Merge remote-tracking branch 'origin/master' into llvm12

7 files changed, 4501 insertions, 696 deletions

diff --git a/src/link/MachO.zig b/src/link/MachO.zig
index fe8fcd6281..517fabaf3e 100644
--- a/src/link/MachO.zig
+++ b/src/link/MachO.zig
@@ -11,7 +11,9 @@ const codegen = @import("../codegen.zig");
 const aarch64 = @import("../codegen/aarch64.zig");
 const math = std.math;
 const mem = std.mem;
+const meta = std.meta;
 
+const bind = @import("MachO/bind.zig");
 const trace = @import("../tracy.zig").trace;
 const build_options = @import("build_options");
 const Module = @import("../Module.zig");
@@ -24,9 +26,9 @@ const target_util = @import("../target.zig");
 const DebugSymbols = @import("MachO/DebugSymbols.zig");
 const Trie = @import("MachO/Trie.zig");
 const CodeSignature = @import("MachO/CodeSignature.zig");
+const Zld = @import("MachO/Zld.zig");
 
 usingnamespace @import("MachO/commands.zig");
-usingnamespace @import("MachO/imports.zig");
 
 pub const base_tag: File.Tag = File.Tag.macho;
 
@@ -87,14 +89,12 @@ code_signature_cmd_index: ?u16 = null,
 
 /// Index into __TEXT,__text section.
 text_section_index: ?u16 = null,
-/// Index into __TEXT,__ziggot section.
-got_section_index: ?u16 = null,
 /// Index into __TEXT,__stubs section.
 stubs_section_index: ?u16 = null,
 /// Index into __TEXT,__stub_helper section.
 stub_helper_section_index: ?u16 = null,
 /// Index into __DATA_CONST,__got section.
-data_got_section_index: ?u16 = null,
+got_section_index: ?u16 = null,
 /// Index into __DATA,__la_symbol_ptr section.
 la_symbol_ptr_section_index: ?u16 = null,
 /// Index into __DATA,__data section.
@@ -104,16 +104,16 @@ entry_addr: ?u64 = null,
 
 /// Table of all local symbols
 /// Internally references string table for names (which are optional).
-local_symbols: std.ArrayListUnmanaged(macho.nlist_64) = .{},
+locals: std.ArrayListUnmanaged(macho.nlist_64) = .{},
 /// Table of all global symbols
-global_symbols: std.ArrayListUnmanaged(macho.nlist_64) = .{},
+globals: std.ArrayListUnmanaged(macho.nlist_64) = .{},
 /// Table of all extern nonlazy symbols, indexed by name.
-extern_nonlazy_symbols: std.StringArrayHashMapUnmanaged(ExternSymbol) = .{},
+nonlazy_imports: std.StringArrayHashMapUnmanaged(Import) = .{},
 /// Table of all extern lazy symbols, indexed by name.
-extern_lazy_symbols: std.StringArrayHashMapUnmanaged(ExternSymbol) = .{},
+lazy_imports: std.StringArrayHashMapUnmanaged(Import) = .{},
 
-local_symbol_free_list: std.ArrayListUnmanaged(u32) = .{},
-global_symbol_free_list: std.ArrayListUnmanaged(u32) = .{},
+locals_free_list: std.ArrayListUnmanaged(u32) = .{},
+globals_free_list: std.ArrayListUnmanaged(u32) = .{},
 offset_table_free_list: std.ArrayListUnmanaged(u32) = .{},
 
 stub_helper_stubs_start_off: ?u64 = null,
@@ -122,8 +122,8 @@ stub_helper_stubs_start_off: ?u64 = null,
 string_table: std.ArrayListUnmanaged(u8) = .{},
 string_table_directory: std.StringHashMapUnmanaged(u32) = .{},
 
-/// Table of trampolines to the actual symbols in __text section.
-offset_table: std.ArrayListUnmanaged(u64) = .{},
+/// Table of GOT entries.
+offset_table: std.ArrayListUnmanaged(GOTEntry) = .{},
 
 error_flags: File.ErrorFlags = File.ErrorFlags{},
 
@@ -154,14 +154,19 @@ string_table_needs_relocation: bool = false,
 /// allocate a fresh text block, which will have ideal capacity, and then grow it
 /// by 1 byte. It will then have -1 overcapacity.
 text_block_free_list: std.ArrayListUnmanaged(*TextBlock) = .{},
+
 /// Pointer to the last allocated text block
 last_text_block: ?*TextBlock = null,
+
 /// A list of all PIE fixups required for this run of the linker.
 /// Warning, this is currently NOT thread-safe. See the TODO below.
 /// TODO Move this list inside `updateDecl` where it should be allocated
 /// prior to calling `generateSymbol`, and then immediately deallocated
 /// rather than sitting in the global scope.
-pie_fixups: std.ArrayListUnmanaged(PieFixup) = .{},
+/// TODO We should also rewrite this using generic relocations common to all
+/// backends.
+pie_fixups: std.ArrayListUnmanaged(PIEFixup) = .{},
+
 /// A list of all stub (extern decls) fixups required for this run of the linker.
 /// Warning, this is currently NOT thread-safe. See the TODO below.
 /// TODO Move this list inside `updateDecl` where it should be allocated
@@ -169,14 +174,42 @@ pie_fixups: std.ArrayListUnmanaged(PieFixup) = .{},
 /// rather than sitting in the global scope.
 stub_fixups: std.ArrayListUnmanaged(StubFixup) = .{},
 
-pub const PieFixup = struct {
-    /// Target address we wanted to address in absolute terms.
-    address: u64,
-    /// Where in the byte stream we should perform the fixup.
-    start: usize,
-    /// The length of the byte stream. For x86_64, this will be
-    /// variable. For aarch64, it will be fixed at 4 bytes.
-    len: usize,
+pub const GOTEntry = struct {
+    /// GOT entry can either be a local pointer or an extern (nonlazy) import.
+    kind: enum {
+        Local,
+        Extern,
+    },
+
+    /// Id to the macho.nlist_64 from the respective table: either locals or nonlazy imports.
+    /// TODO I'm more and more inclined to just manage a single, max two symbol tables
+    ///  rather than 4 as we currently do, but I'll follow up in the future PR.
+    symbol: u32,
+
+    /// Index of this entry in the GOT.
+    index: u32,
+};
+
+pub const Import = struct {
+    /// MachO symbol table entry.
+    symbol: macho.nlist_64,
+
+    /// Id of the dynamic library where the specified entries can be found.
+    dylib_ordinal: i64,
+
+    /// Index of this import within the import list.
+    index: u32,
+};
+
+pub const PIEFixup = struct {
+    /// Target VM address of this relocation.
+    target_addr: u64,
+
+    /// Offset within the byte stream.
+    offset: usize,
+
+    /// Size of the relocation.
+    size: usize,
 };
 
 pub const StubFixup = struct {
@@ -260,9 +293,9 @@ pub const TextBlock = struct {
     /// File offset relocation happens transparently, so it is not included in
     /// this calculation.
     fn capacity(self: TextBlock, macho_file: MachO) u64 {
-        const self_sym = macho_file.local_symbols.items[self.local_sym_index];
+        const self_sym = macho_file.locals.items[self.local_sym_index];
         if (self.next) |next| {
-            const next_sym = macho_file.local_symbols.items[next.local_sym_index];
+            const next_sym = macho_file.locals.items[next.local_sym_index];
             return next_sym.n_value - self_sym.n_value;
         } else {
             // We are the last block.
@@ -274,8 +307,8 @@ pub const TextBlock = struct {
     fn freeListEligible(self: TextBlock, macho_file: MachO) bool {
         // No need to keep a free list node for the last block.
         const next = self.next orelse return false;
-        const self_sym = macho_file.local_symbols.items[self.local_sym_index];
-        const next_sym = macho_file.local_symbols.items[next.local_sym_index];
+        const self_sym = macho_file.locals.items[self.local_sym_index];
+        const next_sym = macho_file.locals.items[next.local_sym_index];
         const cap = next_sym.n_value - self_sym.n_value;
         const ideal_cap = padToIdeal(self.size);
         if (cap <= ideal_cap) return false;
@@ -344,7 +377,7 @@ pub fn openPath(allocator: *Allocator, sub_path: []const u8, options: link.Optio
     };
 
     // Index 0 is always a null symbol.
-    try self.local_symbols.append(allocator, .{
+    try self.locals.append(allocator, .{
         .n_strx = 0,
         .n_type = 0,
         .n_sect = 0,
@@ -600,7 +633,74 @@ fn linkWithLLD(self: *MachO, comp: *Compilation) !void {
         if (!mem.eql(u8, the_object_path, full_out_path)) {
             try fs.cwd().copyFile(the_object_path, fs.cwd(), full_out_path, .{});
         }
-    } else {
+    } else outer: {
+        const use_zld = blk: {
+            if (self.base.options.is_native_os and self.base.options.system_linker_hack) {
+                // If the user forces the use of ld64, make sure we are running native!
+                break :blk false;
+            }
+
+            if (self.base.options.target.cpu.arch == .aarch64) {
+                // On aarch64, always use zld.
+                break :blk true;
+            }
+
+            if (self.base.options.link_libcpp or
+                self.base.options.output_mode == .Lib or
+                self.base.options.linker_script != null)
+            {
+                // Fallback to LLD in this handful of cases on x86_64 only.
+                break :blk false;
+            }
+
+            break :blk true;
+        };
+
+        if (use_zld) {
+            var zld = Zld.init(self.base.allocator);
+            defer zld.deinit();
+            zld.arch = target.cpu.arch;
+
+            var input_files = std.ArrayList([]const u8).init(self.base.allocator);
+            defer input_files.deinit();
+            // Positional arguments to the linker such as object files.
+            try input_files.appendSlice(self.base.options.objects);
+            for (comp.c_object_table.items()) |entry| {
+                try input_files.append(entry.key.status.success.object_path);
+            }
+            if (module_obj_path) |p| {
+                try input_files.append(p);
+            }
+            try input_files.append(comp.compiler_rt_static_lib.?.full_object_path);
+            // libc++ dep
+            if (self.base.options.link_libcpp) {
+                try input_files.append(comp.libcxxabi_static_lib.?.full_object_path);
+                try input_files.append(comp.libcxx_static_lib.?.full_object_path);
+            }
+
+            if (self.base.options.verbose_link) {
+                var argv = std.ArrayList([]const u8).init(self.base.allocator);
+                defer argv.deinit();
+
+                try argv.append("zig");
+                try argv.append("ld");
+
+                try argv.ensureCapacity(input_files.items.len);
+                for (input_files.items) |f| {
+                    argv.appendAssumeCapacity(f);
+                }
+
+                try argv.append("-o");
+                try argv.append(full_out_path);
+
+                Compilation.dump_argv(argv.items);
+            }
+
+            try zld.link(input_files.items, full_out_path);
+
+            break :outer;
+        }
+
         // Create an LLD command line and invoke it.
         var argv = std.ArrayList([]const u8).init(self.base.allocator);
         defer argv.deinit();
@@ -644,9 +744,7 @@ fn linkWithLLD(self: *MachO, comp: *Compilation) !void {
             try argv.append("defs");
         }
 
-        if (is_dyn_lib) {
-            try argv.append("-static");
-        } else {
+        if (is_exe_or_dyn_lib) {
             try argv.append("-dynamic");
         }
 
@@ -836,7 +934,7 @@ fn linkWithLLD(self: *MachO, comp: *Compilation) !void {
                         }
                     },
                     else => {
-                        log.err("{s} terminated", .{ argv.items[0] });
+                        log.err("{s} terminated", .{argv.items[0]});
                         return error.LLDCrashed;
                     },
                 }
@@ -873,119 +971,6 @@ fn linkWithLLD(self: *MachO, comp: *Compilation) !void {
                     log.warn("unexpected LLD stderr:\n{s}", .{stderr});
                 }
             }
-
-            // At this stage, LLD has done its job. It is time to patch the resultant
-            // binaries up!
-            const out_file = try directory.handle.openFile(self.base.options.emit.?.sub_path, .{ .write = true });
-            try self.parseFromFile(out_file);
-
-            if (self.libsystem_cmd_index == null and self.header.?.filetype == macho.MH_EXECUTE) {
-                const text_segment = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
-                const text_section = text_segment.sections.items[self.text_section_index.?];
-                const after_last_cmd_offset = self.header.?.sizeofcmds + @sizeOf(macho.mach_header_64);
-                const needed_size = padToIdeal(@sizeOf(macho.linkedit_data_command));
-
-                if (needed_size + after_last_cmd_offset > text_section.offset) {
-                    log.err("Unable to extend padding between the end of load commands and start of __text section.", .{});
-                    log.err("Re-run the linker with '-headerpad 0x{x}' option if available, or", .{needed_size});
-                    log.err("fall back to the system linker by exporting 'ZIG_SYSTEM_LINKER_HACK=1'.", .{});
-                    return error.NotEnoughPadding;
-                }
-
-                // Calculate next available dylib ordinal.
-                const next_ordinal = blk: {
-                    var ordinal: u32 = 1;
-                    for (self.load_commands.items) |cmd| {
-                        switch (cmd) {
-                            .Dylib => ordinal += 1,
-                            else => {},
-                        }
-                    }
-                    break :blk ordinal;
-                };
-
-                // Add load dylib load command
-                self.libsystem_cmd_index = @intCast(u16, self.load_commands.items.len);
-                const cmdsize = @intCast(u32, mem.alignForwardGeneric(
-                    u64,
-                    @sizeOf(macho.dylib_command) + mem.lenZ(LIB_SYSTEM_PATH),
-                    @sizeOf(u64),
-                ));
-                // TODO Find a way to work out runtime version from the OS version triple stored in std.Target.
-                // In the meantime, we're gonna hardcode to the minimum compatibility version of 0.0.0.
-                const min_version = 0x0;
-                var dylib_cmd = emptyGenericCommandWithData(macho.dylib_command{
-                    .cmd = macho.LC_LOAD_DYLIB,
-                    .cmdsize = cmdsize,
-                    .dylib = .{
-                        .name = @sizeOf(macho.dylib_command),
-                        .timestamp = 2, // not sure why not simply 0; this is reverse engineered from Mach-O files
-                        .current_version = min_version,
-                        .compatibility_version = min_version,
-                    },
-                });
-                dylib_cmd.data = try self.base.allocator.alloc(u8, cmdsize - dylib_cmd.inner.dylib.name);
-                mem.set(u8, dylib_cmd.data, 0);
-                mem.copy(u8, dylib_cmd.data, mem.spanZ(LIB_SYSTEM_PATH));
-                try self.load_commands.append(self.base.allocator, .{ .Dylib = dylib_cmd });
-                self.header_dirty = true;
-                self.load_commands_dirty = true;
-
-                if (self.symtab_cmd_index == null or self.dysymtab_cmd_index == null) {
-                    log.err("Incomplete Mach-O binary: no LC_SYMTAB or LC_DYSYMTAB load command found!", .{});
-                    log.err("Without the symbol table, it is not possible to patch up the binary for cross-compilation.", .{});
-                    return error.NoSymbolTableFound;
-                }
-
-                // Patch dyld info
-                try self.fixupBindInfo(next_ordinal);
-                try self.fixupLazyBindInfo(next_ordinal);
-
-                // Write updated load commands and the header
-                try self.writeLoadCommands();
-                try self.writeHeader();
-
-                assert(!self.header_dirty);
-                assert(!self.load_commands_dirty);
-            }
-            if (self.code_signature_cmd_index == null) outer: {
-                if (target.cpu.arch != .aarch64) break :outer; // This is currently needed only for aarch64 targets.
-                const text_segment = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
-                const text_section = text_segment.sections.items[self.text_section_index.?];
-                const after_last_cmd_offset = self.header.?.sizeofcmds + @sizeOf(macho.mach_header_64);
-                const needed_size = padToIdeal(@sizeOf(macho.linkedit_data_command));
-
-                if (needed_size + after_last_cmd_offset > text_section.offset) {
-                    log.err("Unable to extend padding between the end of load commands and start of __text section.", .{});
-                    log.err("Re-run the linker with '-headerpad 0x{x}' option if available, or", .{needed_size});
-                    log.err("fall back to the system linker by exporting 'ZIG_SYSTEM_LINKER_HACK=1'.", .{});
-                    return error.NotEnoughPadding;
-                }
-
-                // Add code signature load command
-                self.code_signature_cmd_index = @intCast(u16, self.load_commands.items.len);
-                try self.load_commands.append(self.base.allocator, .{
-                    .LinkeditData = .{
-                        .cmd = macho.LC_CODE_SIGNATURE,
-                        .cmdsize = @sizeOf(macho.linkedit_data_command),
-                        .dataoff = 0,
-                        .datasize = 0,
-                    },
-                });
-                self.header_dirty = true;
-                self.load_commands_dirty = true;
-
-                // Pad out space for code signature
-                try self.writeCodeSignaturePadding();
-                // Write updated load commands and the header
-                try self.writeLoadCommands();
-                try self.writeHeader();
-                // Generate adhoc code signature
-                try self.writeCodeSignature();
-
-                assert(!self.header_dirty);
-                assert(!self.load_commands_dirty);
-            }
         }
     }
 
