Skip to content

zeeeeeeeeeeeen/javy

 
 

Repository files navigation

Javy: A JavaScript to WebAssembly toolchain

Build status

About this repo

Introduction: Run your JavaScript on WebAssembly. Javy takes your JavaScript code, and executes it in a WebAssembly embedded JavaScript runtime.

Contributing

We welcome feedback, bug reports and bug fixes. We're also happy to discuss feature development but please discuss the features in an issue before contributing. All contributors will be prompted to sign our CLA.

Build

  • rustup
  • Stable Rust (rustup install stable && rustup default stable)
  • wasm32-wasi, can be installed via rustup target add wasm32-wasi
  • cmake, depending on your operating system and architecture, it might not be installed by default. On Mac it can be installed with homebrew via brew install cmake
  • Rosetta 2 if running MacOS on Apple Silicon, can be installed via softwareupdate --install-rosetta
  • Install the wasi-sdk by running make download-wasi-sdk

Development

  • wasmtime-cli, can be installed via cargo install wasmtime-cli (required for cargo-wasi)
  • cargo-wasi, can be installed via cargo install cargo-wasi
  • wizer, can be installed via cargo install wizer --all-features

Building

After all the dependencies are installed, run make. You should now have access to the executable in target/release/javy

Alternatively you can run make && cargo install --path crates/cli. After running the previous command you'll have a global installation of the executable.

Compiling to WebAssembly

Define your JavaScript like:

// Read input from stdin
const input = readInput();
// Call the function with the input
const result = foo(input);
// Write the result to stdout
writeOutput(result);

// The main function.
function foo(input) {
    return { foo: input.n + 1, newBar: input.bar + "!" };
}

// Read input from stdin
function readInput() {
    const chunkSize = 1024;
    const inputChunks = [];
    let totalBytes = 0;

    // Read all the available bytes
    while (1) {
        const buffer = new Uint8Array(chunkSize);
        // Stdin file descriptor
        const fd = 0;
        const bytesRead = Javy.IO.readSync(fd, buffer);

        totalBytes += bytesRead;
        if (bytesRead === 0) {
            break;
        }
        inputChunks.push(buffer.subarray(0, bytesRead));
    }

    // Assemble input into a single Uint8Array
    const { finalBuffer } = inputChunks.reduce((context, chunk) => {
        context.finalBuffer.set(chunk, context.bufferOffset);
        context.bufferOffset += chunk.length;
        return context;
    }, { bufferOffset: 0, finalBuffer: new Uint8Array(totalBytes) });

    return JSON.parse(new TextDecoder().decode(finalBuffer));
}

// Write output to stdout
function writeOutput(output) {
    const encodedOutput = new TextEncoder().encode(JSON.stringify(output));
    const buffer = new Uint8Array(encodedOutput);
    // Stdout file descriptor
    const fd = 1;
    Javy.IO.writeSync(fd, buffer);
}

Create a WebAssembly binary from your JavaScript by:

javy compile index.js -o destination/index.wasm

For more information on the commands you can run javy --help

You can then execute your WebAssembly binary using a WebAssembly engine:

$ echo '{ "n": 2, "bar": "baz" }' | wasmtime index.wasm
{"foo":3,"newBar":"baz!"}%   

Invoking Javy-generated modules programatically

Javy-generated modules are by design WASI only and follow the command pattern. Any input must be passed via stdin and any output will be placed in stdout. This is especially important when invoking Javy modules from a custom embedding.

In a runtime like Wasmtime, wasmtime-wasi can be used to set the input and retrieve the output.

Creating and using dynamically linked modules

In some scenarios, you may want or need to generate much smaller Wasm modules with Javy. Using the -d flag when invoking Javy will create a dynamically linked module which will have a much smaller file size than a statically linked module. Statically linked modules embed the JS engine inside the module while dynamically linked modules rely on Wasm imports to provide the JS engine. Dynamically linked modules have special requirements that statically linked modules do not and will not execute in WebAssembly runtimes that do not meet these requirements.

To successfully instantiate and run a dynamically linked Javy module, the execution environment must provide a javy_quickjs_provider_v1 namespace for importing that links to the exports provided by the javy_quickjs_provider.wasm module. Dynamically linked modules cannot be instantiated in environments that do not provide this import.

Dynamically linked Javy modules are tied to QuickJS since they use QuickJS's bytecode representation.

Obtaining the QuickJS provider module

The javy_quickjs_provider.wasm module is available as an asset on the Javy release you are using. It can also be obtained by running javy emit-provider -o <path> to write the module into <path>.

Creating and running a dynamically linked module on the CLI

$ echo 'console.log("hello world!");' > my_code.js
$ javy compile -d -o my_code.wasm my_code.js
$ javy emit-provider -o provider.wasm
$ wasmtime run --preload javy_quickjs_provider_v1=provider.wasm my_code.wasm
hello world!

Using quickjs-wasm-rs to build your own toolchain

The quickjs-wasm-rs crate that is part of this project can be used as part of a Rust crate targeting Wasm to customize how that Rust crate interacts with QuickJS. This may be useful when trying to use JavaScript inside a Wasm module and Javy does not fit your needs as quickjs-wasm-rs contains serializers that make it easier to send structured data (for example, strings or objects) between host code and Wasm code.

Releasing

  1. Create a tag for the new version like v0.2.0
git tag v0.2.0
git push origin --tags
  1. Create a new release from the new tag in github here.
  2. A GitHub Action will trigger for publish.yml when a release is published (i.e. it doesn't run on drafts), creating the artifacts for downloading.

About

JS to WebAssembly toolchain

Resources

License

Code of conduct

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C 87.5%
  • JavaScript 7.3%
  • Rust 4.5%
  • Other 0.7%