times of the file path. The times are specified in milliseconds since 1970.

symlink(target, linkpath)

Create a link at linkpath containing the string target.

readlink(path)

Return [str, err] where str is the link target and err the error code.

readdir(path)

Return [array, err] where array is an array of strings containing the filenames of the directory path. err is the error code.

setReadHandler(fd, func)

Add a read handler to the file handle fd. func is called each time there is data pending for fd. A single read handler per file handle is supported. Use func = null to remove the hander.

setWriteHandler(fd, func)

Add a write handler to the file handle fd. func is called each time data can be written to fd. A single write handler per file handle is supported. Use func = null to remove the hander.

signal(signal, func)

Call the function func when the signal signal happens. Only a single handler per signal number is supported. Use null to set the default handler or undefined to ignore the signal.

SIGINT
SIGABRT
SIGFPE
SIGILL
SIGSEGV
SIGTERM

POSIX signal numbers.

kill(pid, sig)

Send the signal sig to the process pid.

exec(args[, options])

Execute a process with the arguments args. options is an object containing optional parameters:

block

Boolean (default = true). If true, wait until the process is termined. In this case, exec return the exit code if positive or the negated signal number if the process was interrupted by a signal. If false, do not block and return the process id of the child.

usePath

Boolean (default = true). If true, the file is searched in the PATH environment variable.

file

String (default = args[0]). Set the file to be executed.

cwd

String. If present, set the working directory of the new process.

stdin
stdout
stderr

If present, set the handle in the child for stdin, stdout or stderr.

waitpid(pid, options)

waitpid Unix system call. Return the array [ret, status].

WNOHANG

Constant for the options argument of waitpid.

dup(fd)

dup Unix system call.

dup2(oldfd, newfd)

dup2 Unix system call.

pipe()

pipe Unix system call. Return two handles as [read_fd, write_fd] or null in case of error.

sleep(delay_ms)

Sleep during delay_ms milliseconds.

setTimeout(func, delay)

Call the function func after delay ms. Return a handle to the timer.

clearTimeout(handle)

Cancel a timer.

platform

Return a string representing the platform: "linux", "darwin", "win32" or "js".

3.4 QuickJS C API

The C API was designed to be simple and efficient. The C API is defined in the header quickjs.h.

3.4.1 Runtime and contexts

JSRuntime represents a Javascript runtime corresponding to an object heap. Several runtimes can exist at the same time but they cannot exchange objects. Inside a given runtime, no multi-threading is supported.

JSContext represents a Javascript context (or Realm). Each JSContext has its own global objects and system objects. There can be several JSContexts per JSRuntime and they can share objects, similary to frames of the same origin sharing Javascript objects in a web browser.

3.4.2 JSValue

JSValue represents a Javascript value which can be a primitive type or an object. Reference counting is used, so it is important to explicitely duplicate (JS_DupValue(), increment the reference count) or free (JS_FreeValue(), decrement the reference count) JSValues.

3.4.3 C functions

C functions can be created with JS_NewCFunction(). JS_SetPropertyFunctionList() is a shortcut to easily add functions, setters and getters properties to a given object.

Unlike other embedded Javascript engines, there is no implicit stack, so C functions get their parameters as normal C parameters. As a general rule, C functions take constant JSValues as parameters (so they don’t need to free them) and return a newly allocated (=live) JSValue.

3.4.4 Exceptions

Exceptions: most C functions can return a Javascript exception. It must be explicitely tested and handled by the C code. The specific JSValue JS_EXCEPTION indicates that an exception occured. The actual exception object is stored in the JSContext and can be retrieved with JS_GetException().

3.4.5 Script evaluation

Use JS_Eval() to evaluate a script or module source.

If the script or module was compiled to bytecode with qjsc, it can be evaluated by calling js_std_eval_binary(). The advantage is that no compilation is needed so it is faster and smaller because the compiler can be removed from the executable if no eval is required.

Note: the bytecode format is linked to a given QuickJS version. Moreover, no security check is done before its execution. Hence the bytecode should not be loaded from untrusted sources. That’s why there is no option to output the bytecode to a binary file in qjsc.

3.4.6 JS Classes

C opaque data can be attached to a Javascript object. The type of the C opaque data is determined with the class ID (JSClassID) of the object. Hence the first step is to register a new class ID and JS class (JS_NewClassID(), JS_NewClass()). Then you can create objects of this class with JS_NewObjectClass() and get or set the C opaque point with JS_GetOpaque()/JS_SetOpaque().

When defining a new JS class, it is possible to declare a finalizer which is called when the object is destroyed. A gc_mark method can be provided so that the cycle removal algorithm can find the other objects referenced by this object. Other methods are available to define exotic object behaviors.

The Class ID are globally allocated (i.e. for all runtimes). The JSClass are allocated per JSRuntime. JS_SetClassProto() is used to define a prototype for a given class in a given JSContext. JS_NewObjectClass() sets this prototype in the created object.

Examples are available in quickjs-libc.c.

3.4.7 C Modules

Native ES6 modules are supported and can be dynamically or statically linked. Look at the test_bjson and bjson.so examples. The standard library quickjs-libc.c is also a good example of a native module.

3.4.8 Memory handling

Use JS_SetMemoryLimit() to set a global memory allocation limit to a given JSRuntime.