@@ -1021,14 +1006,14 @@ pub fn deinit(self: *MachO) void {
     if (self.d_sym) |*ds| {
         ds.deinit(self.base.allocator);
     }
-    for (self.extern_lazy_symbols.items()) |*entry| {
+    for (self.lazy_imports.items()) |*entry| {
         self.base.allocator.free(entry.key);
     }
-    self.extern_lazy_symbols.deinit(self.base.allocator);
-    for (self.extern_nonlazy_symbols.items()) |*entry| {
+    self.lazy_imports.deinit(self.base.allocator);
+    for (self.nonlazy_imports.items()) |*entry| {
         self.base.allocator.free(entry.key);
     }
-    self.extern_nonlazy_symbols.deinit(self.base.allocator);
+    self.nonlazy_imports.deinit(self.base.allocator);
     self.pie_fixups.deinit(self.base.allocator);
     self.stub_fixups.deinit(self.base.allocator);
     self.text_block_free_list.deinit(self.base.allocator);
@@ -1042,10 +1027,10 @@ pub fn deinit(self: *MachO) void {
     }
     self.string_table_directory.deinit(self.base.allocator);
     self.string_table.deinit(self.base.allocator);
-    self.global_symbols.deinit(self.base.allocator);
-    self.global_symbol_free_list.deinit(self.base.allocator);
-    self.local_symbols.deinit(self.base.allocator);
-    self.local_symbol_free_list.deinit(self.base.allocator);
+    self.globals.deinit(self.base.allocator);
+    self.globals_free_list.deinit(self.base.allocator);
+    self.locals.deinit(self.base.allocator);
+    self.locals_free_list.deinit(self.base.allocator);
     for (self.load_commands.items) |*lc| {
         lc.deinit(self.base.allocator);
     }
@@ -1100,7 +1085,7 @@ fn shrinkTextBlock(self: *MachO, text_block: *TextBlock, new_block_size: u64) vo
 }
 
 fn growTextBlock(self: *MachO, text_block: *TextBlock, new_block_size: u64, alignment: u64) !u64 {
-    const sym = self.local_symbols.items[text_block.local_sym_index];
+    const sym = self.locals.items[text_block.local_sym_index];
     const align_ok = mem.alignBackwardGeneric(u64, sym.n_value, alignment) == sym.n_value;
     const need_realloc = !align_ok or new_block_size > text_block.capacity(self.*);
     if (!need_realloc) return sym.n_value;
@@ -1110,34 +1095,41 @@ fn growTextBlock(self: *MachO, text_block: *TextBlock, new_block_size: u64, alig
 pub fn allocateDeclIndexes(self: *MachO, decl: *Module.Decl) !void {
     if (decl.link.macho.local_sym_index != 0) return;
 
-    try self.local_symbols.ensureCapacity(self.base.allocator, self.local_symbols.items.len + 1);
+    try self.locals.ensureCapacity(self.base.allocator, self.locals.items.len + 1);
     try self.offset_table.ensureCapacity(self.base.allocator, self.offset_table.items.len + 1);
 
-    if (self.local_symbol_free_list.popOrNull()) |i| {
+    if (self.locals_free_list.popOrNull()) |i| {
         log.debug("reusing symbol index {d} for {s}", .{ i, decl.name });
         decl.link.macho.local_sym_index = i;
     } else {
-        log.debug("allocating symbol index {d} for {s}", .{ self.local_symbols.items.len, decl.name });
-        decl.link.macho.local_sym_index = @intCast(u32, self.local_symbols.items.len);
-        _ = self.local_symbols.addOneAssumeCapacity();
+        log.debug("allocating symbol index {d} for {s}", .{ self.locals.items.len, decl.name });
+        decl.link.macho.local_sym_index = @intCast(u32, self.locals.items.len);
+        _ = self.locals.addOneAssumeCapacity();
     }
 
     if (self.offset_table_free_list.popOrNull()) |i| {
+        log.debug("reusing offset table entry index {d} for {s}", .{ i, decl.name });
         decl.link.macho.offset_table_index = i;
     } else {
+        log.debug("allocating offset table entry index {d} for {s}", .{ self.offset_table.items.len, decl.name });
         decl.link.macho.offset_table_index = @intCast(u32, self.offset_table.items.len);
         _ = self.offset_table.addOneAssumeCapacity();
         self.offset_table_count_dirty = true;
+        self.rebase_info_dirty = true;
     }
 
-    self.local_symbols.items[decl.link.macho.local_sym_index] = .{
+    self.locals.items[decl.link.macho.local_sym_index] = .{
         .n_strx = 0,
         .n_type = 0,
         .n_sect = 0,
         .n_desc = 0,
         .n_value = 0,
     };
-    self.offset_table.items[decl.link.macho.offset_table_index] = 0;
+    self.offset_table.items[decl.link.macho.offset_table_index] = .{
+        .kind = .Local,
+        .symbol = decl.link.macho.local_sym_index,
+        .index = decl.link.macho.offset_table_index,
+    };
 }
 
 pub fn updateDecl(self: *MachO, module: *Module, decl: *Module.Decl) !void {
@@ -1180,8 +1172,9 @@ pub fn updateDecl(self: *MachO, module: *Module, decl: *Module.Decl) !void {
         .externally_managed => |x| x,
         .appended => code_buffer.items,
         .fail => |em| {
-            // Clear any PIE fixups and stub fixups for this decl.
+            // Clear any PIE fixups for this decl.
             self.pie_fixups.shrinkRetainingCapacity(0);
+            // Clear any stub fixups for this decl.
             self.stub_fixups.shrinkRetainingCapacity(0);
             decl.analysis = .codegen_failure;
             try module.failed_decls.put(module.gpa, decl, em);
@@ -1191,7 +1184,7 @@ pub fn updateDecl(self: *MachO, module: *Module, decl: *Module.Decl) !void {
 
     const required_alignment = typed_value.ty.abiAlignment(self.base.options.target);
     assert(decl.link.macho.local_sym_index != 0); // Caller forgot to call allocateDeclIndexes()
-    const symbol = &self.local_symbols.items[decl.link.macho.local_sym_index];
+    const symbol = &self.locals.items[decl.link.macho.local_sym_index];
 
     if (decl.link.macho.size != 0) {
         const capacity = decl.link.macho.capacity(self.*);
@@ -1200,9 +1193,12 @@ pub fn updateDecl(self: *MachO, module: *Module, decl: *Module.Decl) !void {
             const vaddr = try self.growTextBlock(&decl.link.macho, code.len, required_alignment);
             log.debug("growing {s} from 0x{x} to 0x{x}", .{ decl.name, symbol.n_value, vaddr });
             if (vaddr != symbol.n_value) {
-                symbol.n_value = vaddr;
                 log.debug(" (writing new offset table entry)", .{});
-                self.offset_table.items[decl.link.macho.offset_table_index] = vaddr;
+                self.offset_table.items[decl.link.macho.offset_table_index] = .{
+                    .kind = .Local,
+                    .symbol = decl.link.macho.local_sym_index,
+                    .index = decl.link.macho.offset_table_index,
+                };
                 try self.writeOffsetTableEntry(decl.link.macho.offset_table_index);
             }
         } else if (code.len < decl.link.macho.size) {
@@ -1231,7 +1227,11 @@ pub fn updateDecl(self: *MachO, module: *Module, decl: *Module.Decl) !void {
             .n_desc = 0,
             .n_value = addr,
         };
-        self.offset_table.items[decl.link.macho.offset_table_index] = addr;
+        self.offset_table.items[decl.link.macho.offset_table_index] = .{
+            .kind = .Local,
+            .symbol = decl.link.macho.local_sym_index,
+            .index = decl.link.macho.offset_table_index,
+        };
 
         try self.writeLocalSymbol(decl.link.macho.local_sym_index);
         if (self.d_sym) |*ds|
@@ -1239,30 +1239,48 @@ pub fn updateDecl(self: *MachO, module: *Module, decl: *Module.Decl) !void {
         try self.writeOffsetTableEntry(decl.link.macho.offset_table_index);
     }
 
-    // Perform PIE fixups (if any)
-    const text_segment = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
-    const got_section = text_segment.sections.items[self.got_section_index.?];
+    // Calculate displacements to target addr (if any).
     while (self.pie_fixups.popOrNull()) |fixup| {
-        const target_addr = fixup.address;
-        const this_addr = symbol.n_value + fixup.start;
+        assert(fixup.size == 4);
+        const this_addr = symbol.n_value + fixup.offset;
+        const target_addr = fixup.target_addr;
+
         switch (self.base.options.target.cpu.arch) {
             .x86_64 => {
-                assert(target_addr >= this_addr + fixup.len);
-                const displacement = try math.cast(u32, target_addr - this_addr - fixup.len);
-                var placeholder = code_buffer.items[fixup.start + fixup.len - @sizeOf(u32) ..][0..@sizeOf(u32)];
-                mem.writeIntSliceLittle(u32, placeholder, displacement);
+                const displacement = try math.cast(u32, target_addr - this_addr - 4);
+                mem.writeIntLittle(u32, code_buffer.items[fixup.offset..][0..4], displacement);
             },
             .aarch64 => {
-                assert(target_addr >= this_addr);
-                const displacement = try math.cast(u27, target_addr - this_addr);
-                var placeholder = code_buffer.items[fixup.start..][0..fixup.len];
-                mem.writeIntSliceLittle(u32, placeholder, aarch64.Instruction.b(@as(i28, displacement)).toU32());
+                // TODO optimize instruction based on jump length (use ldr(literal) + nop if possible).
+                {
+                    const inst = code_buffer.items[fixup.offset..][0..4];
+                    var parsed = mem.bytesAsValue(meta.TagPayload(
+                        aarch64.Instruction,
+                        aarch64.Instruction.PCRelativeAddress,
+                    ), inst);
+                    const this_page = @intCast(i32, this_addr >> 12);
+                    const target_page = @intCast(i32, target_addr >> 12);
+                    const pages = @bitCast(u21, @intCast(i21, target_page - this_page));
+                    parsed.immhi = @truncate(u19, pages >> 2);
+                    parsed.immlo = @truncate(u2, pages);
+                }
+                {
+                    const inst = code_buffer.items[fixup.offset + 4 ..][0..4];
+                    var parsed = mem.bytesAsValue(meta.TagPayload(
+                        aarch64.Instruction,
+                        aarch64.Instruction.LoadStoreRegister,
+                    ), inst);
+                    const narrowed = @truncate(u12, target_addr);
+                    const offset = try math.divExact(u12, narrowed, 8);
+                    parsed.offset = offset;
+                }
             },
             else => unreachable, // unsupported target architecture
         }
     }
 
     // Resolve stubs (if any)
+    const text_segment = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
     const stubs = text_segment.sections.items[self.stubs_section_index.?];
     for (self.stub_fixups.items) |fixup| {
         const stub_addr = stubs.addr + fixup.symbol * stubs.reserved2;
@@ -1287,9 +1305,6 @@ pub fn updateDecl(self: *MachO, module: *Module, decl: *Module.Decl) !void {
             try self.writeStubInStubHelper(fixup.symbol);
             try self.writeLazySymbolPointer(fixup.symbol);
 
-            const extern_sym = &self.extern_lazy_symbols.items()[fixup.symbol].value;
-            extern_sym.segment = self.data_segment_cmd_index.?;
-            extern_sym.offset = fixup.symbol * @sizeOf(u64);
             self.rebase_info_dirty = true;
             self.lazy_binding_info_dirty = true;
         }
@@ -1331,9 +1346,9 @@ pub fn updateDeclExports(
     const tracy = trace(@src());
     defer tracy.end();
 
-    try self.global_symbols.ensureCapacity(self.base.allocator, self.global_symbols.items.len + exports.len);
+    try self.globals.ensureCapacity(self.base.allocator, self.globals.items.len + exports.len);
     if (decl.link.macho.local_sym_index == 0) return;
-    const decl_sym = &self.local_symbols.items[decl.link.macho.local_sym_index];
+    const decl_sym = &self.locals.items[decl.link.macho.local_sym_index];
 
     for (exports) |exp| {
         if (exp.options.section) |section_name| {
@@ -1366,7 +1381,7 @@ pub fn updateDeclExports(
         };
         const n_type = decl_sym.n_type | macho.N_EXT;
         if (exp.link.macho.sym_index) |i| {
-            const sym = &self.global_symbols.items[i];
+            const sym = &self.globals.items[i];
             sym.* = .{
                 .n_strx = try self.updateString(sym.n_strx, exp.options.name),
                 .n_type = n_type,
@@ -1376,12 +1391,12 @@ pub fn updateDeclExports(
             };
         } else {
             const name_str_index = try self.makeString(exp.options.name);
-            const i = if (self.global_symbol_free_list.popOrNull()) |i| i else blk: {
-                _ = self.global_symbols.addOneAssumeCapacity();
+            const i = if (self.globals_free_list.popOrNull()) |i| i else blk: {
+                _ = self.globals.addOneAssumeCapacity();
                 self.export_info_dirty = true;
-                break :blk self.global_symbols.items.len - 1;
+                break :blk self.globals.items.len - 1;
             };
-            self.global_symbols.items[i] = .{
+            self.globals.items[i] = .{
                 .n_strx = name_str_index,
                 .n_type = n_type,
                 .n_sect = @intCast(u8, self.text_section_index.?) + 1,
@@ -1396,18 +1411,18 @@ pub fn updateDeclExports(
 
 pub fn deleteExport(self: *MachO, exp: Export) void {
     const sym_index = exp.sym_index orelse return;
-    self.global_symbol_free_list.append(self.base.allocator, sym_index) catch {};
-    self.global_symbols.items[sym_index].n_type = 0;
+    self.globals_free_list.append(self.base.allocator, sym_index) catch {};
+    self.globals.items[sym_index].n_type = 0;
 }
 
 pub fn freeDecl(self: *MachO, decl: *Module.Decl) void {
     // Appending to free lists is allowed to fail because the free lists are heuristics based anyway.
     self.freeTextBlock(&decl.link.macho);
     if (decl.link.macho.local_sym_index != 0) {
-        self.local_symbol_free_list.append(self.base.allocator, decl.link.macho.local_sym_index) catch {};
+        self.locals_free_list.append(self.base.allocator, decl.link.macho.local_sym_index) catch {};
         self.offset_table_free_list.append(self.base.allocator, decl.link.macho.offset_table_index) catch {};
 
-        self.local_symbols.items[decl.link.macho.local_sym_index].n_type = 0;
+        self.locals.items[decl.link.macho.local_sym_index].n_type = 0;
 
         decl.link.macho.local_sym_index = 0;
     }
@@ -1415,7 +1430,7 @@ pub fn freeDecl(self: *MachO, decl: *Module.Decl) void {
 
 pub fn getDeclVAddr(self: *MachO, decl: *const Module.Decl) u64 {
     assert(decl.link.macho.local_sym_index != 0);
-    return self.local_symbols.items[decl.link.macho.local_sym_index].n_value;
+    return self.locals.items[decl.link.macho.local_sym_index].n_value;
 }
 
 pub fn populateMissingMetadata(self: *MachO) !void {
@@ -1555,39 +1570,6 @@ pub fn populateMissingMetadata(self: *MachO) !void {
         self.header_dirty = true;
         self.load_commands_dirty = true;
     }
-    if (self.got_section_index == null) {
-        const text_segment = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
-        self.got_section_index = @intCast(u16, text_segment.sections.items.len);
-
-        const alignment: u2 = switch (self.base.options.target.cpu.arch) {
-            .x86_64 => 0,
-            .aarch64 => 2,
-            else => unreachable, // unhandled architecture type
-        };
-        const flags = macho.S_REGULAR | macho.S_ATTR_PURE_INSTRUCTIONS | macho.S_ATTR_SOME_INSTRUCTIONS;
-        const needed_size = @sizeOf(u64) * self.base.options.symbol_count_hint;
-        const off = text_segment.findFreeSpace(needed_size, @alignOf(u64), self.header_pad);
-        assert(off + needed_size <= text_segment.inner.fileoff + text_segment.inner.filesize); // TODO Must expand __TEXT segment.
-
-        log.debug("found __ziggot section free space 0x{x} to 0x{x}", .{ off, off + needed_size });
-
-        try text_segment.addSection(self.base.allocator, .{
-            .sectname = makeStaticString("__ziggot"),
-            .segname = makeStaticString("__TEXT"),
-            .addr = text_segment.inner.vmaddr + off,
-            .size = needed_size,
-            .offset = @intCast(u32, off),
-            .@"align" = alignment,
-            .reloff = 0,
-            .nreloc = 0,
-            .flags = flags,
-            .reserved1 = 0,
-            .reserved2 = 0,
-            .reserved3 = 0,
-        });
-        self.header_dirty = true;
-        self.load_commands_dirty = true;
-    }
     if (self.stubs_section_index == null) {
         const text_segment = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
         self.stubs_section_index = @intCast(u16, text_segment.sections.items.len);
@@ -1599,7 +1581,7 @@ pub fn populateMissingMetadata(self: *MachO) !void {
         };
         const stub_size: u4 = switch (self.base.options.target.cpu.arch) {
             .x86_64 => 6,
-            .aarch64 => 2 * @sizeOf(u32),
+            .aarch64 => 3 * @sizeOf(u32),
             else => unreachable, // unhandled architecture type
         };
         const flags = macho.S_SYMBOL_STUBS | macho.S_ATTR_PURE_INSTRUCTIONS | macho.S_ATTR_SOME_INSTRUCTIONS;
@@ -1688,9 +1670,9 @@ pub fn populateMissingMetadata(self: *MachO) !void {
         self.header_dirty = true;
         self.load_commands_dirty = true;
     }
-    if (self.data_got_section_index == null) {
+    if (self.got_section_index == null) {
         const dc_segment = &self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
-        self.data_got_section_index = @intCast(u16, dc_segment.sections.items.len);
+        self.got_section_index = @intCast(u16, dc_segment.sections.items.len);
 
