Safety Checker

Safety Checker

This skill helps identify unsafe operations and potential undefined behavior in Zig code, ensuring memory safety and correctness.

Zig Safety Principles

Zig provides safety by default in Debug and ReleaseSafe builds:

• Bounds checking on arrays and slices
• Integer overflow detection
• Null pointer checks for optionals
• Alignment verification

However, unsafe operations can bypass these checks and cause undefined behavior.

Unsafe Operations to Watch For

1. Pointer Casts Without Validation

Type Punning with @ptrCast

// DANGEROUS: No validation of pointer validity
pub fn setHandler(self: *Server, handler: *anyopaque) void {
    self.handler = handler;
}

pub fn getHandler(self: *Server) *Handler {
    const handler_ptr = self.handler.?;
    return @ptrCast(@alignCast(handler_ptr));  // Unsafe cast!
}

Problems:

• No type safety: *anyopaque can point to anything
• No alignment verification before cast
• No null check (using .? panics)
• Could point to wrong type

Better approach:

// SAFE: Use proper optional type
pub const Server = struct {
    handler: ?*Handler,  // Type-safe optional pointer

    pub fn setHandler(self: *Server, handler: *Handler) void {
        self.handler = handler;  // Type checked
    }

    pub fn getHandler(self: *Server) ?*Handler {
        return self.handler;  // No cast needed
    }
};

If type erasure is truly needed:

// SAFER: Validate before casting
pub fn getHandler(self: *Server) ServerError!*Handler {
    const handler_ptr = self.handler orelse return error.NoHandler;

    // Validate alignment
    const addr = @intFromPtr(handler_ptr);
    if (addr % @alignOf(Handler) != 0) {
        return error.InvalidAlignment;
    }

    // Cast with explicit alignment
    const aligned_ptr: *align(@alignOf(Handler)) anyopaque = @alignCast(handler_ptr);
    return @ptrCast(aligned_ptr);
}

2. Missing Bounds Checks

Array/Slice Access

// DANGEROUS: No bounds check
pub fn readIntBig(buffer: []const u8, comptime T: type) T {
    const size = @sizeOf(T);
    if (buffer.len < size) @panic("Buffer too small");  // Panic!

    const bytes = buffer[0..size];
    return std.mem.readInt(T, bytes[0..size], .big);
}

Problems:

• Panic on invalid input (should return error)
• Potential crash if panic is reached in release mode
• No way for caller to handle error

Better approach:

// SAFE: Return error, not panic
pub fn readIntBig(buffer: []const u8, comptime T: type) ParseError!T {
    const size = @sizeOf(T);
    if (buffer.len < size) return error.BufferTooSmall;

    const bytes = buffer[0..size];
    return std.mem.readInt(T, bytes[0..size], .big);
}

Unchecked Indexing

// DANGEROUS: Assumes index is valid
pub fn getRegister(self: *Server, index: u8) Value {
    return self.registers[index];  // Could be out of bounds!
}

// SAFE: Check bounds explicitly
pub fn getRegister(self: *Server, index: u8) ServerError!Value {
    if (index >= self.registers.len) return error.RegisterOutOfBounds;
    return self.registers[index];
}

// ALSO SAFE: Use optional for invalid indices
pub fn getRegister(self: *Server, index: u8) ?Value {
    if (index >= self.registers.len) return null;
    return self.registers[index];
}

3. Unsafe Pointer Arithmetic

// DANGEROUS: Manual pointer arithmetic
pub fn advance(ptr: [*]const u8, offset: usize) [*]const u8 {
    return ptr + offset;  // No bounds check!
}

// SAFE: Use slices with bounds
pub fn advance(slice: []const u8, offset: usize) ?[]const u8 {
    if (offset >= slice.len) return null;
    return slice[offset..];
}

4. Uninitialized Memory

// DANGEROUS: Reading uninitialized memory
pub fn createValue() Value {
    var value: Value = undefined;  // Uninitialized!
    if (condition) {
        value = Value.makeInt(42);
    }
    return value;  // May return uninitialized data!
}

// SAFE: Initialize before use
pub fn createValue() Value {
    if (condition) {
        return Value.makeInt(42);
    }
    return Value.nil();  // Always initialized
}

5. Null Pointer Dereference

// DANGEROUS: Force-unwrapping optionals
pub fn process(self: *Server) void {
    const handler = self.handler.?;  // Panics if null!
    handler.handle(self);
}

// SAFE: Check for null
pub fn process(self: *Server) ServerError!void {
    const handler = self.handler orelse return error.NoHandler;
    try handler.handle(self);
}

// ALSO SAFE: Use if for optional handling
pub fn process(self: *Server) void {
    if (self.handler) |handler| {
        handler.handle(self) catch |err| {
            std.log.err("Handle failed: {}", .{err});
        };
    }
}

6. Integer Overflow

// DANGEROUS: Unchecked arithmetic in ReleaseFast
pub fn allocate(size: usize, count: usize) ![]u8 {
    const total = size * count;  // Can overflow!
    return try allocator.alloc(u8, total);
}

