Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

FFI and Memory Layout

Overview

libxev-go uses FFI (Foreign Function Interface) via the jupiterrider/ffi library to call into libxev (written in Zig) without requiring cgo. This approach provides several benefits:

  • Pure Go build (no C compiler required)
  • Better cross-compilation support
  • Smaller binary size
  • Better goroutine integration

However, FFI requires careful attention to memory layout and struct alignment between Go and Zig.

The Zig Field Reordering Issue

Problem

Zig automatically reorders struct fields by their alignment to minimize padding and ensure optimal memory access. This means fields are not necessarily laid out in memory in the order they are declared.

Example: xev.Options

In libxev’s Zig code, Options is declared as:

pub const Options = struct {
    entries: u32 = 256,
    thread_pool: ?*xev.ThreadPool = null,
};

You might expect the memory layout to be:

[entries: 4 bytes][padding: 4 bytes][thread_pool: 8 bytes]

But Zig reorders fields by alignment, resulting in:

[thread_pool: 8 bytes][entries: 4 bytes][padding: 4 bytes]

The 8-byte pointer comes first, followed by the 4-byte integer.

Go Struct Declaration

To match this layout in Go, you cannot declare fields in source order:

// WRONG - fields in source order
type LoopOptions struct {
    Entries    uint32      // offset 0
    _          uint32      // padding
    ThreadPool *ThreadPool // offset 8
}

Instead, you must declare them in alignment order to match Zig’s layout:

// CORRECT - fields in alignment order
type LoopOptions struct {
    ThreadPool *ThreadPool // offset 0 (8 bytes)
    Entries    uint32      // offset 8 (4 bytes)
    _          uint32      // padding to 16 bytes
}

Debugging Memory Layout Issues

Symptoms

When Go and Zig struct layouts don’t match, you’ll typically see:

  1. SIGSEGV crashes with suspicious addresses like 0x100 (256) or other small offsets
  2. Garbage values when printing struct fields (e.g., expecting 256 but seeing 1687552)
  3. Crashes in memory allocators (libsystem_malloc.dylib)

Diagnostic Approach

  1. Check field offsets in Zig:

    std.debug.print("offsetof(entries): {}\n", .{@offsetOf(xev.Options, "entries")});
    std.debug.print("offsetof(thread_pool): {}\n", .{@offsetOf(xev.Options, "thread_pool")});
    
  2. Check field offsets in Go:

    import "unsafe"
    
    fmt.Printf("Entries offset: %d\n", unsafe.Offsetof(opts.Entries))
    fmt.Printf("ThreadPool offset: %d\n", unsafe.Offsetof(opts.ThreadPool))
    
  3. Dump raw bytes:

    const bytes: [*]const u8 = @ptrCast(options);
    for (0..16) |i| {
        std.debug.print("{x:0>2} ", .{bytes[i]});
    }
    
  4. Compare layouts: If offsets don’t match between Go and Zig, you need to reorder Go fields.

Best Practices

1. Order by Alignment

When creating Go structs that map to Zig structs:

  1. List all fields with their sizes and alignments
  2. Sort by alignment (descending): 8-byte pointers first, then 4-byte ints, etc.
  3. Add explicit padding to match the total struct size

Example:

type MyStruct struct {
    // 8-byte aligned fields first
    Pointer1 *SomeType
    Pointer2 *AnotherType

    // 4-byte aligned fields
    Count  uint32
    Flags  uint32

    // 2-byte aligned fields
    ShortVal uint16

    // Explicit padding to match Zig struct size
    _ [6]byte
}

2. Document the Layout

Always add comments explaining:

  • The Zig struct being mirrored
  • Why fields are in a particular order
  • The total size and padding
// LoopOptions matches xev.Options in libxev.
// IMPORTANT: Zig reorders struct fields by alignment!
// Actual memory layout:
//   thread_pool: ?*ThreadPool (8 bytes) at offset 0
//   entries: u32 (4 bytes) at offset 8
//   (4 bytes padding to 16 bytes total)
type LoopOptions struct {
    ThreadPool *ThreadPool
    Entries    uint32
    _          uint32
}

3. Verify with Tests

Create size verification tests:

func TestLayoutSizes(t *testing.T) {
    // Call Zig function that returns struct size
    zigSize := cxev.GetOptionsSize()
    goSize := unsafe.Sizeof(cxev.LoopOptions{})

    if zigSize != goSize {
        t.Errorf("size mismatch: Zig=%d Go=%d", zigSize, goSize)
    }
}

4. Export Size/Offset Functions from Zig

Add debug exports to verify layouts:

export fn xev_options_sizeof() usize {
    return @sizeOf(xev.Options);
}

export fn xev_options_field_offsets(entries_offset: *usize, tp_offset: *usize) void {
    entries_offset.* = @offsetOf(xev.Options, "entries");
    tp_offset.* = @offsetOf(xev.Options, "thread_pool");
}

Common Pitfalls

  1. Assuming declaration order equals memory order - Zig reorders by alignment
  2. Forgetting to add padding - Structs may have trailing padding
  3. Not checking on both 32-bit and 64-bit - Pointer sizes differ
  4. Ignoring warnings - If the code “works sometimes,” there’s likely a layout bug

Tools

  • @offsetOf in Zig - Get field offset at compile time
  • unsafe.Offsetof in Go - Get field offset
  • @sizeOf / unsafe.Sizeof - Get struct sizes
  • Debug prints with raw byte dumps - Visualize actual memory layout

For the specific issue that led to this documentation:

  • Issue: SIGSEGV at addr=0x100 during File operations with thread pool
  • Root cause: Go’s LoopOptions fields were in declaration order, but Zig’s Options had fields reordered by alignment
  • Fix: Reordered Go struct fields to match Zig’s alignment-based layout
  • Commit: f0e0ea3 “fix: add padding to LoopOptions for proper memory alignment”