         const flags = macho.S_NON_LAZY_SYMBOL_POINTERS;
         const needed_size = @sizeOf(u64) * self.base.options.symbol_count_hint;
@@ -2062,12 +2044,12 @@ pub fn populateMissingMetadata(self: *MachO) !void {
         self.header_dirty = true;
         self.load_commands_dirty = true;
     }
-    if (!self.extern_nonlazy_symbols.contains("dyld_stub_binder")) {
-        const index = @intCast(u32, self.extern_nonlazy_symbols.items().len);
+    if (!self.nonlazy_imports.contains("dyld_stub_binder")) {
+        const index = @intCast(u32, self.nonlazy_imports.items().len);
         const name = try self.base.allocator.dupe(u8, "dyld_stub_binder");
         const offset = try self.makeString("dyld_stub_binder");
-        try self.extern_nonlazy_symbols.putNoClobber(self.base.allocator, name, .{
-            .inner = .{
+        try self.nonlazy_imports.putNoClobber(self.base.allocator, name, .{
+            .symbol = .{
                 .n_strx = offset,
                 .n_type = std.macho.N_UNDF | std.macho.N_EXT,
                 .n_sect = 0,
@@ -2075,68 +2057,19 @@ pub fn populateMissingMetadata(self: *MachO) !void {
                 .n_value = 0,
             },
             .dylib_ordinal = 1, // TODO this is currently hardcoded.
-            .segment = self.data_const_segment_cmd_index.?,
-            .offset = index * @sizeOf(u64),
+            .index = index,
+        });
+        const off_index = @intCast(u32, self.offset_table.items.len);
+        try self.offset_table.append(self.base.allocator, .{
+            .kind = .Extern,
+            .symbol = index,
+            .index = off_index,
         });
+        try self.writeOffsetTableEntry(off_index);
         self.binding_info_dirty = true;
     }
     if (self.stub_helper_stubs_start_off == null) {
-        const text_segment = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
-        const stub_helper = &text_segment.sections.items[self.stub_helper_section_index.?];
-        const data_segment = &self.load_commands.items[self.data_segment_cmd_index.?].Segment;
-        const data = &data_segment.sections.items[self.data_section_index.?];
-        const data_const_segment = &self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
-        const got = &data_const_segment.sections.items[self.data_got_section_index.?];
-        switch (self.base.options.target.cpu.arch) {
-            .x86_64 => {
-                const code_size = 15;
-                var code: [code_size]u8 = undefined;
-                // lea %r11, [rip + disp]
-                code[0] = 0x4c;
-                code[1] = 0x8d;
-                code[2] = 0x1d;
-                {
-                    const displacement = try math.cast(u32, data.addr - stub_helper.addr - 7);
-                    mem.writeIntLittle(u32, code[3..7], displacement);
-                }
-                // push %r11
-                code[7] = 0x41;
-                code[8] = 0x53;
-                // jmp [rip + disp]
-                code[9] = 0xff;
-                code[10] = 0x25;
-                {
-                    const displacement = try math.cast(u32, got.addr - stub_helper.addr - code_size);
-                    mem.writeIntLittle(u32, code[11..], displacement);
-                }
-                self.stub_helper_stubs_start_off = stub_helper.offset + code_size;
-                try self.base.file.?.pwriteAll(&code, stub_helper.offset);
-            },
-            .aarch64 => {
-                var code: [4 * @sizeOf(u32)]u8 = undefined;
-                {
-                    const displacement = try math.cast(i21, data.addr - stub_helper.addr);
-                    mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.adr(.x17, displacement).toU32());
-                }
-                mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.stp(
-                    .x16,
-                    .x17,
-                    aarch64.Register.sp,
-                    aarch64.Instruction.LoadStorePairOffset.pre_index(-16),
-                ).toU32());
-                {
-                    const displacement = try math.divExact(u64, got.addr - stub_helper.addr - 2 * @sizeOf(u32), 4);
-                    const literal = try math.cast(u19, displacement);
-                    mem.writeIntLittle(u32, code[8..12], aarch64.Instruction.ldr(.x16, .{
-                        .literal = literal,
-                    }).toU32());
-                }
-                mem.writeIntLittle(u32, code[12..16], aarch64.Instruction.br(.x16).toU32());
-                self.stub_helper_stubs_start_off = stub_helper.offset + 4 * @sizeOf(u32);
-                try self.base.file.?.pwriteAll(&code, stub_helper.offset);
-            },
-            else => unreachable,
-        }
+        try self.writeStubHelperPreamble();
     }
 }
 
@@ -2161,7 +2094,7 @@ fn allocateTextBlock(self: *MachO, text_block: *TextBlock, new_block_size: u64,
             const big_block = self.text_block_free_list.items[i];
             // We now have a pointer to a live text block that has too much capacity.
             // Is it enough that we could fit this new text block?
-            const sym = self.local_symbols.items[big_block.local_sym_index];
+            const sym = self.locals.items[big_block.local_sym_index];
             const capacity = big_block.capacity(self.*);
             const ideal_capacity = padToIdeal(capacity);
             const ideal_capacity_end_vaddr = sym.n_value + ideal_capacity;
@@ -2192,7 +2125,7 @@ fn allocateTextBlock(self: *MachO, text_block: *TextBlock, new_block_size: u64,
             }
             break :blk new_start_vaddr;
         } else if (self.last_text_block) |last| {
-            const last_symbol = self.local_symbols.items[last.local_sym_index];
+            const last_symbol = self.locals.items[last.local_sym_index];
             // TODO We should pad out the excess capacity with NOPs. For executables,
             // no padding seems to be OK, but it will probably not be for objects.
             const ideal_capacity = padToIdeal(last.size);
@@ -2290,12 +2223,12 @@ fn updateString(self: *MachO, old_str_off: u32, new_name: []const u8) !u32 {
 }
 
 pub fn addExternSymbol(self: *MachO, name: []const u8) !u32 {
-    const index = @intCast(u32, self.extern_lazy_symbols.items().len);
+    const index = @intCast(u32, self.lazy_imports.items().len);
     const offset = try self.makeString(name);
     const sym_name = try self.base.allocator.dupe(u8, name);
     const dylib_ordinal = 1; // TODO this is now hardcoded, since we only support libSystem.
-    try self.extern_lazy_symbols.putNoClobber(self.base.allocator, sym_name, .{
-        .inner = .{
+    try self.lazy_imports.putNoClobber(self.base.allocator, sym_name, .{
+        .symbol = .{
             .n_strx = offset,
             .n_type = macho.N_UNDF | macho.N_EXT,
             .n_sect = 0,
@@ -2303,6 +2236,7 @@ pub fn addExternSymbol(self: *MachO, name: []const u8) !u32 {
             .n_value = 0,
         },
         .dylib_ordinal = dylib_ordinal,
+        .index = index,
     });
     log.debug("adding new extern symbol '{s}' with dylib ordinal '{}'", .{ name, dylib_ordinal });
     return index;
@@ -2461,41 +2395,29 @@ fn findFreeSpaceLinkedit(self: *MachO, object_size: u64, min_alignment: u16, sta
 }
 
 fn writeOffsetTableEntry(self: *MachO, index: usize) !void {
-    const text_segment = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
-    const sect = &text_segment.sections.items[self.got_section_index.?];
+    const seg = &self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+    const sect = &seg.sections.items[self.got_section_index.?];
     const off = sect.offset + @sizeOf(u64) * index;
-    const vmaddr = sect.addr + @sizeOf(u64) * index;
 
     if (self.offset_table_count_dirty) {
         // TODO relocate.
         self.offset_table_count_dirty = false;
     }
 
-    var code: [8]u8 = undefined;
-    switch (self.base.options.target.cpu.arch) {
-        .x86_64 => {
-            const pos_symbol_off = try math.cast(u31, vmaddr - self.offset_table.items[index] + 7);
-            const symbol_off = @bitCast(u32, @as(i32, pos_symbol_off) * -1);
-            // lea %rax, [rip - disp]
-            code[0] = 0x48;
-            code[1] = 0x8D;
-            code[2] = 0x5;
-            mem.writeIntLittle(u32, code[3..7], symbol_off);
-            // ret
-            code[7] = 0xC3;
-        },
-        .aarch64 => {
-            const pos_symbol_off = try math.cast(u20, vmaddr - self.offset_table.items[index]);
-            const symbol_off = @as(i21, pos_symbol_off) * -1;
-            // adr x0, #-disp
-            mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.adr(.x0, symbol_off).toU32());
-            // ret x28
-            mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.ret(.x28).toU32());
-        },
-        else => unreachable, // unsupported target architecture
-    }
-    log.debug("writing offset table entry 0x{x} at 0x{x}", .{ self.offset_table.items[index], off });
-    try self.base.file.?.pwriteAll(&code, off);
+    const got_entry = self.offset_table.items[index];
+    const sym = blk: {
+        switch (got_entry.kind) {
+            .Local => {
+                break :blk self.locals.items[got_entry.symbol];
+            },
+            .Extern => {
+                break :blk self.nonlazy_imports.items()[got_entry.symbol].value.symbol;
+            },
+        }
+    };
+    const sym_name = self.getString(sym.n_strx);
+    log.debug("writing offset table entry [ 0x{x} => 0x{x} ({s}) ]", .{ off, sym.n_value, sym_name });
+    try self.base.file.?.pwriteAll(mem.asBytes(&sym.n_value), off);
 }
 
 fn writeLazySymbolPointer(self: *MachO, index: u32) !void {
@@ -2518,6 +2440,133 @@ fn writeLazySymbolPointer(self: *MachO, index: u32) !void {
     try self.base.file.?.pwriteAll(&buf, off);
 }
 
+fn writeStubHelperPreamble(self: *MachO) !void {
+    const text_segment = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    const stub_helper = &text_segment.sections.items[self.stub_helper_section_index.?];
+    const data_const_segment = &self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+    const got = &data_const_segment.sections.items[self.got_section_index.?];
+    const data_segment = &self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+    const data = &data_segment.sections.items[self.data_section_index.?];
+
+    switch (self.base.options.target.cpu.arch) {
+        .x86_64 => {
+            const code_size = 15;
+            var code: [code_size]u8 = undefined;
+            // lea %r11, [rip + disp]
+            code[0] = 0x4c;
+            code[1] = 0x8d;
+            code[2] = 0x1d;
+            {
+                const target_addr = data.addr;
+                const displacement = try math.cast(u32, target_addr - stub_helper.addr - 7);
+                mem.writeIntLittle(u32, code[3..7], displacement);
+            }
+            // push %r11
+            code[7] = 0x41;
+            code[8] = 0x53;
+            // jmp [rip + disp]
+            code[9] = 0xff;
+            code[10] = 0x25;
+            {
+                const displacement = try math.cast(u32, got.addr - stub_helper.addr - code_size);
+                mem.writeIntLittle(u32, code[11..], displacement);
+            }
+            try self.base.file.?.pwriteAll(&code, stub_helper.offset);
+            self.stub_helper_stubs_start_off = stub_helper.offset + code_size;
+        },
+        .aarch64 => {
+            var code: [6 * @sizeOf(u32)]u8 = undefined;
+
+            data_blk_outer: {
+                const this_addr = stub_helper.addr;
+                const target_addr = data.addr;
+                data_blk: {
+                    const displacement = math.cast(i21, target_addr - this_addr) catch |_| break :data_blk;
+                    // adr x17, disp
+                    mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.adr(.x17, displacement).toU32());
+                    // nop
+                    mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.nop().toU32());
+                    break :data_blk_outer;
+                }
+                data_blk: {
+                    const new_this_addr = this_addr + @sizeOf(u32);
+                    const displacement = math.cast(i21, target_addr - new_this_addr) catch |_| break :data_blk;
+                    // nop
+                    mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.nop().toU32());
+                    // adr x17, disp
+                    mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.adr(.x17, displacement).toU32());
+                    break :data_blk_outer;
+                }
+                // Jump is too big, replace adr with adrp and add.
+                const this_page = @intCast(i32, this_addr >> 12);
+                const target_page = @intCast(i32, target_addr >> 12);
+                const pages = @intCast(i21, target_page - this_page);
+                // adrp x17, pages
+                mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.adrp(.x17, pages).toU32());
+                const narrowed = @truncate(u12, target_addr);
+                mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.add(.x17, .x17, narrowed, false).toU32());
+            }
+
+            // stp x16, x17, [sp, #-16]!
+            mem.writeIntLittle(u32, code[8..12], aarch64.Instruction.stp(
+                .x16,
+                .x17,
+                aarch64.Register.sp,
+                aarch64.Instruction.LoadStorePairOffset.pre_index(-16),
+            ).toU32());
+
+            binder_blk_outer: {
+                const this_addr = stub_helper.addr + 3 * @sizeOf(u32);
+                const target_addr = got.addr;
+                binder_blk: {
+                    const displacement = math.divExact(u64, target_addr - this_addr, 4) catch |_| break :binder_blk;
+                    const literal = math.cast(u18, displacement) catch |_| break :binder_blk;
+                    // ldr x16, label
+                    mem.writeIntLittle(u32, code[12..16], aarch64.Instruction.ldr(.x16, .{
+                        .literal = literal,
+                    }).toU32());
+                    // nop
+                    mem.writeIntLittle(u32, code[16..20], aarch64.Instruction.nop().toU32());
+                    break :binder_blk_outer;
+                }
+                binder_blk: {
+                    const new_this_addr = this_addr + @sizeOf(u32);
+                    const displacement = math.divExact(u64, target_addr - new_this_addr, 4) catch |_| break :binder_blk;
+                    const literal = math.cast(u18, displacement) catch |_| break :binder_blk;
+                    // nop
+                    mem.writeIntLittle(u32, code[12..16], aarch64.Instruction.nop().toU32());
+                    // ldr x16, label
+                    mem.writeIntLittle(u32, code[16..20], aarch64.Instruction.ldr(.x16, .{
+                        .literal = literal,
+                    }).toU32());
+                    break :binder_blk_outer;
+                }
+                // Jump is too big, replace ldr with adrp and ldr(register).
+                const this_page = @intCast(i32, this_addr >> 12);
+                const target_page = @intCast(i32, target_addr >> 12);
+                const pages = @intCast(i21, target_page - this_page);
+                // adrp x16, pages
+                mem.writeIntLittle(u32, code[12..16], aarch64.Instruction.adrp(.x16, pages).toU32());
+                const narrowed = @truncate(u12, target_addr);
+                const offset = try math.divExact(u12, narrowed, 8);
+                // ldr x16, x16, offset
+                mem.writeIntLittle(u32, code[16..20], aarch64.Instruction.ldr(.x16, .{
+                    .register = .{
+                        .rn = .x16,
+                        .offset = aarch64.Instruction.LoadStoreOffset.imm(offset),
+                    },
+                }).toU32());
+            }
+
+            // br x16
+            mem.writeIntLittle(u32, code[20..24], aarch64.Instruction.br(.x16).toU32());
+            try self.base.file.?.pwriteAll(&code, stub_helper.offset);
+            self.stub_helper_stubs_start_off = stub_helper.offset + code.len;
+        },
+        else => unreachable,
+    }
+}
+
 fn writeStub(self: *MachO, index: u32) !void {
     const text_segment = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
     const stubs = text_segment.sections.items[self.stubs_section_index.?];
@@ -2527,9 +2576,12 @@ fn writeStub(self: *MachO, index: u32) !void {
     const stub_off = stubs.offset + index * stubs.reserved2;
     const stub_addr = stubs.addr + index * stubs.reserved2;
     const la_ptr_addr = la_symbol_ptr.addr + index * @sizeOf(u64);
+
     log.debug("writing stub at 0x{x}", .{stub_off});
+
     var code = try self.base.allocator.alloc(u8, stubs.reserved2);
     defer self.base.allocator.free(code);
+
     switch (self.base.options.target.cpu.arch) {
         .x86_64 => {
             assert(la_ptr_addr >= stub_addr + stubs.reserved2);
@@ -2541,12 +2593,50 @@ fn writeStub(self: *MachO, index: u32) !void {
         },
         .aarch64 => {
             assert(la_ptr_addr >= stub_addr);
-            const displacement = try math.divExact(u64, la_ptr_addr - stub_addr, 4);
-            const literal = try math.cast(u19, displacement);
-            mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.ldr(.x16, .{
-                .literal = literal,
-            }).toU32());
-            mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.br(.x16).toU32());
+            outer: {
+                const this_addr = stub_addr;
+                const target_addr = la_ptr_addr;
+                inner: {
+                    const displacement = math.divExact(u64, target_addr - this_addr, 4) catch |_| break :inner;
+                    const literal = math.cast(u18, displacement) catch |_| break :inner;
+                    // ldr x16, literal
+                    mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.ldr(.x16, .{
+                        .literal = literal,
+                    }).toU32());
+                    // nop
+                    mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.nop().toU32());
+                    break :outer;
+                }
+                inner: {
+                    const new_this_addr = this_addr + @sizeOf(u32);
+                    const displacement = math.divExact(u64, target_addr - new_this_addr, 4) catch |_| break :inner;
+                    const literal = math.cast(u18, displacement) catch |_| break :inner;
+                    // nop
+                    mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.nop().toU32());
+                    // ldr x16, literal
+                    mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.ldr(.x16, .{
+                        .literal = literal,
+                    }).toU32());
+                    break :outer;
+                }
+                // Use adrp followed by ldr(register).
+                const this_page = @intCast(i32, this_addr >> 12);
+                const target_page = @intCast(i32, target_addr >> 12);
+                const pages = @intCast(i21, target_page - this_page);
+                // adrp x16, pages
+                mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.adrp(.x16, pages).toU32());
+                const narrowed = @truncate(u12, target_addr);
+                const offset = try math.divExact(u12, narrowed, 8);
+                // ldr x16, x16, offset
+                mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.ldr(.x16, .{
+                    .register = .{
+                        .rn = .x16,
+                        .offset = aarch64.Instruction.LoadStoreOffset.imm(offset),
+                    },
+                }).toU32());
+            }
+            // br x16
+            mem.writeIntLittle(u32, code[8..12], aarch64.Instruction.br(.x16).toU32());
         },
         else => unreachable,
     }
@@ -2563,8 +2653,10 @@ fn writeStubInStubHelper(self: *MachO, index: u32) !void {
         else => unreachable,
     };
     const stub_off = self.stub_helper_stubs_start_off.? + index * stub_size;
+
     var code = try self.base.allocator.alloc(u8, stub_size);
     defer self.base.allocator.free(code);
+
     switch (self.base.options.target.cpu.arch) {
         .x86_64 => {
             const displacement = try math.cast(
@@ -2579,12 +2671,19 @@ fn writeStubInStubHelper(self: *MachO, index: u32) !void {
             mem.writeIntLittle(u32, code[6..][0..4], @bitCast(u32, displacement));
         },
         .aarch64 => {
-            const displacement = try math.cast(i28, @intCast(i64, stub_helper.offset) - @intCast(i64, stub_off) - 4);
+            const literal = blk: {
+                const div_res = try math.divExact(u64, stub_size - @sizeOf(u32), 4);
+                break :blk try math.cast(u18, div_res);
+            };
+            // ldr w16, literal
             mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.ldr(.w16, .{
-                .literal = @divExact(stub_size - @sizeOf(u32), 4),
+                .literal = literal,
             }).toU32());
+            const displacement = try math.cast(i28, @intCast(i64, stub_helper.offset) - @intCast(i64, stub_off) - 4);
+            // b disp
             mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.b(displacement).toU32());
-            mem.writeIntLittle(u32, code[8..12], 0x0); // Just a placeholder populated in `populateLazyBindOffsetsInStubHelper`.
+            // Just a placeholder populated in `populateLazyBindOffsetsInStubHelper`.
+            mem.writeIntLittle(u32, code[8..12], 0x0);
         },
         else => unreachable,
     }
@@ -2593,9 +2692,9 @@ fn writeStubInStubHelper(self: *MachO, index: u32) !void {
 
 fn relocateSymbolTable(self: *MachO) !void {
     const symtab = &self.load_commands.items[self.symtab_cmd_index.?].Symtab;
-    const nlocals = self.local_symbols.items.len;
-    const nglobals = self.global_symbols.items.len;
-    const nundefs = self.extern_lazy_symbols.items().len + self.extern_nonlazy_symbols.items().len;
+    const nlocals = self.locals.items.len;
+    const nglobals = self.globals.items.len;
+    const nundefs = self.lazy_imports.items().len + self.nonlazy_imports.items().len;
     const nsyms = nlocals + nglobals + nundefs;
 
     if (symtab.nsyms < nsyms) {
@@ -2630,7 +2729,7 @@ fn writeLocalSymbol(self: *MachO, index: usize) !void {
     const symtab = &self.load_commands.items[self.symtab_cmd_index.?].Symtab;
     const off = symtab.symoff + @sizeOf(macho.nlist_64) * index;
     log.debug("writing local symbol {} at 0x{x}", .{ index, off });
-    try self.base.file.?.pwriteAll(mem.asBytes(&self.local_symbols.items[index]), off);
+    try self.base.file.?.pwriteAll(mem.asBytes(&self.locals.items[index]), off);
 }
 
 fn writeAllGlobalAndUndefSymbols(self: *MachO) !void {
@@ -2639,18 +2738,18 @@ fn writeAllGlobalAndUndefSymbols(self: *MachO) !void {
 
     try self.relocateSymbolTable();
     const symtab = &self.load_commands.items[self.symtab_cmd_index.?].Symtab;
-    const nlocals = self.local_symbols.items.len;
-    const nglobals = self.global_symbols.items.len;
+    const nlocals = self.locals.items.len;
+    const nglobals = self.globals.items.len;
 
