Foreign Function Interface (FFI)

Deno's Foreign Function Interface (FFI) allows JavaScript and TypeScript code to call functions in dynamic libraries written in languages like C, C++, or Rust. This enables you to integrate native code performance and capabilities directly into your Deno applications.

Deno FFI Reference Docs

Introduction to FFI Jump to heading#

FFI provides a bridge between Deno's JavaScript runtime and native code. This allows you to:

• Use existing native libraries within your Deno applications
• Implement performance-critical code in languages like Rust or C
• Access operating system APIs and hardware features not directly available in JavaScript

Deno's FFI implementation is based on the Deno.dlopen API, which loads dynamic libraries and creates JavaScript bindings to the functions they export.

Security considerations Jump to heading#

FFI requires explicit permission using the --allow-ffi flag, as native code runs outside of Deno's security sandbox:

deno run --allow-ffi my_ffi_script.ts

Basic usage Jump to heading#

The basic pattern for using FFI in Deno involves:

1. Defining the interface for the native functions you want to call
2. Loading the dynamic library using Deno.dlopen()
3. Calling the loaded functions

Here's a simple example loading a C library:

const dylib = Deno.dlopen("libexample.so", {
  add: { parameters: ["i32", "i32"], result: "i32" },
});

console.log(dylib.symbols.add(5, 3)); // 8

dylib.close();

Loading libraries Jump to heading#

The first argument to Deno.dlopen() is the path to the dynamic library. Deno hands this path to the operating system's dynamic loader (dlopen on Linux and macOS, LoadLibrary on Windows), so the way the file is located follows the OS rules, not Deno's module resolution.

There are two ways to specify the library:

• A path with a separator (for example "./libexample.so" or "/usr/lib/libexample.so") is opened from that exact location. A relative path like "./libexample.so" is resolved against the current working directory of the process, not against the location of your .ts file. This is a common source of "could not open library" errors: running the same script from a different directory changes where Deno looks.
• A bare name with no separator (for example "libexample.so") is left to the OS to find on its standard search path. On Linux that means LD_LIBRARY_PATH and the system cache (such as /usr/lib); on macOS the DYLD_* paths; on Windows the executable's directory, the system directories, and PATH. Use this form for libraries that are already installed system-wide.

Resolving the path relative to your module Jump to heading#

To load a library that ships next to your source file regardless of the current working directory, resolve the path against import.meta.url instead of using a bare relative string:

// Always points at libexample.so sitting next to this module.
const path = new URL("./libexample.so", import.meta.url).pathname;

const dylib = Deno.dlopen(
  path,
  {
    add: { parameters: ["i32", "i32"], result: "i32" },
  } as const,
);

This pattern also works inside an executable produced by deno compile (see below).

Bundling a library with deno compile Jump to heading#

Deno.dlopen needs a real file on disk, so a dynamic library is not embedded in a compiled binary automatically. Include it explicitly with the --include flag:

deno compile --allow-ffi --include libexample.so main.ts

At runtime Deno unpacks the included library to a temporary directory and points import.meta.url at it, so a module that resolves its path with new URL("./libexample.so", import.meta.url).pathname (as shown above) finds the bundled copy and the resulting binary runs on a machine that does not have the library installed. If you do not bundle the library, ship it alongside the executable and load it by a path relative to the binary, or rely on the system search path described above.

Handling load failures Jump to heading#

Deno.dlopen throws synchronously when the library cannot be loaded or when a declared symbol is missing, so wrap the call in a try/catch to fail gracefully:

let dylib;
try {
  dylib = Deno.dlopen(
    "./libexample.so",
    {
      add: { parameters: ["i32", "i32"], result: "i32" },
    } as const,
  );
} catch (err) {
  // A missing or unreadable file reports "Could not open library: ...".
  // A missing function reports "Failed to register symbol <name>: ...".
  console.error("Failed to load native library:", err.message);
  Deno.exit(1);
}

The error message distinguishes the two common cases: the file itself could not be opened (wrong path, missing file, or an architecture mismatch), or the file loaded but one of the symbols you declared was not found in it.

Supported types Jump to heading#

Deno's FFI supports a variety of data types for parameters and return values:

As of Deno 1.25, the pointer type has been split into a pointer and a buffer type to ensure users take advantage of optimizations for Typed Arrays, and as of Deno 1.31 the JavaScript representation of pointer has become an opaque pointer object or null for null pointers.

• [1] void type can only be used as a result type.
• [2] buffer type accepts TypedArrays as parameter, but it always returns a pointer object or null when used as result type like the pointer type.
• [3] function type works exactly the same as the pointer type as a parameter and result type.
• [4] struct type is for passing and returning C structs by value (copy). The struct array must enumerate each of the struct's fields' type in order. The structs are padded automatically: Packed structs can be defined by using an appropriate amount of u8 fields to avoid padding. Only TypedArrays are supported as structs, and structs are always returned as Uint8Arrays.

Working with structs Jump to heading#

To pass or return a C struct by value, describe its layout with { struct: [...] } — an array that lists each field's FFI type in declaration order. Struct values are passed as a TypedArray whose bytes match the C layout, and structs returned by value come back as a Uint8Array of the right length. The struct array in the type table earlier on this page is the authoritative shape.

