Bun FFI Native Binding Skill

Build native extensions for JavaScript using Bun's tight integration with Zig and C via FFI.

When to Use

• Hot paths: Compute-intensive operations (crypto, compression, math)
• System integration: Direct OS/hardware access
• Large data processing: Batch operations on arrays/buffers
• Legacy libraries: Wrap existing C/Zig libraries

Two Approaches

1. Zig Bindgen (Recommended)

Zig functions compiled directly into Bun with zero-overhead bindings.

Setup:

bun add -d @zig/build

Zig function (

src/math.zig

):

const std = @import("std");
const jsc = @import("jsc");

pub fn add(global: *jsc.JSGlobalObject, a: i32, b: i32) !i32 {
    return std.math.add(i32, a, b) catch {
        return global.throwPretty("Integer overflow", .{});
    };
}

Binding declaration (

src/bindings.ts

):

import { t, fn } from "bindgen";

export const add = fn({
  args: { global: t.globalObject, a: t.i32, b: t.i32 },
  ret: t.i32
});

Usage (

index.ts

):

import { add } from "bun:math";
console.log(add(2, 3)); // 5

2. C FFI (Dynamic Loading)

Load C libraries at runtime without compilation.

C function (

lib.c

):

int add(int a, int b) {
    return a + b;
}

Compile:

gcc -shared -fPIC -o lib.so lib.c

Load in Bun (

index.ts

):

import { dlopen, FFIType } from "bun:ffi";

const lib = dlopen("./lib.so", {
  add: { args: [FFIType.i32, FFIType.i32], returns: FFIType.i32 }
});

console.log(lib.symbols.add(2, 3)); // 5

Performance Considerations

Bridge Cost

• Overhead: 10-100 nanoseconds per call
• Dominates: Tiny functions called repeatedly
• Solution: Batch operations

Data Conversion

• Overhead: Proportional to payload size
• Dominates: Complex object marshaling
• Solution: Use typed arrays, avoid JSON

Rule of Thumb

If work per call > bridge cost → native wins

Critical Edge Cases

See references/EDGE_CASES.md for:

• Exception boundaries (panics crash runtime)
• Memory ownership (who frees allocations?)
• Struct alignment (layout assumptions)
• GC interaction (pinning references)
• Thread safety (event loop constraints)
• ABI compatibility (C calling convention)

Best Practices

1. Minimize boundary crossings — batch processing in native code
2. Use typed arrays — zero-copy buffer mapping
3. Avoid per-call allocation — reuse buffers
4. Binary formats — faster than JSON serialization
5. Stable APIs — version struct layouts
6. Error handling — convert panics to JS exceptions

Optimization Checklist

• Minimize JS → native calls
• Avoid JSON across boundary
• Use typed arrays/buffers
• Batch processing in native
• Convert errors to JS exceptions
• No Zig panics escape to JS
• No global mutable state
• Benchmark boundary latency
• No per-call memory allocation
• Thread safety verified

Example: Batch Array Processing

Zig (

src/process.zig

):

pub fn processArray(global: *jsc.JSGlobalObject, ptr: [*]u32, len: usize) !u32 {
    var sum: u32 = 0;
    for (0..len) |i| {
        sum +|= ptr[i];
    }
    return sum;
}

JS (

index.ts

):

const data = new Uint32Array([1, 2, 3, 4, 5]);
const sum = processArray(data.buffer, data.length);

This avoids 5 separate JS→native calls and marshals data once.

See Also

• Complete SDK Reference
• Zig Documentation
• Bun FFI Docs