-    const nundefs = self.extern_lazy_symbols.items().len + self.extern_nonlazy_symbols.items().len;
+    const nundefs = self.lazy_imports.items().len + self.nonlazy_imports.items().len;
     var undefs = std.ArrayList(macho.nlist_64).init(self.base.allocator);
     defer undefs.deinit();
     try undefs.ensureCapacity(nundefs);
-    for (self.extern_lazy_symbols.items()) |entry| {
-        undefs.appendAssumeCapacity(entry.value.inner);
+    for (self.lazy_imports.items()) |entry| {
+        undefs.appendAssumeCapacity(entry.value.symbol);
     }
-    for (self.extern_nonlazy_symbols.items()) |entry| {
-        undefs.appendAssumeCapacity(entry.value.inner);
+    for (self.nonlazy_imports.items()) |entry| {
+        undefs.appendAssumeCapacity(entry.value.symbol);
     }
 
     const locals_off = symtab.symoff;
@@ -2659,7 +2758,7 @@ fn writeAllGlobalAndUndefSymbols(self: *MachO) !void {
     const globals_off = locals_off + locals_size;
     const globals_size = nglobals * @sizeOf(macho.nlist_64);
     log.debug("writing global symbols from 0x{x} to 0x{x}", .{ globals_off, globals_size + globals_off });
-    try self.base.file.?.pwriteAll(mem.sliceAsBytes(self.global_symbols.items), globals_off);
+    try self.base.file.?.pwriteAll(mem.sliceAsBytes(self.globals.items), globals_off);
 
     const undefs_off = globals_off + globals_size;
     const undefs_size = nundefs * @sizeOf(macho.nlist_64);
@@ -2685,15 +2784,15 @@ fn writeIndirectSymbolTable(self: *MachO) !void {
     const text_segment = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
     const stubs = &text_segment.sections.items[self.stubs_section_index.?];
     const data_const_seg = &self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
-    const got = &data_const_seg.sections.items[self.data_got_section_index.?];
+    const got = &data_const_seg.sections.items[self.got_section_index.?];
     const data_segment = &self.load_commands.items[self.data_segment_cmd_index.?].Segment;
     const la_symbol_ptr = &data_segment.sections.items[self.la_symbol_ptr_section_index.?];
     const dysymtab = &self.load_commands.items[self.dysymtab_cmd_index.?].Dysymtab;
 
-    const lazy = self.extern_lazy_symbols.items();
-    const nonlazy = self.extern_nonlazy_symbols.items();
+    const lazy = self.lazy_imports.items();
+    const got_entries = self.offset_table.items;
     const allocated_size = self.allocatedSizeLinkedit(dysymtab.indirectsymoff);
-    const nindirectsyms = @intCast(u32, lazy.len * 2 + nonlazy.len);
+    const nindirectsyms = @intCast(u32, lazy.len * 2 + got_entries.len);
     const needed_size = @intCast(u32, nindirectsyms * @sizeOf(u32));
 
     if (needed_size > allocated_size) {
@@ -2712,20 +2811,27 @@ fn writeIndirectSymbolTable(self: *MachO) !void {
     var writer = stream.writer();
 
     stubs.reserved1 = 0;
-    for (self.extern_lazy_symbols.items()) |_, i| {
+    for (lazy) |_, i| {
         const symtab_idx = @intCast(u32, dysymtab.iundefsym + i);
         try writer.writeIntLittle(u32, symtab_idx);
     }
 
     const base_id = @intCast(u32, lazy.len);
     got.reserved1 = base_id;
-    for (self.extern_nonlazy_symbols.items()) |_, i| {
-        const symtab_idx = @intCast(u32, dysymtab.iundefsym + i + base_id);
-        try writer.writeIntLittle(u32, symtab_idx);
+    for (got_entries) |entry| {
+        switch (entry.kind) {
+            .Local => {
+                try writer.writeIntLittle(u32, macho.INDIRECT_SYMBOL_LOCAL);
+            },
+            .Extern => {
+                const symtab_idx = @intCast(u32, dysymtab.iundefsym + entry.index + base_id);
+                try writer.writeIntLittle(u32, symtab_idx);
+            },
+        }
     }
 
-    la_symbol_ptr.reserved1 = got.reserved1 + @intCast(u32, nonlazy.len);
-    for (self.extern_lazy_symbols.items()) |_, i| {
+    la_symbol_ptr.reserved1 = got.reserved1 + @intCast(u32, got_entries.len);
+    for (lazy) |_, i| {
         const symtab_idx = @intCast(u32, dysymtab.iundefsym + i);
         try writer.writeIntLittle(u32, symtab_idx);
     }
@@ -2791,7 +2897,7 @@ fn writeCodeSignature(self: *MachO) !void {
 
 fn writeExportTrie(self: *MachO) !void {
     if (!self.export_info_dirty) return;
-    if (self.global_symbols.items.len == 0) return;
+    if (self.globals.items.len == 0) return;
 
     const tracy = trace(@src());
     defer tracy.end();
@@ -2800,7 +2906,7 @@ fn writeExportTrie(self: *MachO) !void {
     defer trie.deinit();
 
     const text_segment = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
-    for (self.global_symbols.items) |symbol| {
+    for (self.globals.items) |symbol| {
         // TODO figure out if we should put all global symbols into the export trie
         const name = self.getString(symbol.n_strx);
         assert(symbol.n_value >= text_segment.inner.vmaddr);
@@ -2842,14 +2948,48 @@ fn writeRebaseInfoTable(self: *MachO) !void {
     const tracy = trace(@src());
     defer tracy.end();
 
-    const size = try rebaseInfoSize(self.extern_lazy_symbols.items());
+    var pointers = std.ArrayList(bind.Pointer).init(self.base.allocator);
+    defer pointers.deinit();
+
+    if (self.got_section_index) |idx| {
+        const seg = self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+        const sect = seg.sections.items[idx];
+        const base_offset = sect.addr - seg.inner.vmaddr;
+        const segment_id = self.data_const_segment_cmd_index.?;
+
+        for (self.offset_table.items) |entry| {
+            if (entry.kind == .Extern) continue;
+            try pointers.append(.{
+                .offset = base_offset + entry.index * @sizeOf(u64),
+                .segment_id = segment_id,
+            });
+        }
+    }
+
+    if (self.la_symbol_ptr_section_index) |idx| {
+        try pointers.ensureCapacity(pointers.items.len + self.lazy_imports.items().len);
+        const seg = self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+        const sect = seg.sections.items[idx];
+        const base_offset = sect.addr - seg.inner.vmaddr;
+        const segment_id = self.data_segment_cmd_index.?;
+
+        for (self.lazy_imports.items()) |entry| {
+            pointers.appendAssumeCapacity(.{
+                .offset = base_offset + entry.value.index * @sizeOf(u64),
+                .segment_id = segment_id,
+            });
+        }
+    }
+
+    std.sort.sort(bind.Pointer, pointers.items, {}, bind.pointerCmp);
+
+    const size = try bind.rebaseInfoSize(pointers.items);
     var buffer = try self.base.allocator.alloc(u8, @intCast(usize, size));
     defer self.base.allocator.free(buffer);
 
     var stream = std.io.fixedBufferStream(buffer);
-    try writeRebaseInfo(self.extern_lazy_symbols.items(), stream.writer());
+    try bind.writeRebaseInfo(pointers.items, stream.writer());
 
-    const linkedit_segment = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
     const dyld_info = &self.load_commands.items[self.dyld_info_cmd_index.?].DyldInfoOnly;
     const allocated_size = self.allocatedSizeLinkedit(dyld_info.rebase_off);
     const needed_size = mem.alignForwardGeneric(u64, buffer.len, @alignOf(u64));
@@ -2874,14 +3014,34 @@ fn writeBindingInfoTable(self: *MachO) !void {
     const tracy = trace(@src());
     defer tracy.end();
 
-    const size = try bindInfoSize(self.extern_nonlazy_symbols.items());
+    var pointers = std.ArrayList(bind.Pointer).init(self.base.allocator);
+    defer pointers.deinit();
+
+    if (self.got_section_index) |idx| {
+        const seg = self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+        const sect = seg.sections.items[idx];
+        const base_offset = sect.addr - seg.inner.vmaddr;
+        const segment_id = @intCast(u16, self.data_const_segment_cmd_index.?);
+
+        for (self.offset_table.items) |entry| {
+            if (entry.kind == .Local) continue;
+            const import = self.nonlazy_imports.items()[entry.symbol];
+            try pointers.append(.{
+                .offset = base_offset + entry.index * @sizeOf(u64),
+                .segment_id = segment_id,
+                .dylib_ordinal = import.value.dylib_ordinal,
+                .name = import.key,
+            });
+        }
+    }
+
+    const size = try bind.bindInfoSize(pointers.items);
     var buffer = try self.base.allocator.alloc(u8, @intCast(usize, size));
     defer self.base.allocator.free(buffer);
 
     var stream = std.io.fixedBufferStream(buffer);
-    try writeBindInfo(self.extern_nonlazy_symbols.items(), stream.writer());
+    try bind.writeBindInfo(pointers.items, stream.writer());
 
-    const linkedit_segment = self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
     const dyld_info = &self.load_commands.items[self.dyld_info_cmd_index.?].DyldInfoOnly;
     const allocated_size = self.allocatedSizeLinkedit(dyld_info.bind_off);
     const needed_size = mem.alignForwardGeneric(u64, buffer.len, @alignOf(u64));
@@ -2903,14 +3063,36 @@ fn writeBindingInfoTable(self: *MachO) !void {
 fn writeLazyBindingInfoTable(self: *MachO) !void {
     if (!self.lazy_binding_info_dirty) return;
 
-    const size = try lazyBindInfoSize(self.extern_lazy_symbols.items());
+    const tracy = trace(@src());
+    defer tracy.end();
+
+    var pointers = std.ArrayList(bind.Pointer).init(self.base.allocator);
+    defer pointers.deinit();
+
+    if (self.la_symbol_ptr_section_index) |idx| {
+        try pointers.ensureCapacity(self.lazy_imports.items().len);
+        const seg = self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+        const sect = seg.sections.items[idx];
+        const base_offset = sect.addr - seg.inner.vmaddr;
+        const segment_id = @intCast(u16, self.data_segment_cmd_index.?);
+
+        for (self.lazy_imports.items()) |entry| {
+            pointers.appendAssumeCapacity(.{
+                .offset = base_offset + entry.value.index * @sizeOf(u64),
+                .segment_id = segment_id,
+                .dylib_ordinal = entry.value.dylib_ordinal,
+                .name = entry.key,
+            });
+        }
+    }
+
+    const size = try bind.lazyBindInfoSize(pointers.items);
     var buffer = try self.base.allocator.alloc(u8, @intCast(usize, size));
     defer self.base.allocator.free(buffer);
 
     var stream = std.io.fixedBufferStream(buffer);
-    try writeLazyBindInfo(self.extern_lazy_symbols.items(), stream.writer());
+    try bind.writeLazyBindInfo(pointers.items, stream.writer());
 
-    const linkedit_segment = self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
     const dyld_info = &self.load_commands.items[self.dyld_info_cmd_index.?].DyldInfoOnly;
     const allocated_size = self.allocatedSizeLinkedit(dyld_info.lazy_bind_off);
     const needed_size = mem.alignForwardGeneric(u64, buffer.len, @alignOf(u64));
@@ -2931,7 +3113,7 @@ fn writeLazyBindingInfoTable(self: *MachO) !void {
 }
 
 fn populateLazyBindOffsetsInStubHelper(self: *MachO, buffer: []const u8) !void {
-    if (self.extern_lazy_symbols.items().len == 0) return;
+    if (self.lazy_imports.items().len == 0) return;
 
     var stream = std.io.fixedBufferStream(buffer);
     var reader = stream.reader();
@@ -2977,7 +3159,7 @@ fn populateLazyBindOffsetsInStubHelper(self: *MachO, buffer: []const u8) !void {
             else => {},
         }
     }
-    assert(self.extern_lazy_symbols.items().len <= offsets.items.len);
+    assert(self.lazy_imports.items().len <= offsets.items.len);
 
     const stub_size: u4 = switch (self.base.options.target.cpu.arch) {
         .x86_64 => 10,
@@ -2990,7 +3172,7 @@ fn populateLazyBindOffsetsInStubHelper(self: *MachO, buffer: []const u8) !void {
         else => unreachable,
     };
     var buf: [@sizeOf(u32)]u8 = undefined;
-    for (self.extern_lazy_symbols.items()) |_, i| {
+    for (self.lazy_imports.items()) |_, i| {
         const placeholder_off = self.stub_helper_stubs_start_off.? + i * stub_size + off;
         mem.writeIntLittle(u32, &buf, offsets.items[i]);
         try self.base.file.?.pwriteAll(&buf, placeholder_off);
@@ -3104,177 +3286,6 @@ fn writeHeader(self: *MachO) !void {
     self.header_dirty = false;
 }
 