Custom memory allocation functions can be provided with JS_NewRuntime2().

The maximum system stack size can be set with JS_SetMaxStackSize().

3.4.9 Execution timeout and interrupts

Use JS_SetInterruptHandler() to set a callback which is regularly called by the engine when it is executing code. This callback can be used to implement an execution timeout.

It is used by the command line interpreter to implement a Ctrl-C handler.

4 Internals

4.1 Bytecode

The compiler generates bytecode directly with no intermediate representation such as a parse tree, hence it is very fast. Several optimizations passes are done over the generated bytecode.

A stack-based bytecode was chosen because it is simple and generates compact code.

For each function, the maximum stack size is computed at compile time so that no runtime stack overflow tests are needed.

A separate compressed line number table is maintained for the debug information.

Access to closure variables is optimized and is almost as fast as local variables.

Direct eval in strict mode is optimized.

4.2 Executable generation

4.2.1 qjsc compiler

The qjsc compiler generates C sources from Javascript files. By default the C sources are compiled with the system compiler (gcc or clang).

The generated C source contains the bytecode of the compiled functions or modules. If a full complete executable is needed, it also contains a main() function with the necessary C code to initialize the Javascript engine and to load and execute the compiled functions and modules.

Javascript code can be mixed with C modules.

In order to have smaller executables, specific Javascript features can be disabled, in particular eval or the regular expressions. The code removal relies on the Link Time Optimization of the system compiler.

4.2.2 Binary JSON

qjsc works by compiling scripts or modules and then serializing them to a binary format. A subset of this format (without functions or modules) can be used as binary JSON. The example test_bjson.js shows how to use it.

Warning: the binary JSON format may change without notice, so it should not be used to store persistent data. The test_bjson.js example is only used to test the binary object format functions.

4.3 Runtime

4.3.1 Strings

Strings are stored either as an 8 bit or a 16 bit array of characters. Hence random access to characters is always fast.

The C API provides functions to convert Javascript Strings to C UTF-8 encoded strings. The most common case where the Javascript string contains only ASCII characters involves no copying.

4.3.2 Objects

The object shapes (object prototype, property names and flags) are shared between objects to save memory.

Arrays with no holes (except at the end of the array) are optimized.

TypedArray accesses are optimized.

4.3.3 Atoms

Object property names and some strings are stored as Atoms (unique strings) to save memory and allow fast comparison. Atoms are represented as a 32 bit integer. Half of the atom range is reserved for immediate integer literals from 0 to 2^{31}-1.

4.3.4 Numbers

Numbers are represented either as 32-bit signed integers or 64-bit IEEE-754 floating point values. Most operations have fast paths for the 32-bit integer case.

4.3.5 Garbage collection

Reference counting is used to free objects automatically and deterministically. A separate cycle removal pass is done when the allocated memory becomes too large. The cycle removal algorithm only uses the reference counts and the object content, so no explicit garbage collection roots need to be manipulated in the C code.

4.3.6 JSValue

It is a Javascript value which can be a primitive type (such as Number, String, ...) or an Object. NaN boxing is used in the 32-bit version to store 64-bit floating point numbers. The representation is optimized so that 32-bit integers and reference counted values can be efficiently tested.

In 64-bit code, JSValue are 128-bit large and no NaN boxing is used. The rationale is that in 64-bit code memory usage is less critical.

In both cases (32 or 64 bits), JSValue exactly fits two CPU registers, so it can be efficiently returned by C functions.

4.3.7 Function call

The engine is optimized so that function calls are fast. The system stack holds the Javascript parameters and local variables.

4.4 RegExp

A specific regular expression engine was developped. It is both small and efficient and supports all the ES2020 features including the Unicode properties. As the Javascript compiler, it directly generates bytecode without a parse tree.

Backtracking with an explicit stack is used so that there is no recursion on the system stack. Simple quantizers are specifically optimized to avoid recursions.

Infinite recursions coming from quantizers with empty terms are avoided.

The full regexp library weights about 15 KiB (x86 code), excluding the Unicode library.

4.5 Unicode

A specific Unicode library was developped so that there is no dependency on an external large Unicode library such as ICU. All the Unicode tables are compressed while keeping a reasonnable access speed.

The library supports case conversion, Unicode normalization, Unicode script queries, Unicode general category queries and all Unicode binary properties.

The full Unicode library weights about 45 KiB (x86 code).

4.6 BigInt and BigFloat

BigInt and BigFloat are implemented with the libbf library7. It weights about 60 KiB (x86 code) and provides arbitrary precision IEEE 754 floating point operations and transcendental functions with exact rounding.

5 License

QuickJS is released under the MIT license.

Unless otherwise specified, the QuickJS sources are copyright Fabrice Bellard and Charlie Gordon.

Footnotes

(1)

https://www.ecma-international.org/ecma-262/10.0

(2)

https://github.com/tc39/test262

(3)

https://tc39.github.io/ecma262/

(4)

The old ES5.1 tests can be extracted with git clone --single-branch --branch es5-tests https://github.com/tc39/test262.git test262o

(5)

https://github.com/bterlson/test262-harness

(6)

We believe the current specification of tails calls is too complicated and presents limited practical interests.

(7)

https://bellard.org/libbf