// SAFE: Use checked arithmetic
pub fn allocate(size: usize, count: usize) ![]u8 {
    const total = std.math.mul(usize, size, count) catch {
        return error.ArithmeticOverflow;
    };
    return try allocator.alloc(u8, total);
}

// ALSO SAFE: Wrapping arithmetic when overflow is intended
pub fn hash(value: u32) u32 {
    return value *% 31;  // *% = wrapping multiplication
}

7. Alignment Issues

// DANGEROUS: Assuming alignment
pub fn readU32(bytes: []const u8) u32 {
    const ptr: *const u32 = @ptrCast(bytes.ptr);  // May be misaligned!
    return ptr.*;
}

// SAFE: Use std.mem functions
pub fn readU32(bytes: []const u8) !u32 {
    if (bytes.len < 4) return error.BufferTooSmall;
    return std.mem.readInt(u32, bytes[0..4], .little);
}

// SAFE: Check alignment explicitly
pub fn readU32(bytes: []align(4) const u8) u32 {
    const ptr: *const u32 = @ptrCast(bytes.ptr);  // Compiler ensures alignment
    return ptr.*;
}

8. Unsafe Memory Operations

// DANGEROUS: @memcpy with wrong size
pub fn copy(dest: []u8, src: []const u8) void {
    @memcpy(dest, src);  // Panics if dest.len < src.len
}

// SAFE: Check sizes first
pub fn copy(dest: []u8, src: []const u8) !void {
    if (dest.len < src.len) return error.DestinationTooSmall;
    @memcpy(dest[0..src.len], src);
}

Safety Checklist

When Writing Low-Level Code:

• No @ptrCast without validation
• Check alignment before pointer casts
• Validate buffer sizes before access
• Use errors instead of panics for invalid input
• Initialize all variables before use
• Check for null before unwrapping optionals
• Use checked arithmetic for user-controlled values
• Verify alignment assumptions
• Validate sizes for @memcpy, @memset

When Reviewing Code:

• Look for @ptrCast - is it necessary? Is it safe?
• Look for @alignCast - is alignment guaranteed?
• Check all array/slice indexing - bounds checked?
• Look for .? unwrapping - can it be null?
• Check for undefined - is it initialized on all paths?
• Look for panics - should they be errors instead?
• Check arithmetic - can it overflow?
• Verify @memcpy sizes match

Red Flags

Immediate Review Needed:

1. @ptrCast without validation - Almost always wrong
2. @panic for user input - Should be errors
3. .? on optional without null check - May crash
4. Array access without bounds check - May crash
5. undefined without initialization on all paths - UB
6. Unchecked arithmetic on user input - May overflow

Consider Refactoring:

1. Type erasure with *anyopaque - Prefer proper types
2. Manual pointer arithmetic - Use slices instead
3. Assuming alignment - Make it explicit
4. Complex pointer operations - Simplify if possible

Testing Safety

Test edge cases and error conditions:

test "register access out of bounds" {
    var server = try Server.init(std.testing.allocator);
    defer server.deinit();

    // Should return error, not crash
    try std.testing.expectError(
        ServerError.RegisterOutOfBounds,
        server.getRegister(255)
    );
}

test "stack overflow protection" {
    var server = try Server.init(std.testing.allocator);
    defer server.deinit();

    // Fill stack
    while (server.sp < server.stack.len) {
        try server.push(Value.makeInt(0));
    }

    // Next push should error, not crash
    try std.testing.expectError(
        ServerError.StackOverflow,
        server.push(Value.makeInt(1))
    );
}

test "buffer too small handling" {
    const data = [_]u8{ 1, 2 };  // Only 2 bytes

    // Should error, not crash
    try std.testing.expectError(
        ParseError.BufferTooSmall,
        readIntBig(&data, u32)  // Needs 4 bytes
    );
}

Build Modes and Safety

• Debug: All safety checks enabled (bounds, overflow, etc.)
• ReleaseSafe: Safety checks enabled, optimizations on
• ReleaseFast: Safety checks disabled for performance
• ReleaseSmall: Like ReleaseFast but optimizes for size

Golden Rule: Code should be correct in ReleaseFast even though safety checks are disabled. Never rely on runtime checks for correctness in release builds.

// BAD: Relies on Debug mode checks
pub fn access(slice: []u8, index: usize) u8 {
    return slice[index];  // Crashes in ReleaseFast if out of bounds!
}

// GOOD: Explicit check in all modes
pub fn access(slice: []u8, index: usize) ?u8 {
    if (index >= slice.len) return null;
    return slice[index];
}

Summary

Safety in Zig is about:

1. Validation - Check inputs and bounds explicitly
2. Error handling - Return errors, don't panic
3. Type safety - Avoid type erasure when possible
4. Explicit intent - Make assumptions clear in types
5. Testing - Test error paths and edge cases

When in doubt, prefer safety over performance. Optimize only when profiling proves it necessary.