-/// Parse MachO contents from existing binary file.
-fn parseFromFile(self: *MachO, file: fs.File) !void {
-    self.base.file = file;
-    var reader = file.reader();
-    const header = try reader.readStruct(macho.mach_header_64);
-    try self.load_commands.ensureCapacity(self.base.allocator, header.ncmds);
-    var i: u16 = 0;
-    while (i < header.ncmds) : (i += 1) {
-        const cmd = try LoadCommand.read(self.base.allocator, reader);
-        switch (cmd.cmd()) {
-            macho.LC_SEGMENT_64 => {
-                const x = cmd.Segment;
-                if (parseAndCmpName(&x.inner.segname, "__PAGEZERO")) {
-                    self.pagezero_segment_cmd_index = i;
-                } else if (parseAndCmpName(&x.inner.segname, "__LINKEDIT")) {
-                    self.linkedit_segment_cmd_index = i;
-                } else if (parseAndCmpName(&x.inner.segname, "__TEXT")) {
-                    self.text_segment_cmd_index = i;
-                    for (x.sections.items) |sect, j| {
-                        if (parseAndCmpName(&sect.sectname, "__text")) {
-                            self.text_section_index = @intCast(u16, j);
-                        }
-                    }
-                } else if (parseAndCmpName(&x.inner.segname, "__DATA")) {
-                    self.data_segment_cmd_index = i;
-                } else if (parseAndCmpName(&x.inner.segname, "__DATA_CONST")) {
-                    self.data_const_segment_cmd_index = i;
-                }
-            },
-            macho.LC_DYLD_INFO_ONLY => {
-                self.dyld_info_cmd_index = i;
-            },
-            macho.LC_SYMTAB => {
-                self.symtab_cmd_index = i;
-            },
-            macho.LC_DYSYMTAB => {
-                self.dysymtab_cmd_index = i;
-            },
-            macho.LC_LOAD_DYLINKER => {
-                self.dylinker_cmd_index = i;
-            },
-            macho.LC_VERSION_MIN_MACOSX, macho.LC_VERSION_MIN_IPHONEOS, macho.LC_VERSION_MIN_WATCHOS, macho.LC_VERSION_MIN_TVOS => {
-                self.version_min_cmd_index = i;
-            },
-            macho.LC_SOURCE_VERSION => {
-                self.source_version_cmd_index = i;
-            },
-            macho.LC_UUID => {
-                self.uuid_cmd_index = i;
-            },
-            macho.LC_MAIN => {
-                self.main_cmd_index = i;
-            },
-            macho.LC_LOAD_DYLIB => {
-                const x = cmd.Dylib;
-                if (parseAndCmpName(x.data, mem.spanZ(LIB_SYSTEM_PATH))) {
-                    self.libsystem_cmd_index = i;
-                }
-            },
-            macho.LC_FUNCTION_STARTS => {
-                self.function_starts_cmd_index = i;
-            },
-            macho.LC_DATA_IN_CODE => {
-                self.data_in_code_cmd_index = i;
-            },
-            macho.LC_CODE_SIGNATURE => {
-                self.code_signature_cmd_index = i;
-            },
-            else => {
-                log.warn("Unknown load command detected: 0x{x}.", .{cmd.cmd()});
-            },
-        }
-        self.load_commands.appendAssumeCapacity(cmd);
-    }
-    self.header = header;
-}
-
-fn parseAndCmpName(name: []const u8, needle: []const u8) bool {
-    const len = mem.indexOfScalar(u8, name, @as(u8, 0)) orelse name.len;
-    return mem.eql(u8, name[0..len], needle);
-}
-
-fn parseSymbolTable(self: *MachO) !void {
-    const symtab = self.load_commands.items[self.symtab_cmd_index.?].Symtab;
-    const dysymtab = self.load_commands.items[self.dysymtab_cmd_index.?].Dysymtab;
-
-    var buffer = try self.base.allocator.alloc(macho.nlist_64, symtab.nsyms);
-    defer self.base.allocator.free(buffer);
-    const nread = try self.base.file.?.preadAll(@ptrCast([*]u8, buffer)[0 .. symtab.nsyms * @sizeOf(macho.nlist_64)], symtab.symoff);
-    assert(@divExact(nread, @sizeOf(macho.nlist_64)) == buffer.len);
-
-    try self.local_symbols.ensureCapacity(self.base.allocator, dysymtab.nlocalsym);
-    try self.global_symbols.ensureCapacity(self.base.allocator, dysymtab.nextdefsym);
-    try self.undef_symbols.ensureCapacity(self.base.allocator, dysymtab.nundefsym);
-
-    self.local_symbols.appendSliceAssumeCapacity(buffer[dysymtab.ilocalsym .. dysymtab.ilocalsym + dysymtab.nlocalsym]);
-    self.global_symbols.appendSliceAssumeCapacity(buffer[dysymtab.iextdefsym .. dysymtab.iextdefsym + dysymtab.nextdefsym]);
-    self.undef_symbols.appendSliceAssumeCapacity(buffer[dysymtab.iundefsym .. dysymtab.iundefsym + dysymtab.nundefsym]);
-}
-
-fn parseStringTable(self: *MachO) !void {
-    const symtab = self.load_commands.items[self.symtab_cmd_index.?].Symtab;
-
-    var buffer = try self.base.allocator.alloc(u8, symtab.strsize);
-    defer self.base.allocator.free(buffer);
-    const nread = try self.base.file.?.preadAll(buffer, symtab.stroff);
-    assert(nread == buffer.len);
-
-    try self.string_table.ensureCapacity(self.base.allocator, symtab.strsize);
-    self.string_table.appendSliceAssumeCapacity(buffer);
-}
-
-fn fixupBindInfo(self: *MachO, dylib_ordinal: u32) !void {
-    const dyld_info = self.load_commands.items[self.dyld_info_cmd_index.?].DyldInfoOnly;
-    var buffer = try self.base.allocator.alloc(u8, dyld_info.bind_size);
-    defer self.base.allocator.free(buffer);
-    const nread = try self.base.file.?.preadAll(buffer, dyld_info.bind_off);
-    assert(nread == buffer.len);
-    try self.fixupInfoCommon(buffer, dylib_ordinal);
-    try self.base.file.?.pwriteAll(buffer, dyld_info.bind_off);
-}
-
-fn fixupLazyBindInfo(self: *MachO, dylib_ordinal: u32) !void {
-    const dyld_info = self.load_commands.items[self.dyld_info_cmd_index.?].DyldInfoOnly;
-    var buffer = try self.base.allocator.alloc(u8, dyld_info.lazy_bind_size);
-    defer self.base.allocator.free(buffer);
-    const nread = try self.base.file.?.preadAll(buffer, dyld_info.lazy_bind_off);
-    assert(nread == buffer.len);
-    try self.fixupInfoCommon(buffer, dylib_ordinal);
-    try self.base.file.?.pwriteAll(buffer, dyld_info.lazy_bind_off);
-}
-
-fn fixupInfoCommon(self: *MachO, buffer: []u8, dylib_ordinal: u32) !void {
-    var stream = std.io.fixedBufferStream(buffer);
-    var reader = stream.reader();
-
-    while (true) {
-        const inst = reader.readByte() catch |err| switch (err) {
-            error.EndOfStream => break,
-            else => return err,
-        };
-        const imm: u8 = inst & macho.BIND_IMMEDIATE_MASK;
-        const opcode: u8 = inst & macho.BIND_OPCODE_MASK;
-
-        switch (opcode) {
-            macho.BIND_OPCODE_SET_SYMBOL_TRAILING_FLAGS_IMM => {
-                var next = try reader.readByte();
-                while (next != @as(u8, 0)) {
-                    next = try reader.readByte();
-                }
-            },
-            macho.BIND_OPCODE_SET_SEGMENT_AND_OFFSET_ULEB => {
-                _ = try std.leb.readULEB128(u64, reader);
-            },
-            macho.BIND_OPCODE_SET_DYLIB_SPECIAL_IMM, macho.BIND_OPCODE_SET_DYLIB_ORDINAL_IMM => {
-                // Perform the fixup.
-                try stream.seekBy(-1);
-                var writer = stream.writer();
-                try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_ORDINAL_IMM | @truncate(u4, dylib_ordinal));
-            },
-            macho.BIND_OPCODE_SET_DYLIB_ORDINAL_ULEB => {
-                _ = try std.leb.readULEB128(u64, reader);
-            },
-            macho.BIND_OPCODE_SET_ADDEND_SLEB => {
-                _ = try std.leb.readILEB128(i64, reader);
-            },
-            else => {},
-        }
-    }
-}
-
 pub fn padToIdeal(actual_size: anytype) @TypeOf(actual_size) {
     // TODO https://github.com/ziglang/zig/issues/1284
     return std.math.add(@TypeOf(actual_size), actual_size, actual_size / ideal_factor) catch
diff --git a/src/link/MachO/Archive.zig b/src/link/MachO/Archive.zig
new file mode 100644
index 0000000000..bb6fb68231
--- /dev/null
+++ b/src/link/MachO/Archive.zig
@@ -0,0 +1,278 @@
+const Archive = @This();
+
+const std = @import("std");
+const assert = std.debug.assert;
+const fs = std.fs;
+const log = std.log.scoped(.archive);
+const macho = std.macho;
+const mem = std.mem;
+
+const Allocator = mem.Allocator;
+const Object = @import("Object.zig");
+const parseName = @import("Zld.zig").parseName;
+
+usingnamespace @import("commands.zig");
+
+allocator: *Allocator,
+file: fs.File,
+header: ar_hdr,
+name: []u8,
+
+objects: std.ArrayListUnmanaged(Object) = .{},
+
+/// Parsed table of contents.
+/// Each symbol name points to a list of all definition
+/// sites within the current static archive.
+toc: std.StringArrayHashMapUnmanaged(std.ArrayListUnmanaged(u32)) = .{},
+
+// Archive files start with the ARMAG identifying string.  Then follows a
+// `struct ar_hdr', and as many bytes of member file data as its `ar_size'
+// member indicates, for each member file.
+/// String that begins an archive file.
+const ARMAG: *const [SARMAG:0]u8 = "!<arch>\n";
+/// Size of that string.
+const SARMAG: u4 = 8;
+
+/// String in ar_fmag at the end of each header.
+const ARFMAG: *const [2:0]u8 = "`\n";
+
+const ar_hdr = extern struct {
+    /// Member file name, sometimes / terminated.
+    ar_name: [16]u8,
+
+    /// File date, decimal seconds since Epoch.
+    ar_date: [12]u8,
+
+    /// User ID, in ASCII format.
+    ar_uid: [6]u8,
+
+    /// Group ID, in ASCII format.
+    ar_gid: [6]u8,
+
+    /// File mode, in ASCII octal.
+    ar_mode: [8]u8,
+
+    /// File size, in ASCII decimal.
+    ar_size: [10]u8,
+
+    /// Always contains ARFMAG.
+    ar_fmag: [2]u8,
+
+    const NameOrLength = union(enum) {
+        Name: []const u8,
+        Length: u64,
+    };
+    pub fn nameOrLength(self: ar_hdr) !NameOrLength {
+        const value = getValue(&self.ar_name);
+        const slash_index = mem.indexOf(u8, value, "/") orelse return error.MalformedArchive;
+        const len = value.len;
+        if (slash_index == len - 1) {
+            // Name stored directly
+            return NameOrLength{ .Name = value };
+        } else {
+            // Name follows the header directly and its length is encoded in
+            // the name field.
+            const length = try std.fmt.parseInt(u64, value[slash_index + 1 ..], 10);
+            return NameOrLength{ .Length = length };
+        }
+    }
+
+    pub fn size(self: ar_hdr) !u64 {
+        const value = getValue(&self.ar_size);
+        return std.fmt.parseInt(u64, value, 10);
+    }
+
+    fn getValue(raw: []const u8) []const u8 {
+        return mem.trimRight(u8, raw, &[_]u8{@as(u8, 0x20)});
+    }
+};
+
+pub fn deinit(self: *Archive) void {
+    self.allocator.free(self.name);
+    for (self.objects.items) |*object| {
+        object.deinit();
+    }
+    self.objects.deinit(self.allocator);
+    for (self.toc.items()) |*entry| {
+        self.allocator.free(entry.key);
+        entry.value.deinit(self.allocator);
+    }
+    self.toc.deinit(self.allocator);
+    self.file.close();
+}
+
+/// Caller owns the returned Archive instance and is responsible for calling
+/// `deinit` to free allocated memory.
+pub fn initFromFile(allocator: *Allocator, arch: std.Target.Cpu.Arch, ar_name: []const u8, file: fs.File) !Archive {
+    var reader = file.reader();
+    var magic = try readMagic(allocator, reader);
+    defer allocator.free(magic);
+
+    if (!mem.eql(u8, magic, ARMAG)) {
+        // Reset file cursor.
+        try file.seekTo(0);
+        return error.NotArchive;
+    }
+
+    const header = try reader.readStruct(ar_hdr);
+
+    if (!mem.eql(u8, &header.ar_fmag, ARFMAG))
+        return error.MalformedArchive;
+
+    var embedded_name = try getName(allocator, header, reader);
+    log.debug("parsing archive '{s}' at '{s}'", .{ embedded_name, ar_name });
+    defer allocator.free(embedded_name);
+
+    var name = try allocator.dupe(u8, ar_name);
+    var self = Archive{
+        .allocator = allocator,
+        .file = file,
+        .header = header,
+        .name = name,
+    };
+
+    var object_offsets = try self.readTableOfContents(reader);
+    defer self.allocator.free(object_offsets);
+
+    var i: usize = 1;
+    while (i < object_offsets.len) : (i += 1) {
+        const offset = object_offsets[i];
+        try reader.context.seekTo(offset);
+        try self.readObject(arch, ar_name, reader);
+    }
+
+    return self;
+}
+
+fn readTableOfContents(self: *Archive, reader: anytype) ![]u32 {
+    const symtab_size = try reader.readIntLittle(u32);
+    var symtab = try self.allocator.alloc(u8, symtab_size);
+    defer self.allocator.free(symtab);
+    try reader.readNoEof(symtab);
+
+    const strtab_size = try reader.readIntLittle(u32);
+    var strtab = try self.allocator.alloc(u8, strtab_size);
+    defer self.allocator.free(strtab);
+    try reader.readNoEof(strtab);
+
+    var symtab_stream = std.io.fixedBufferStream(symtab);
+    var symtab_reader = symtab_stream.reader();
+
+    var object_offsets = std.ArrayList(u32).init(self.allocator);
+    try object_offsets.append(0);
+    var last: usize = 0;
+
+    while (true) {
+        const n_strx = symtab_reader.readIntLittle(u32) catch |err| switch (err) {
+            error.EndOfStream => break,
+            else => |e| return e,
+        };
+        const object_offset = try symtab_reader.readIntLittle(u32);
+
+        const sym_name = mem.spanZ(@ptrCast([*:0]const u8, strtab.ptr + n_strx));
+        const owned_name = try self.allocator.dupe(u8, sym_name);
+        const res = try self.toc.getOrPut(self.allocator, owned_name);
+        defer if (res.found_existing) self.allocator.free(owned_name);
+
+        if (!res.found_existing) {
+            res.entry.value = .{};
+        }
+
+        try res.entry.value.append(self.allocator, object_offset);
+
+        // TODO This will go once we properly use archive's TOC to pick
+        // an object which defines a missing symbol rather than pasting in
+        // all of the objects always.
+        // Here, we assume that symbols are NOT sorted in any way, and
+        // they point to objects in sequence.
+        if (object_offsets.items[last] != object_offset) {
+            try object_offsets.append(object_offset);
+            last += 1;
+        }
+    }
+
+    return object_offsets.toOwnedSlice();
+}
+
+fn readObject(self: *Archive, arch: std.Target.Cpu.Arch, ar_name: []const u8, reader: anytype) !void {
+    const object_header = try reader.readStruct(ar_hdr);
+
+    if (!mem.eql(u8, &object_header.ar_fmag, ARFMAG))
+        return error.MalformedArchive;
+
+    var object_name = try getName(self.allocator, object_header, reader);
+    log.debug("extracting object '{s}' from archive '{s}'", .{ object_name, self.name });
+
+    const offset = @intCast(u32, try reader.context.getPos());
+    const header = try reader.readStruct(macho.mach_header_64);
+
+    const this_arch: std.Target.Cpu.Arch = switch (header.cputype) {
+        macho.CPU_TYPE_ARM64 => .aarch64,
+        macho.CPU_TYPE_X86_64 => .x86_64,
+        else => |value| {
+            log.err("unsupported cpu architecture 0x{x}", .{value});
+            return error.UnsupportedCpuArchitecture;
+        },
+    };
+    if (this_arch != arch) {
+        log.err("mismatched cpu architecture: found {s}, expected {s}", .{ this_arch, arch });
+        return error.MismatchedCpuArchitecture;
+    }
+
+    // TODO Implement std.fs.File.clone() or similar.
+    var new_file = try fs.cwd().openFile(ar_name, .{});
+    var object = Object{
+        .allocator = self.allocator,
+        .name = object_name,
+        .ar_name = try mem.dupe(self.allocator, u8, ar_name),
+        .file = new_file,
+        .header = header,
+    };
+
+    try object.readLoadCommands(reader, .{ .offset = offset });
+
+    if (object.symtab_cmd_index != null) {
+        try object.readSymtab();
+        try object.readStrtab();
+    }
+
+    if (object.data_in_code_cmd_index != null) try object.readDataInCode();
+
+    log.debug("\n\n", .{});
+    log.debug("{s} defines symbols", .{object.name});
+    for (object.symtab.items) |sym| {
+        const symname = object.getString(sym.n_strx);
+        log.debug("'{s}': {}", .{ symname, sym });
+    }
+
+    try self.objects.append(self.allocator, object);
+}
+
+fn readMagic(allocator: *Allocator, reader: anytype) ![]u8 {
+    var magic = std.ArrayList(u8).init(allocator);
+    try magic.ensureCapacity(SARMAG);
+    var i: usize = 0;
+    while (i < SARMAG) : (i += 1) {
+        const next = try reader.readByte();
+        magic.appendAssumeCapacity(next);
+    }
+    return magic.toOwnedSlice();
+}
+
+fn getName(allocator: *Allocator, header: ar_hdr, reader: anytype) ![]u8 {
+    const name_or_length = try header.nameOrLength();
+    var name: []u8 = undefined;
+    switch (name_or_length) {
+        .Name => |n| {
+            name = try allocator.dupe(u8, n);
+        },
+        .Length => |len| {
+            var n = try allocator.alloc(u8, len);
+            defer allocator.free(n);
+            try reader.readNoEof(n);
+            const actual_len = mem.indexOfScalar(u8, n, @as(u8, 0)) orelse n.len;
+            name = try allocator.dupe(u8, n[0..actual_len]);
+        },
+    }
+    return name;
+}
diff --git a/src/link/MachO/DebugSymbols.zig b/src/link/MachO/DebugSymbols.zig
index 042c1a12cf..a81fd00c0a 100644
--- a/src/link/MachO/DebugSymbols.zig
+++ b/src/link/MachO/DebugSymbols.zig
@@ -839,8 +839,8 @@ fn findFreeSpaceLinkedit(self: *DebugSymbols, object_size: u64, min_alignment: u
 
 fn relocateSymbolTable(self: *DebugSymbols) !void {
     const symtab = &self.load_commands.items[self.symtab_cmd_index.?].Symtab;
-    const nlocals = self.base.local_symbols.items.len;
-    const nglobals = self.base.global_symbols.items.len;
+    const nlocals = self.base.locals.items.len;
+    const nglobals = self.base.globals.items.len;
     const nsyms = nlocals + nglobals;
 
     if (symtab.nsyms < nsyms) {
@@ -875,7 +875,7 @@ pub fn writeLocalSymbol(self: *DebugSymbols, index: usize) !void {
     const symtab = &self.load_commands.items[self.symtab_cmd_index.?].Symtab;
     const off = symtab.symoff + @sizeOf(macho.nlist_64) * index;
     log.debug("writing dSym local symbol {} at 0x{x}", .{ index, off });
-    try self.file.pwriteAll(mem.asBytes(&self.base.local_symbols.items[index]), off);
+    try self.file.pwriteAll(mem.asBytes(&self.base.locals.items[index]), off);
 }
 
 fn writeStringTable(self: *DebugSymbols) !void {
@@ -1057,7 +1057,7 @@ pub fn commitDeclDebugInfo(
     var dbg_info_buffer = &debug_buffers.dbg_info_buffer;
     var dbg_info_type_relocs = &debug_buffers.dbg_info_type_relocs;
 