Suppose you have this small C library that operates on a 2D Point:

typedef struct {
  double x;
  double y;
} Point;

double distance(Point a, Point b) {
  double dx = a.x - b.x;
  double dy = a.y - b.y;
  return __builtin_sqrt(dx * dx + dy * dy);
}

Point midpoint(Point a, Point b) {
  Point m;
  m.x = (a.x + b.x) / 2.0;
  m.y = (a.y + b.y) / 2.0;
  return m;
}

Build it as a shared library. The compiler flags and output filename vary by platform:

cc -shared -fPIC -O2 -o libpoint.so point.c

cc -dynamiclib -O2 -o libpoint.dylib point.c

cl /LD /O2 point.c /Fe:point.dll

Then call into it from Deno, using the filename for your platform in Deno.dlopen. Note that the struct definition is an array of field types in declaration order, not an object with named fields:

// `Point` mirrors the C `struct Point { double x; double y; }`.
const Point = { struct: ["f64", "f64"] } as const;

const lib = Deno.dlopen(
  "./libpoint.so",
  {
    distance: { parameters: [Point, Point], result: "f64" },
    midpoint: { parameters: [Point, Point], result: Point },
  } as const,
);

// Build struct values as a TypedArray whose bytes match the C layout.
// Two f64 fields → two slots in a Float64Array.
const a = new Float64Array([1.0, 2.0]); // Point { x: 1.0, y: 2.0 }
const b = new Float64Array([4.0, 6.0]); // Point { x: 4.0, y: 6.0 }

// FFI reads the underlying bytes, so pass the buffer as a Uint8Array view.
const aBytes = new Uint8Array(a.buffer);
const bBytes = new Uint8Array(b.buffer);

console.log("distance =", lib.symbols.distance(aBytes, bBytes));

// A struct returned by value comes back as a Uint8Array sized to the struct.
// Wrap it in a Float64Array to read the fields back out.
const midBytes = lib.symbols.midpoint(aBytes, bBytes);
const mid = new Float64Array(midBytes.buffer);
console.log("midpoint =", { x: mid[0], y: mid[1] });

lib.close();

Run it with the --allow-ffi permission:

deno run --allow-ffi point.ts

You should see:

distance = 5
midpoint = { x: 2.5, y: 4 }

A few things to keep in mind when working with structs:

• Layout matches the C compiler. Deno pads struct fields the same way your C compiler does. If you need a packed struct, pad it explicitly with u8 fields, as noted in the type table above.
• Field order is positional. The struct array is just types, in declaration order — there are no field names on the JavaScript side. The TypedArray you pass must lay the fields out in the same order.
• Returned structs are bytes. A struct result is always a Uint8Array; view it through the appropriate TypedArray (or a DataView) to read the fields.

Working with callbacks Jump to heading#

You can pass JavaScript functions as callbacks to native code:

const signatures = {
  setCallback: {
    parameters: ["function"],
    result: "void",
  },
  runCallback: {
    parameters: [],
    result: "void",
  },
} as const;

// Create a callback function
const callback = new Deno.UnsafeCallback(
  { parameters: ["i32"], result: "void" } as const,
  (value) => {
    console.log("Callback received:", value);
  },
);

// Pass the callback to the native library
dylib.symbols.setCallback(callback.pointer);

// Later, this will trigger our JavaScript function
dylib.symbols.runCallback();

// Always clean up when done
callback.close();

Best practices with FFI Jump to heading#

1. Always close resources. Close libraries with dylib.close() and callbacks with callback.close() when done.

2. Prefer TypeScript. Use TypeScript for better type-checking when working with FFI.

3. Wrap FFI calls in try/catch blocks to handle errors gracefully.

4. Be extremely careful when using FFI, as native code can bypass Deno's security sandbox.

5. Keep the FFI interface as small as possible to reduce the attack surface.

Examples Jump to heading#

Using a Rust library Jump to heading#

Here's an example of creating and using a Rust library with Deno:

First, create a Rust library:

// lib.rs
#[unsafe(no_mangle)]
pub extern "C" fn fibonacci(n: u32) -> u32 {
  if n <= 1 {
    return n;
  }
  fibonacci(n - 1) + fibonacci(n - 2)
}

Compile it as a dynamic library:

rustc --crate-type cdylib lib.rs

Then use it from Deno:

const libName = {
  windows: "./lib.dll",
  linux: "./liblib.so",
  darwin: "./liblib.dylib",
}[Deno.build.os];

const dylib = Deno.dlopen(
  libName,
  {
    fibonacci: { parameters: ["u32"], result: "u32" },
  } as const,
);

// Calculate the 10th Fibonacci number
const result = dylib.symbols.fibonacci(10);
console.log(`Fibonacci(10) = ${result}`); // 55

dylib.close();

Examples Jump to heading#

• Netsaur
• WebView_deno
• Deno_sdl2
• Deno FFI examples repository

These community-maintained repos includes working examples of FFI integrations with various native libraries across different operating systems.

Related Approaches to Native Code Integration Jump to heading#

While Deno's FFI provides a direct way to call native functions, there are other approaches to integrate native code:

Using Node-API (N-API) with Deno Jump to heading#

Deno supports Node-API (N-API) for compatibility with native Node.js addons. This enables you to use existing native modules written for Node.js.

Directly loading a Node-API addon:

import process from "node:process";
process.dlopen(module, "./native_module.node", 0);

Using an npm package that uses a Node-API addon:

import someNativeAddon from "npm:some-native-addon";
console.log(someNativeAddon.doSomething());

How is this different from FFI?

Node-API support is ideal for leveraging existing Node.js native modules, whereas FFI is best for direct, lightweight calls to native libraries.

Alternatives to FFI Jump to heading#

Before using FFI, consider these alternatives:

• WebAssembly, for portable native code that runs within Deno's sandbox.
• Use Deno.command to execute external binaries and subprocesses with controlled permissions.
• Check whether Deno's native APIs already provide the functionality you need.

Deno's FFI capabilities provide powerful integration with native code, enabling performance optimizations and access to system-level functionality. However, this power comes with significant security considerations. Always be cautious when working with FFI and ensure you trust the native libraries you're using.