-    const symbol = self.base.local_symbols.items[decl.link.macho.local_sym_index];
+    const symbol = self.base.locals.items[decl.link.macho.local_sym_index];
     const text_block = &decl.link.macho;
     // If the Decl is a function, we need to update the __debug_line program.
     const typed_value = decl.typed_value.most_recent.typed_value;
diff --git a/src/link/MachO/Object.zig b/src/link/MachO/Object.zig
new file mode 100644
index 0000000000..a8293d6410
--- /dev/null
+++ b/src/link/MachO/Object.zig
@@ -0,0 +1,229 @@
+const Object = @This();
+
+const std = @import("std");
+const assert = std.debug.assert;
+const fs = std.fs;
+const io = std.io;
+const log = std.log.scoped(.object);
+const macho = std.macho;
+const mem = std.mem;
+
+const Allocator = mem.Allocator;
+const parseName = @import("Zld.zig").parseName;
+
+usingnamespace @import("commands.zig");
+
+allocator: *Allocator,
+file: fs.File,
+name: []u8,
+ar_name: ?[]u8 = null,
+
+header: macho.mach_header_64,
+
+load_commands: std.ArrayListUnmanaged(LoadCommand) = .{},
+
+segment_cmd_index: ?u16 = null,
+symtab_cmd_index: ?u16 = null,
+dysymtab_cmd_index: ?u16 = null,
+build_version_cmd_index: ?u16 = null,
+data_in_code_cmd_index: ?u16 = null,
+text_section_index: ?u16 = null,
+
+// __DWARF segment sections
+dwarf_debug_info_index: ?u16 = null,
+dwarf_debug_abbrev_index: ?u16 = null,
+dwarf_debug_str_index: ?u16 = null,
+dwarf_debug_line_index: ?u16 = null,
+dwarf_debug_ranges_index: ?u16 = null,
+
+symtab: std.ArrayListUnmanaged(macho.nlist_64) = .{},
+strtab: std.ArrayListUnmanaged(u8) = .{},
+
+data_in_code_entries: std.ArrayListUnmanaged(macho.data_in_code_entry) = .{},
+
+pub fn deinit(self: *Object) void {
+    for (self.load_commands.items) |*lc| {
+        lc.deinit(self.allocator);
+    }
+    self.load_commands.deinit(self.allocator);
+    self.symtab.deinit(self.allocator);
+    self.strtab.deinit(self.allocator);
+    self.data_in_code_entries.deinit(self.allocator);
+    self.allocator.free(self.name);
+    if (self.ar_name) |v| {
+        self.allocator.free(v);
+    }
+    self.file.close();
+}
+
+/// Caller owns the returned Object instance and is responsible for calling
+/// `deinit` to free allocated memory.
+pub fn initFromFile(allocator: *Allocator, arch: std.Target.Cpu.Arch, name: []const u8, file: fs.File) !Object {
+    var reader = file.reader();
+    const header = try reader.readStruct(macho.mach_header_64);
+
+    if (header.filetype != macho.MH_OBJECT) {
+        // Reset file cursor.
+        try file.seekTo(0);
+        return error.NotObject;
+    }
+
+    const this_arch: std.Target.Cpu.Arch = switch (header.cputype) {
+        macho.CPU_TYPE_ARM64 => .aarch64,
+        macho.CPU_TYPE_X86_64 => .x86_64,
+        else => |value| {
+            log.err("unsupported cpu architecture 0x{x}", .{value});
+            return error.UnsupportedCpuArchitecture;
+        },
+    };
+    if (this_arch != arch) {
+        log.err("mismatched cpu architecture: found {s}, expected {s}", .{ this_arch, arch });
+        return error.MismatchedCpuArchitecture;
+    }
+
+    var self = Object{
+        .allocator = allocator,
+        .name = try allocator.dupe(u8, name),
+        .file = file,
+        .header = header,
+    };
+
+    try self.readLoadCommands(reader, .{});
+
+    if (self.symtab_cmd_index != null) {
+        try self.readSymtab();
+        try self.readStrtab();
+    }
+
+    if (self.data_in_code_cmd_index != null) try self.readDataInCode();
+
+    log.debug("\n\n", .{});
+    log.debug("{s} defines symbols", .{self.name});
+    for (self.symtab.items) |sym| {
+        const symname = self.getString(sym.n_strx);
+        log.debug("'{s}': {}", .{ symname, sym });
+    }
+
+    return self;
+}
+
+pub const ReadOffset = struct {
+    offset: ?u32 = null,
+};
+
+pub fn readLoadCommands(self: *Object, reader: anytype, offset: ReadOffset) !void {
+    const offset_mod = offset.offset orelse 0;
+    try self.load_commands.ensureCapacity(self.allocator, self.header.ncmds);
+
+    var i: u16 = 0;
+    while (i < self.header.ncmds) : (i += 1) {
+        var cmd = try LoadCommand.read(self.allocator, reader);
+        switch (cmd.cmd()) {
+            macho.LC_SEGMENT_64 => {
+                self.segment_cmd_index = i;
+                var seg = cmd.Segment;
+                for (seg.sections.items) |*sect, j| {
+                    const index = @intCast(u16, j);
+                    const segname = parseName(&sect.segname);
+                    const sectname = parseName(&sect.sectname);
+                    if (mem.eql(u8, segname, "__DWARF")) {
+                        if (mem.eql(u8, sectname, "__debug_info")) {
+                            self.dwarf_debug_info_index = index;
+                        } else if (mem.eql(u8, sectname, "__debug_abbrev")) {
+                            self.dwarf_debug_abbrev_index = index;
+                        } else if (mem.eql(u8, sectname, "__debug_str")) {
+                            self.dwarf_debug_str_index = index;
+                        } else if (mem.eql(u8, sectname, "__debug_line")) {
+                            self.dwarf_debug_line_index = index;
+                        } else if (mem.eql(u8, sectname, "__debug_ranges")) {
+                            self.dwarf_debug_ranges_index = index;
+                        }
+                    } else if (mem.eql(u8, segname, "__TEXT")) {
+                        if (mem.eql(u8, sectname, "__text")) {
+                            self.text_section_index = index;
+                        }
+                    }
+
+                    sect.offset += offset_mod;
+                    if (sect.reloff > 0)
+                        sect.reloff += offset_mod;
+                }
+
+                seg.inner.fileoff += offset_mod;
+            },
+            macho.LC_SYMTAB => {
+                self.symtab_cmd_index = i;
+                cmd.Symtab.symoff += offset_mod;
+                cmd.Symtab.stroff += offset_mod;
+            },
+            macho.LC_DYSYMTAB => {
+                self.dysymtab_cmd_index = i;
+            },
+            macho.LC_BUILD_VERSION => {
+                self.build_version_cmd_index = i;
+            },
+            macho.LC_DATA_IN_CODE => {
+                self.data_in_code_cmd_index = i;
+                cmd.LinkeditData.dataoff += offset_mod;
+            },
+            else => {
+                log.debug("Unknown load command detected: 0x{x}.", .{cmd.cmd()});
+            },
+        }
+        self.load_commands.appendAssumeCapacity(cmd);
+    }
+}
+
+pub fn readSymtab(self: *Object) !void {
+    const symtab_cmd = self.load_commands.items[self.symtab_cmd_index.?].Symtab;
+    var buffer = try self.allocator.alloc(u8, @sizeOf(macho.nlist_64) * symtab_cmd.nsyms);
+    defer self.allocator.free(buffer);
+    _ = try self.file.preadAll(buffer, symtab_cmd.symoff);
+    try self.symtab.ensureCapacity(self.allocator, symtab_cmd.nsyms);
+    // TODO this align case should not be needed.
+    // Probably a bug in stage1.
+    const slice = @alignCast(@alignOf(macho.nlist_64), mem.bytesAsSlice(macho.nlist_64, buffer));
+    self.symtab.appendSliceAssumeCapacity(slice);
+}
+
+pub fn readStrtab(self: *Object) !void {
+    const symtab_cmd = self.load_commands.items[self.symtab_cmd_index.?].Symtab;
+    var buffer = try self.allocator.alloc(u8, symtab_cmd.strsize);
+    defer self.allocator.free(buffer);
+    _ = try self.file.preadAll(buffer, symtab_cmd.stroff);
+    try self.strtab.ensureCapacity(self.allocator, symtab_cmd.strsize);
+    self.strtab.appendSliceAssumeCapacity(buffer);
+}
+
+pub fn getString(self: *const Object, str_off: u32) []const u8 {
+    assert(str_off < self.strtab.items.len);
+    return mem.spanZ(@ptrCast([*:0]const u8, self.strtab.items.ptr + str_off));
+}
+
+pub fn readSection(self: Object, allocator: *Allocator, index: u16) ![]u8 {
+    const seg = self.load_commands.items[self.segment_cmd_index.?].Segment;
+    const sect = seg.sections.items[index];
+    var buffer = try allocator.alloc(u8, sect.size);
+    _ = try self.file.preadAll(buffer, sect.offset);
+    return buffer;
+}
+
+pub fn readDataInCode(self: *Object) !void {
+    const index = self.data_in_code_cmd_index orelse return;
+    const data_in_code = self.load_commands.items[index].LinkeditData;
+
+    var buffer = try self.allocator.alloc(u8, data_in_code.datasize);
+    defer self.allocator.free(buffer);
+
+    _ = try self.file.preadAll(buffer, data_in_code.dataoff);
+
+    var stream = io.fixedBufferStream(buffer);
+    var reader = stream.reader();
+    while (true) {
+        const dice = reader.readStruct(macho.data_in_code_entry) catch |err| switch (err) {
+            error.EndOfStream => break,
+            else => |e| return e,
+        };
+        try self.data_in_code_entries.append(self.allocator, dice);
+    }
+}
diff --git a/src/link/MachO/Zld.zig b/src/link/MachO/Zld.zig
new file mode 100644
index 0000000000..9340474d19
--- /dev/null
+++ b/src/link/MachO/Zld.zig
@@ -0,0 +1,3294 @@
+const Zld = @This();
+
+const std = @import("std");
+const assert = std.debug.assert;
+const dwarf = std.dwarf;
+const leb = std.leb;
+const mem = std.mem;
+const meta = std.meta;
+const fs = std.fs;
+const macho = std.macho;
+const math = std.math;
+const log = std.log.scoped(.zld);
+const aarch64 = @import("../../codegen/aarch64.zig");
+
+const Allocator = mem.Allocator;
+const CodeSignature = @import("CodeSignature.zig");
+const Archive = @import("Archive.zig");
+const Object = @import("Object.zig");
+const Trie = @import("Trie.zig");
+
+usingnamespace @import("commands.zig");
+usingnamespace @import("bind.zig");
+
+allocator: *Allocator,
+
+arch: ?std.Target.Cpu.Arch = null,
+page_size: ?u16 = null,
+file: ?fs.File = null,
+out_path: ?[]const u8 = null,
+
+// TODO Eventually, we will want to keep track of the  archives themselves to be able to exclude objects
+// contained within from landing in the final artifact. For now however, since we don't optimise the binary
+// at all, we just move all objects from the archives into the final artifact.
+objects: std.ArrayListUnmanaged(Object) = .{},
+
+load_commands: std.ArrayListUnmanaged(LoadCommand) = .{},
+
+pagezero_segment_cmd_index: ?u16 = null,
+text_segment_cmd_index: ?u16 = null,
+data_const_segment_cmd_index: ?u16 = null,
+data_segment_cmd_index: ?u16 = null,
+linkedit_segment_cmd_index: ?u16 = null,
+dyld_info_cmd_index: ?u16 = null,
+symtab_cmd_index: ?u16 = null,
+dysymtab_cmd_index: ?u16 = null,
+dylinker_cmd_index: ?u16 = null,
+libsystem_cmd_index: ?u16 = null,
+data_in_code_cmd_index: ?u16 = null,
+function_starts_cmd_index: ?u16 = null,
+main_cmd_index: ?u16 = null,
+version_min_cmd_index: ?u16 = null,
+source_version_cmd_index: ?u16 = null,
+uuid_cmd_index: ?u16 = null,
+code_signature_cmd_index: ?u16 = null,
+
+// __TEXT segment sections
+text_section_index: ?u16 = null,
+stubs_section_index: ?u16 = null,
+stub_helper_section_index: ?u16 = null,
+text_const_section_index: ?u16 = null,
+cstring_section_index: ?u16 = null,
+
+// __DATA_CONST segment sections
+got_section_index: ?u16 = null,
+mod_init_func_section_index: ?u16 = null,
+mod_term_func_section_index: ?u16 = null,
+data_const_section_index: ?u16 = null,
+
+// __DATA segment sections
+tlv_section_index: ?u16 = null,
+tlv_data_section_index: ?u16 = null,
+tlv_bss_section_index: ?u16 = null,
+la_symbol_ptr_section_index: ?u16 = null,
+data_section_index: ?u16 = null,
+bss_section_index: ?u16 = null,
+
+locals: std.StringArrayHashMapUnmanaged(std.ArrayListUnmanaged(Symbol)) = .{},
+exports: std.StringArrayHashMapUnmanaged(macho.nlist_64) = .{},
+nonlazy_imports: std.StringArrayHashMapUnmanaged(Import) = .{},
+lazy_imports: std.StringArrayHashMapUnmanaged(Import) = .{},
+tlv_bootstrap: ?Import = null,
+threadlocal_offsets: std.ArrayListUnmanaged(u64) = .{},
+local_rebases: std.ArrayListUnmanaged(Pointer) = .{},
+nonlazy_pointers: std.StringArrayHashMapUnmanaged(GotEntry) = .{},
+
+strtab: std.ArrayListUnmanaged(u8) = .{},
+
+stub_helper_stubs_start_off: ?u64 = null,
+
+mappings: std.AutoHashMapUnmanaged(MappingKey, SectionMapping) = .{},
+unhandled_sections: std.AutoHashMapUnmanaged(MappingKey, u0) = .{},
+
+// TODO this will require scanning the relocations at least one to work out
+// the exact amount of local GOT indirections. For the time being, set some
+// default value.
+const max_local_got_indirections: u16 = 1000;
+
+const GotEntry = struct {
+    index: u32,
+    target_addr: u64,
+};
+
+const MappingKey = struct {
+    object_id: u16,
+    source_sect_id: u16,
+};
+
+const SectionMapping = struct {
+    source_sect_id: u16,
+    target_seg_id: u16,
+    target_sect_id: u16,
+    offset: u32,
+};
+
+const Symbol = struct {
+    inner: macho.nlist_64,
+    tt: Type,
+    object_id: u16,
+
+    const Type = enum {
+        Local,
+        WeakGlobal,
+        Global,
+    };
+};
+
+const DebugInfo = struct {
+    inner: dwarf.DwarfInfo,
+    debug_info: []u8,
+    debug_abbrev: []u8,
+    debug_str: []u8,
+    debug_line: []u8,
+    debug_ranges: []u8,
+
+    pub fn parseFromObject(allocator: *Allocator, object: Object) !?DebugInfo {
+        var debug_info = blk: {
+            const index = object.dwarf_debug_info_index orelse return null;
+            break :blk try object.readSection(allocator, index);
+        };
+        var debug_abbrev = blk: {
+            const index = object.dwarf_debug_abbrev_index orelse return null;
+            break :blk try object.readSection(allocator, index);
+        };
+        var debug_str = blk: {
+            const index = object.dwarf_debug_str_index orelse return null;
+            break :blk try object.readSection(allocator, index);
+        };
+        var debug_line = blk: {
+            const index = object.dwarf_debug_line_index orelse return null;
+            break :blk try object.readSection(allocator, index);
+        };
+        var debug_ranges = blk: {
+            if (object.dwarf_debug_ranges_index) |ind| {
+                break :blk try object.readSection(allocator, ind);
+            }
+            break :blk try allocator.alloc(u8, 0);
+        };
+
+        var inner: dwarf.DwarfInfo = .{
+            .endian = .Little,
+            .debug_info = debug_info,
+            .debug_abbrev = debug_abbrev,
+            .debug_str = debug_str,
+            .debug_line = debug_line,
+            .debug_ranges = debug_ranges,
+        };
+        try dwarf.openDwarfDebugInfo(&inner, allocator);
+
+        return DebugInfo{
+            .inner = inner,
+            .debug_info = debug_info,
+            .debug_abbrev = debug_abbrev,
+            .debug_str = debug_str,
+            .debug_line = debug_line,
+            .debug_ranges = debug_ranges,
+        };
+    }
+
+    pub fn deinit(self: *DebugInfo, allocator: *Allocator) void {
+        allocator.free(self.debug_info);
+        allocator.free(self.debug_abbrev);
+        allocator.free(self.debug_str);
+        allocator.free(self.debug_line);
+        allocator.free(self.debug_ranges);
+        self.inner.abbrev_table_list.deinit();
+        self.inner.compile_unit_list.deinit();
+        self.inner.func_list.deinit();
+    }
+};
+
+pub const Import = struct {
+    /// MachO symbol table entry.
+    symbol: macho.nlist_64,
+
+    /// Id of the dynamic library where the specified entries can be found.
+    dylib_ordinal: i64,
+
+    /// Index of this import within the import list.
+    index: u32,
+};
+
+/// Default path to dyld
+/// TODO instead of hardcoding it, we should probably look through some env vars and search paths
+/// instead but this will do for now.
+const DEFAULT_DYLD_PATH: [*:0]const u8 = "/usr/lib/dyld";
+
+/// Default lib search path
+/// TODO instead of hardcoding it, we should probably look through some env vars and search paths
+/// instead but this will do for now.
+const DEFAULT_LIB_SEARCH_PATH: []const u8 = "/usr/lib";
+
+const LIB_SYSTEM_NAME: [*:0]const u8 = "System";
+/// TODO we should search for libSystem and fail if it doesn't exist, instead of hardcoding it
+const LIB_SYSTEM_PATH: [*:0]const u8 = DEFAULT_LIB_SEARCH_PATH ++ "/libSystem.B.dylib";
+
+pub fn init(allocator: *Allocator) Zld {
+    return .{ .allocator = allocator };
+}
+
+pub fn deinit(self: *Zld) void {
+    self.threadlocal_offsets.deinit(self.allocator);
+    self.strtab.deinit(self.allocator);
+    self.local_rebases.deinit(self.allocator);
+    for (self.lazy_imports.items()) |*entry| {
+        self.allocator.free(entry.key);
+    }
+    self.lazy_imports.deinit(self.allocator);
+    for (self.nonlazy_imports.items()) |*entry| {
+        self.allocator.free(entry.key);
+    }
+    self.nonlazy_imports.deinit(self.allocator);
+    for (self.nonlazy_pointers.items()) |*entry| {
+        self.allocator.free(entry.key);
+    }
+    self.nonlazy_pointers.deinit(self.allocator);
+    for (self.exports.items()) |*entry| {
+        self.allocator.free(entry.key);
+    }
+    self.exports.deinit(self.allocator);
+    for (self.locals.items()) |*entry| {
+        self.allocator.free(entry.key);
+        entry.value.deinit(self.allocator);
+    }
+    self.locals.deinit(self.allocator);
+    for (self.objects.items) |*object| {
+        object.deinit();
+    }
+    self.objects.deinit(self.allocator);
+    for (self.load_commands.items) |*lc| {
+        lc.deinit(self.allocator);
+    }
+    self.load_commands.deinit(self.allocator);
+    self.mappings.deinit(self.allocator);
+    self.unhandled_sections.deinit(self.allocator);
+    if (self.file) |*f| f.close();
+}
+
+pub fn link(self: *Zld, files: []const []const u8, out_path: []const u8) !void {
+    if (files.len == 0) return error.NoInputFiles;
+    if (out_path.len == 0) return error.EmptyOutputPath;
+
+    if (self.arch == null) {
+        // Try inferring the arch from the object files.
+        self.arch = blk: {
+            const file = try fs.cwd().openFile(files[0], .{});
+            defer file.close();
+            var reader = file.reader();
+            const header = try reader.readStruct(macho.mach_header_64);
+            const arch: std.Target.Cpu.Arch = switch (header.cputype) {
+                macho.CPU_TYPE_X86_64 => .x86_64,
+                macho.CPU_TYPE_ARM64 => .aarch64,
+                else => |value| {
+                    log.err("unsupported cpu architecture 0x{x}", .{value});
+                    return error.UnsupportedCpuArchitecture;
+                },
+            };
+            break :blk arch;
+        };
+    }
+
+    self.page_size = switch (self.arch.?) {
+        .aarch64 => 0x4000,
+        .x86_64 => 0x1000,
+        else => unreachable,
+    };
+    self.out_path = out_path;
+    self.file = try fs.cwd().createFile(out_path, .{
+        .truncate = true,
+        .read = true,
+        .mode = if (std.Target.current.os.tag == .windows) 0 else 0o777,
+    });
+
+    try self.populateMetadata();
+    try self.parseInputFiles(files);
+    try self.sortSections();
+    try self.resolveImports();
+    try self.allocateTextSegment();
+    try self.allocateDataConstSegment();
+    try self.allocateDataSegment();
+    self.allocateLinkeditSegment();
+    try self.writeStubHelperCommon();
+    try self.resolveSymbols();
+    try self.doRelocs();
+    try self.flush();
+}
+
+fn parseInputFiles(self: *Zld, files: []const []const u8) !void {
+    for (files) |file_name| {
+        const file = try fs.cwd().openFile(file_name, .{});
+
+        try_object: {
+            var object = Object.initFromFile(self.allocator, self.arch.?, file_name, file) catch |err| switch (err) {
+                error.NotObject => break :try_object,
+                else => |e| return e,
+            };
+            const index = @intCast(u16, self.objects.items.len);
+            try self.objects.append(self.allocator, object);
+            try self.updateMetadata(index);
+            continue;
+        }
+
+        try_archive: {
+            var archive = Archive.initFromFile(self.allocator, self.arch.?, file_name, file) catch |err| switch (err) {
+                error.NotArchive => break :try_archive,
+                else => |e| return e,
+            };
+            defer archive.deinit();
+            while (archive.objects.popOrNull()) |object| {
+                const index = @intCast(u16, self.objects.items.len);
+                try self.objects.append(self.allocator, object);
+                try self.updateMetadata(index);
+            }
+            continue;
+        }
+
+        log.err("unexpected file type: expected object '.o' or archive '.a': {s}", .{file_name});
+        return error.UnexpectedInputFileType;
+    }
+}
+
+fn mapAndUpdateSections(
+    self: *Zld,
+    object_id: u16,
+    source_sect_id: u16,
+    target_seg_id: u16,
+    target_sect_id: u16,
+) !void {
+    const object = self.objects.items[object_id];
+    const source_seg = object.load_commands.items[object.segment_cmd_index.?].Segment;
+    const source_sect = source_seg.sections.items[source_sect_id];
+    const target_seg = &self.load_commands.items[target_seg_id].Segment;
+    const target_sect = &target_seg.sections.items[target_sect_id];
+
+    const alignment = try math.powi(u32, 2, target_sect.@"align");
+    const offset = mem.alignForwardGeneric(u64, target_sect.size, alignment);
+    const size = mem.alignForwardGeneric(u64, source_sect.size, alignment);
+    const key = MappingKey{
+        .object_id = object_id,
+        .source_sect_id = source_sect_id,
+    };
+    try self.mappings.putNoClobber(self.allocator, key, .{
+        .source_sect_id = source_sect_id,
+        .target_seg_id = target_seg_id,
+        .target_sect_id = target_sect_id,
+        .offset = @intCast(u32, offset),
+    });
+    log.debug("{s}: {s},{s} mapped to {s},{s} from 0x{x} to 0x{x}", .{
+        object.name,
+        parseName(&source_sect.segname),
+        parseName(&source_sect.sectname),
+        parseName(&target_sect.segname),
+        parseName(&target_sect.sectname),
+        offset,
+        offset + size,
+    });
+
+    target_sect.size = offset + size;
+}
+
+fn updateMetadata(self: *Zld, object_id: u16) !void {
+    const object = self.objects.items[object_id];
+    const object_seg = object.load_commands.items[object.segment_cmd_index.?].Segment;
+    const text_seg = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    const data_const_seg = &self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+    const data_seg = &self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+
+    // Create missing metadata
+    for (object_seg.sections.items) |source_sect, id| {
+        if (id == object.text_section_index.?) continue;
+        const segname = parseName(&source_sect.segname);
+        const sectname = parseName(&source_sect.sectname);
+        const flags = source_sect.flags;
+
+        switch (flags) {
+            macho.S_REGULAR, macho.S_4BYTE_LITERALS, macho.S_8BYTE_LITERALS, macho.S_16BYTE_LITERALS => {
+                if (mem.eql(u8, segname, "__TEXT")) {
+                    if (self.text_const_section_index != null) continue;
+
+                    self.text_const_section_index = @intCast(u16, text_seg.sections.items.len);
+                    try text_seg.addSection(self.allocator, .{
+                        .sectname = makeStaticString("__const"),
+                        .segname = makeStaticString("__TEXT"),
+                        .addr = 0,
+                        .size = 0,
+                        .offset = 0,
+                        .@"align" = 0,
+                        .reloff = 0,
+                        .nreloc = 0,
+                        .flags = macho.S_REGULAR,
+                        .reserved1 = 0,
+                        .reserved2 = 0,
+                        .reserved3 = 0,
+                    });
+                } else if (mem.eql(u8, segname, "__DATA")) {
+                    if (!mem.eql(u8, sectname, "__const")) continue;
+                    if (self.data_const_section_index != null) continue;
+
+                    self.data_const_section_index = @intCast(u16, data_const_seg.sections.items.len);
+                    try data_const_seg.addSection(self.allocator, .{
+                        .sectname = makeStaticString("__const"),
+                        .segname = makeStaticString("__DATA_CONST"),
+                        .addr = 0,
+                        .size = 0,
+                        .offset = 0,
+                        .@"align" = 0,
+                        .reloff = 0,
+                        .nreloc = 0,
+                        .flags = macho.S_REGULAR,
+                        .reserved1 = 0,
+                        .reserved2 = 0,
+                        .reserved3 = 0,
+                    });
+                }
+            },
+            macho.S_CSTRING_LITERALS => {
+                if (!mem.eql(u8, segname, "__TEXT")) continue;
+                if (self.cstring_section_index != null) continue;
+
+                self.cstring_section_index = @intCast(u16, text_seg.sections.items.len);
+                try text_seg.addSection(self.allocator, .{
+                    .sectname = makeStaticString("__cstring"),
+                    .segname = makeStaticString("__TEXT"),
+                    .addr = 0,
+                    .size = 0,
+                    .offset = 0,
+                    .@"align" = 0,
+                    .reloff = 0,
+                    .nreloc = 0,
+                    .flags = macho.S_CSTRING_LITERALS,
+                    .reserved1 = 0,
+                    .reserved2 = 0,
+                    .reserved3 = 0,
+                });
+            },
+            macho.S_MOD_INIT_FUNC_POINTERS => {
+                if (!mem.eql(u8, segname, "__DATA")) continue;
+                if (self.mod_init_func_section_index != null) continue;
+
+                self.mod_init_func_section_index = @intCast(u16, data_const_seg.sections.items.len);
+                try data_const_seg.addSection(self.allocator, .{
+                    .sectname = makeStaticString("__mod_init_func"),
+                    .segname = makeStaticString("__DATA_CONST"),
+                    .addr = 0,
+                    .size = 0,
+                    .offset = 0,
+                    .@"align" = 0,
+                    .reloff = 0,
+                    .nreloc = 0,
+                    .flags = macho.S_MOD_INIT_FUNC_POINTERS,
+                    .reserved1 = 0,
+                    .reserved2 = 0,
+                    .reserved3 = 0,
+                });
+            },
+            macho.S_MOD_TERM_FUNC_POINTERS => {
+                if (!mem.eql(u8, segname, "__DATA")) continue;
+                if (self.mod_term_func_section_index != null) continue;
+
+                self.mod_term_func_section_index = @intCast(u16, data_const_seg.sections.items.len);
+                try data_const_seg.addSection(self.allocator, .{
+                    .sectname = makeStaticString("__mod_term_func"),
+                    .segname = makeStaticString("__DATA_CONST"),
+                    .addr = 0,
+                    .size = 0,
+                    .offset = 0,
+                    .@"align" = 0,
+                    .reloff = 0,
+                    .nreloc = 0,
+                    .flags = macho.S_MOD_TERM_FUNC_POINTERS,
+                    .reserved1 = 0,
+                    .reserved2 = 0,
+                    .reserved3 = 0,
+                });
+            },
+            macho.S_ZEROFILL => {
+                if (!mem.eql(u8, segname, "__DATA")) continue;
+                if (self.bss_section_index != null) continue;
+
+                self.bss_section_index = @intCast(u16, data_seg.sections.items.len);
+                try data_seg.addSection(self.allocator, .{
+                    .sectname = makeStaticString("__bss"),
+                    .segname = makeStaticString("__DATA"),
+                    .addr = 0,
+                    .size = 0,
+                    .offset = 0,
+                    .@"align" = 0,
+                    .reloff = 0,
+                    .nreloc = 0,
+                    .flags = macho.S_ZEROFILL,
+                    .reserved1 = 0,
+                    .reserved2 = 0,
+                    .reserved3 = 0,
+                });
+            },
+            macho.S_THREAD_LOCAL_VARIABLES => {
+                if (!mem.eql(u8, segname, "__DATA")) continue;
+                if (self.tlv_section_index != null) continue;
+
+                self.tlv_section_index = @intCast(u16, data_seg.sections.items.len);
+                try data_seg.addSection(self.allocator, .{
+                    .sectname = makeStaticString("__thread_vars"),
+                    .segname = makeStaticString("__DATA"),
+                    .addr = 0,
+                    .size = 0,
+                    .offset = 0,
+                    .@"align" = 0,
+                    .reloff = 0,
+                    .nreloc = 0,
+                    .flags = macho.S_THREAD_LOCAL_VARIABLES,
+                    .reserved1 = 0,
+                    .reserved2 = 0,
+                    .reserved3 = 0,
+                });
+            },
+            macho.S_THREAD_LOCAL_REGULAR => {
+                if (!mem.eql(u8, segname, "__DATA")) continue;
+                if (self.tlv_data_section_index != null) continue;
+
+                self.tlv_data_section_index = @intCast(u16, data_seg.sections.items.len);
+                try data_seg.addSection(self.allocator, .{
+                    .sectname = makeStaticString("__thread_data"),
+                    .segname = makeStaticString("__DATA"),
+                    .addr = 0,
+                    .size = 0,
+                    .offset = 0,
+                    .@"align" = 0,
+                    .reloff = 0,
+                    .nreloc = 0,
+                    .flags = macho.S_THREAD_LOCAL_REGULAR,
+                    .reserved1 = 0,
+                    .reserved2 = 0,
+                    .reserved3 = 0,
+                });
+            },
+            macho.S_THREAD_LOCAL_ZEROFILL => {
+                if (!mem.eql(u8, segname, "__DATA")) continue;
+                if (self.tlv_bss_section_index != null) continue;
+
+                self.tlv_bss_section_index = @intCast(u16, data_seg.sections.items.len);
+                try data_seg.addSection(self.allocator, .{
+                    .sectname = makeStaticString("__thread_bss"),
+                    .segname = makeStaticString("__DATA"),
+                    .addr = 0,
+                    .size = 0,
+                    .offset = 0,
+                    .@"align" = 0,
+                    .reloff = 0,
+                    .nreloc = 0,
+                    .flags = macho.S_THREAD_LOCAL_ZEROFILL,
+                    .reserved1 = 0,
+                    .reserved2 = 0,
+                    .reserved3 = 0,
+                });
+            },
+            else => {
+                log.debug("unhandled section type 0x{x} for '{s}/{s}'", .{ flags, segname, sectname });
+            },
+        }
+    }
+
+    // Find ideal section alignment.
+    for (object_seg.sections.items) |source_sect, id| {
+        if (self.getMatchingSection(source_sect)) |res| {
+            const target_seg = &self.load_commands.items[res.seg].Segment;
+            const target_sect = &target_seg.sections.items[res.sect];
+            target_sect.@"align" = math.max(target_sect.@"align", source_sect.@"align");
+        }
+    }
+
+    // Update section mappings
+    for (object_seg.sections.items) |source_sect, id| {
+        const source_sect_id = @intCast(u16, id);
+        if (self.getMatchingSection(source_sect)) |res| {
+            try self.mapAndUpdateSections(object_id, source_sect_id, res.seg, res.sect);
+            continue;
+        }
+
+        const segname = parseName(&source_sect.segname);
+        const sectname = parseName(&source_sect.sectname);
+        log.debug("section '{s}/{s}' will be unmapped", .{ segname, sectname });
+        try self.unhandled_sections.putNoClobber(self.allocator, .{
+            .object_id = object_id,
+            .source_sect_id = source_sect_id,
+        }, 0);
+    }
+}
+
+const MatchingSection = struct {
+    seg: u16,
+    sect: u16,
+};
+
+fn getMatchingSection(self: *Zld, section: macho.section_64) ?MatchingSection {
+    const segname = parseName(&section.segname);
+    const sectname = parseName(&section.sectname);
+    const res: ?MatchingSection = blk: {
+        switch (section.flags) {
+            macho.S_4BYTE_LITERALS, macho.S_8BYTE_LITERALS, macho.S_16BYTE_LITERALS => {
+                break :blk .{
+                    .seg = self.text_segment_cmd_index.?,
+                    .sect = self.text_const_section_index.?,
+                };
+            },
+            macho.S_CSTRING_LITERALS => {
+                break :blk .{
+                    .seg = self.text_segment_cmd_index.?,
+                    .sect = self.cstring_section_index.?,
+                };
+            },
+            macho.S_MOD_INIT_FUNC_POINTERS => {
+                break :blk .{
+                    .seg = self.data_const_segment_cmd_index.?,
+                    .sect = self.mod_init_func_section_index.?,
+                };
+            },
+            macho.S_MOD_TERM_FUNC_POINTERS => {
+                break :blk .{
+                    .seg = self.data_const_segment_cmd_index.?,
+                    .sect = self.mod_term_func_section_index.?,
+                };
+            },
+            macho.S_ZEROFILL => {
+                break :blk .{
+                    .seg = self.data_segment_cmd_index.?,
+                    .sect = self.bss_section_index.?,
+                };
+            },
+            macho.S_THREAD_LOCAL_VARIABLES => {
+                break :blk .{
+                    .seg = self.data_segment_cmd_index.?,
+                    .sect = self.tlv_section_index.?,
+                };
+            },
+            macho.S_THREAD_LOCAL_REGULAR => {
+                break :blk .{
+                    .seg = self.data_segment_cmd_index.?,
+                    .sect = self.tlv_data_section_index.?,
+                };
+            },
+            macho.S_THREAD_LOCAL_ZEROFILL => {
+                break :blk .{
+                    .seg = self.data_segment_cmd_index.?,
+                    .sect = self.tlv_bss_section_index.?,
+                };
+            },
+            macho.S_REGULAR | macho.S_ATTR_PURE_INSTRUCTIONS | macho.S_ATTR_SOME_INSTRUCTIONS => {
+                break :blk .{
+                    .seg = self.text_segment_cmd_index.?,
+                    .sect = self.text_section_index.?,
+                };
+            },
+            macho.S_REGULAR => {
+                if (mem.eql(u8, segname, "__TEXT")) {
+                    break :blk .{
+                        .seg = self.text_segment_cmd_index.?,
+                        .sect = self.text_const_section_index.?,
+                    };
+                } else if (mem.eql(u8, segname, "__DATA")) {
+                    if (mem.eql(u8, sectname, "__data")) {
+                        break :blk .{
+                            .seg = self.data_segment_cmd_index.?,
+                            .sect = self.data_section_index.?,
+                        };
+                    } else if (mem.eql(u8, sectname, "__const")) {
+                        break :blk .{
+                            .seg = self.data_const_segment_cmd_index.?,
+                            .sect = self.data_const_section_index.?,
+                        };
+                    }
+                }
+                break :blk null;
+            },
+            else => {
+                break :blk null;
+            },
+        }
+    };
+    return res;
+}
+
+fn sortSections(self: *Zld) !void {
+    var text_index_mapping = std.AutoHashMap(u16, u16).init(self.allocator);
+    defer text_index_mapping.deinit();
+    var data_const_index_mapping = std.AutoHashMap(u16, u16).init(self.allocator);
+    defer data_const_index_mapping.deinit();
+    var data_index_mapping = std.AutoHashMap(u16, u16).init(self.allocator);
+    defer data_index_mapping.deinit();
+
+    {
+        // __TEXT segment
+        const seg = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+        var sections = seg.sections.toOwnedSlice(self.allocator);
+        defer self.allocator.free(sections);
+        try seg.sections.ensureCapacity(self.allocator, sections.len);
+
+        const indices = &[_]*?u16{
+            &self.text_section_index,
+            &self.stubs_section_index,
+            &self.stub_helper_section_index,
+            &self.text_const_section_index,
+            &self.cstring_section_index,
+        };
+        for (indices) |maybe_index| {
+            const new_index: u16 = if (maybe_index.*) |index| blk: {
+                const idx = @intCast(u16, seg.sections.items.len);
+                seg.sections.appendAssumeCapacity(sections[index]);
+                try text_index_mapping.putNoClobber(index, idx);
+                break :blk idx;
+            } else continue;
+            maybe_index.* = new_index;
+        }
+    }
+
+    {
+        // __DATA_CONST segment
+        const seg = &self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+        var sections = seg.sections.toOwnedSlice(self.allocator);
+        defer self.allocator.free(sections);
+        try seg.sections.ensureCapacity(self.allocator, sections.len);
+
+        const indices = &[_]*?u16{
+            &self.got_section_index,
+            &self.mod_init_func_section_index,
+            &self.mod_term_func_section_index,
+            &self.data_const_section_index,
+        };
+        for (indices) |maybe_index| {
+            const new_index: u16 = if (maybe_index.*) |index| blk: {
+                const idx = @intCast(u16, seg.sections.items.len);
+                seg.sections.appendAssumeCapacity(sections[index]);
+                try data_const_index_mapping.putNoClobber(index, idx);
+                break :blk idx;
+            } else continue;
+            maybe_index.* = new_index;
+        }
+    }
+
+    {
+        // __DATA segment
+        const seg = &self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+        var sections = seg.sections.toOwnedSlice(self.allocator);
+        defer self.allocator.free(sections);
+        try seg.sections.ensureCapacity(self.allocator, sections.len);
+
+        // __DATA segment
+        const indices = &[_]*?u16{
+            &self.la_symbol_ptr_section_index,
+            &self.tlv_section_index,
+            &self.data_section_index,
+            &self.tlv_data_section_index,
+            &self.tlv_bss_section_index,
+            &self.bss_section_index,
+        };
+        for (indices) |maybe_index| {
+            const new_index: u16 = if (maybe_index.*) |index| blk: {
+                const idx = @intCast(u16, seg.sections.items.len);
+                seg.sections.appendAssumeCapacity(sections[index]);
+                try data_index_mapping.putNoClobber(index, idx);
+                break :blk idx;
+            } else continue;
+            maybe_index.* = new_index;
+        }
+    }
+
+    var it = self.mappings.iterator();
+    while (it.next()) |entry| {
+        const mapping = &entry.value;
+        if (self.text_segment_cmd_index.? == mapping.target_seg_id) {
+            const new_index = text_index_mapping.get(mapping.target_sect_id) orelse unreachable;
+            mapping.target_sect_id = new_index;
+        } else if (self.data_const_segment_cmd_index.? == mapping.target_seg_id) {
+            const new_index = data_const_index_mapping.get(mapping.target_sect_id) orelse unreachable;
+            mapping.target_sect_id = new_index;
+        } else if (self.data_segment_cmd_index.? == mapping.target_seg_id) {
+            const new_index = data_index_mapping.get(mapping.target_sect_id) orelse unreachable;
+            mapping.target_sect_id = new_index;
+        } else unreachable;
+    }
+}
+
+fn resolveImports(self: *Zld) !void {
+    var imports = std.StringArrayHashMap(bool).init(self.allocator);
+    defer imports.deinit();
+
+    for (self.objects.items) |object| {
+        for (object.symtab.items) |sym| {
+            if (isLocal(&sym)) continue;
+
+            const name = object.getString(sym.n_strx);
+            const res = try imports.getOrPut(name);
+            if (isExport(&sym)) {
+                res.entry.value = false;
+                continue;
+            }
+            if (res.found_existing and !res.entry.value)
+                continue;
+            res.entry.value = true;
+        }
+    }
+
+    for (imports.items()) |entry| {
+        if (!entry.value) continue;
+
+        const sym_name = entry.key;
+        const n_strx = try self.makeString(sym_name);
+        var new_sym: macho.nlist_64 = .{
+            .n_strx = n_strx,
+            .n_type = macho.N_UNDF | macho.N_EXT,
+            .n_value = 0,
+            .n_desc = macho.REFERENCE_FLAG_UNDEFINED_NON_LAZY | macho.N_SYMBOL_RESOLVER,
+            .n_sect = 0,
+        };
+        var key = try self.allocator.dupe(u8, sym_name);
+        // TODO handle symbol resolution from non-libc dylibs.
+        const dylib_ordinal = 1;
+
+        // TODO need to rework this. Perhaps should create a set of all possible libc
+        // symbols which are expected to be nonlazy?
+        if (mem.eql(u8, sym_name, "___stdoutp") or
+            mem.eql(u8, sym_name, "___stderrp") or
+            mem.eql(u8, sym_name, "___stdinp") or
+            mem.eql(u8, sym_name, "___stack_chk_guard") or
+            mem.eql(u8, sym_name, "_environ") or
+            mem.eql(u8, sym_name, "__DefaultRuneLocale") or
+            mem.eql(u8, sym_name, "_mach_task_self_"))
+        {
+            log.debug("writing nonlazy symbol '{s}'", .{sym_name});
+            const index = @intCast(u32, self.nonlazy_imports.items().len);
+            try self.nonlazy_imports.putNoClobber(self.allocator, key, .{
+                .symbol = new_sym,
+                .dylib_ordinal = dylib_ordinal,
+                .index = index,
+            });
+        } else if (mem.eql(u8, sym_name, "__tlv_bootstrap")) {
+            log.debug("writing threadlocal symbol '{s}'", .{sym_name});
+            self.tlv_bootstrap = .{
+                .symbol = new_sym,
+                .dylib_ordinal = dylib_ordinal,
+                .index = 0,
+            };
+        } else {
+            log.debug("writing lazy symbol '{s}'", .{sym_name});
+            const index = @intCast(u32, self.lazy_imports.items().len);
+            try self.lazy_imports.putNoClobber(self.allocator, key, .{
+                .symbol = new_sym,
+                .dylib_ordinal = dylib_ordinal,
+                .index = index,
+            });
+        }
+    }
+
+    const n_strx = try self.makeString("dyld_stub_binder");
+    const name = try self.allocator.dupe(u8, "dyld_stub_binder");
+    log.debug("writing nonlazy symbol 'dyld_stub_binder'", .{});
+    const index = @intCast(u32, self.nonlazy_imports.items().len);
+    try self.nonlazy_imports.putNoClobber(self.allocator, name, .{
+        .symbol = .{
+            .n_strx = n_strx,
+            .n_type = std.macho.N_UNDF | std.macho.N_EXT,
+            .n_sect = 0,
+            .n_desc = std.macho.REFERENCE_FLAG_UNDEFINED_NON_LAZY | std.macho.N_SYMBOL_RESOLVER,
+            .n_value = 0,
+        },
+        .dylib_ordinal = 1,
+        .index = index,
+    });
+}
+
+fn allocateTextSegment(self: *Zld) !void {
+    const seg = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    const nexterns = @intCast(u32, self.lazy_imports.items().len);
+
+    const base_vmaddr = self.load_commands.items[self.pagezero_segment_cmd_index.?].Segment.inner.vmsize;
+    seg.inner.fileoff = 0;
+    seg.inner.vmaddr = base_vmaddr;
+
+    // Set stubs and stub_helper sizes
+    const stubs = &seg.sections.items[self.stubs_section_index.?];
+    const stub_helper = &seg.sections.items[self.stub_helper_section_index.?];
+    stubs.size += nexterns * stubs.reserved2;
+
+    const stub_size: u4 = switch (self.arch.?) {
+        .x86_64 => 10,
+        .aarch64 => 3 * @sizeOf(u32),
+        else => unreachable,
+    };
+    stub_helper.size += nexterns * stub_size;
+
+    var sizeofcmds: u64 = 0;
+    for (self.load_commands.items) |lc| {
+        sizeofcmds += lc.cmdsize();
+    }
+
+    try self.allocateSegment(self.text_segment_cmd_index.?, @sizeOf(macho.mach_header_64) + sizeofcmds);
+
+    // Shift all sections to the back to minimize jump size between __TEXT and __DATA segments.
+    var min_alignment: u32 = 0;
+    for (seg.sections.items) |sect| {
+        const alignment = try math.powi(u32, 2, sect.@"align");
+        min_alignment = math.max(min_alignment, alignment);
+    }
+
+    assert(min_alignment > 0);
+    const last_sect_idx = seg.sections.items.len - 1;
+    const last_sect = seg.sections.items[last_sect_idx];
+    const shift: u32 = blk: {
+        const diff = seg.inner.filesize - last_sect.offset - last_sect.size;
+        const factor = @divTrunc(diff, min_alignment);
+        break :blk @intCast(u32, factor * min_alignment);
+    };
+
+    if (shift > 0) {
+        for (seg.sections.items) |*sect| {
+            sect.offset += shift;
+            sect.addr += shift;
+        }
+    }
+}
+
+fn allocateDataConstSegment(self: *Zld) !void {
+    const seg = &self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+    const nonlazy = @intCast(u32, self.nonlazy_imports.items().len);
+
+    const text_seg = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    seg.inner.fileoff = text_seg.inner.fileoff + text_seg.inner.filesize;
+    seg.inner.vmaddr = text_seg.inner.vmaddr + text_seg.inner.vmsize;
+
+    // Set got size
+    const got = &seg.sections.items[self.got_section_index.?];
+    // TODO this will require scanning the relocations at least one to work out
+    // the exact amount of local GOT indirections. For the time being, set some
+    // default value.
+    got.size += (max_local_got_indirections + nonlazy) * @sizeOf(u64);
+
+    try self.allocateSegment(self.data_const_segment_cmd_index.?, 0);
+}
+
+fn allocateDataSegment(self: *Zld) !void {
+    const seg = &self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+    const lazy = @intCast(u32, self.lazy_imports.items().len);
+
+    const data_const_seg = self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+    seg.inner.fileoff = data_const_seg.inner.fileoff + data_const_seg.inner.filesize;
+    seg.inner.vmaddr = data_const_seg.inner.vmaddr + data_const_seg.inner.vmsize;
+
+    // Set la_symbol_ptr and data size
+    const la_symbol_ptr = &seg.sections.items[self.la_symbol_ptr_section_index.?];
+    const data = &seg.sections.items[self.data_section_index.?];
+    la_symbol_ptr.size += lazy * @sizeOf(u64);
+    data.size += @sizeOf(u64); // TODO when do we need more?
+
+    try self.allocateSegment(self.data_segment_cmd_index.?, 0);
+}
+
+fn allocateLinkeditSegment(self: *Zld) void {
+    const seg = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+    const data_seg = self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+    seg.inner.fileoff = data_seg.inner.fileoff + data_seg.inner.filesize;
+    seg.inner.vmaddr = data_seg.inner.vmaddr + data_seg.inner.vmsize;
+}
+
+fn allocateSegment(self: *Zld, index: u16, offset: u64) !void {
+    const base_vmaddr = self.load_commands.items[self.pagezero_segment_cmd_index.?].Segment.inner.vmsize;
+    const seg = &self.load_commands.items[index].Segment;
+
+    // Allocate the sections according to their alignment at the beginning of the segment.
+    var start: u64 = offset;
+    for (seg.sections.items) |*sect| {
+        const alignment = try math.powi(u32, 2, sect.@"align");
+        const start_aligned = mem.alignForwardGeneric(u64, start, alignment);
+        const end_aligned = mem.alignForwardGeneric(u64, start_aligned + sect.size, alignment);
+        sect.offset = @intCast(u32, seg.inner.fileoff + start_aligned);
+        sect.addr = seg.inner.vmaddr + start_aligned;
+        start = end_aligned;
+    }
+
+    const seg_size_aligned = mem.alignForwardGeneric(u64, start, self.page_size.?);
+    seg.inner.filesize = seg_size_aligned;
+    seg.inner.vmsize = seg_size_aligned;
+}
+
+fn writeStubHelperCommon(self: *Zld) !void {
+    const text_segment = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    const stub_helper = &text_segment.sections.items[self.stub_helper_section_index.?];
+    const data_const_segment = &self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+    const got = &data_const_segment.sections.items[self.got_section_index.?];
+    const data_segment = &self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+    const data = &data_segment.sections.items[self.data_section_index.?];
+    const la_symbol_ptr = data_segment.sections.items[self.la_symbol_ptr_section_index.?];
+
+    self.stub_helper_stubs_start_off = blk: {
+        switch (self.arch.?) {
+            .x86_64 => {
+                const code_size = 15;
+                var code: [code_size]u8 = undefined;
+                // lea %r11, [rip + disp]
+                code[0] = 0x4c;
+                code[1] = 0x8d;
+                code[2] = 0x1d;
+                {
+                    const target_addr = data.addr + data.size - @sizeOf(u64);
+                    const displacement = try math.cast(u32, target_addr - stub_helper.addr - 7);
+                    mem.writeIntLittle(u32, code[3..7], displacement);
+                }
+                // push %r11
+                code[7] = 0x41;
+                code[8] = 0x53;
+                // jmp [rip + disp]
+                code[9] = 0xff;
+                code[10] = 0x25;
+                {
+                    const dyld_stub_binder = self.nonlazy_imports.get("dyld_stub_binder").?;
+                    const addr = (got.addr + dyld_stub_binder.index * @sizeOf(u64));
+                    const displacement = try math.cast(u32, addr - stub_helper.addr - code_size);
+                    mem.writeIntLittle(u32, code[11..], displacement);
+                }
+                try self.file.?.pwriteAll(&code, stub_helper.offset);
+                break :blk stub_helper.offset + code_size;
+            },
+            .aarch64 => {
+                var code: [6 * @sizeOf(u32)]u8 = undefined;
+                data_blk_outer: {
+                    const this_addr = stub_helper.addr;
+                    const target_addr = data.addr + data.size - @sizeOf(u64);
+                    data_blk: {
+                        const displacement = math.cast(i21, target_addr - this_addr) catch |_| break :data_blk;
+                        // adr x17, disp
+                        mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.adr(.x17, displacement).toU32());
+                        // nop
+                        mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.nop().toU32());
+                        break :data_blk_outer;
+                    }
+                    data_blk: {
+                        const new_this_addr = this_addr + @sizeOf(u32);
+                        const displacement = math.cast(i21, target_addr - new_this_addr) catch |_| break :data_blk;
+                        // nop
+                        mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.nop().toU32());
+                        // adr x17, disp
+                        mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.adr(.x17, displacement).toU32());
+                        break :data_blk_outer;
+                    }
+                    // Jump is too big, replace adr with adrp and add.
+                    const this_page = @intCast(i32, this_addr >> 12);
+                    const target_page = @intCast(i32, target_addr >> 12);
+                    const pages = @intCast(i21, target_page - this_page);
+                    mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.adrp(.x17, pages).toU32());
+                    const narrowed = @truncate(u12, target_addr);
+                    mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.add(.x17, .x17, narrowed, false).toU32());
+                }
+                // stp x16, x17, [sp, #-16]!
+                code[8] = 0xf0;
+                code[9] = 0x47;
+                code[10] = 0xbf;
+                code[11] = 0xa9;
+                binder_blk_outer: {
+                    const dyld_stub_binder = self.nonlazy_imports.get("dyld_stub_binder").?;
+                    const this_addr = stub_helper.addr + 3 * @sizeOf(u32);
+                    const target_addr = (got.addr + dyld_stub_binder.index * @sizeOf(u64));
+                    binder_blk: {
+                        const displacement = math.divExact(u64, target_addr - this_addr, 4) catch |_| break :binder_blk;
+                        const literal = math.cast(u18, displacement) catch |_| break :binder_blk;
+                        // ldr x16, label
+                        mem.writeIntLittle(u32, code[12..16], aarch64.Instruction.ldr(.x16, .{
+                            .literal = literal,
+                        }).toU32());
+                        // nop
+                        mem.writeIntLittle(u32, code[16..20], aarch64.Instruction.nop().toU32());
+                        break :binder_blk_outer;
+                    }
+                    binder_blk: {
+                        const new_this_addr = this_addr + @sizeOf(u32);
+                        const displacement = math.divExact(u64, target_addr - new_this_addr, 4) catch |_| break :binder_blk;
+                        const literal = math.cast(u18, displacement) catch |_| break :binder_blk;
+                        log.debug("2: disp=0x{x}, literal=0x{x}", .{ displacement, literal });
+                        // Pad with nop to please division.
+                        // nop
+                        mem.writeIntLittle(u32, code[12..16], aarch64.Instruction.nop().toU32());
+                        // ldr x16, label
+                        mem.writeIntLittle(u32, code[16..20], aarch64.Instruction.ldr(.x16, .{
+                            .literal = literal,
+                        }).toU32());
+                        break :binder_blk_outer;
+                    }
+                    // Use adrp followed by ldr(immediate).
+                    const this_page = @intCast(i32, this_addr >> 12);
+                    const target_page = @intCast(i32, target_addr >> 12);
+                    const pages = @intCast(i21, target_page - this_page);
+                    mem.writeIntLittle(u32, code[12..16], aarch64.Instruction.adrp(.x16, pages).toU32());
+                    const narrowed = @truncate(u12, target_addr);
+                    const offset = try math.divExact(u12, narrowed, 8);
+                    mem.writeIntLittle(u32, code[16..20], aarch64.Instruction.ldr(.x16, .{
+                        .register = .{
+                            .rn = .x16,
+                            .offset = aarch64.Instruction.LoadStoreOffset.imm(offset),
+                        },
+                    }).toU32());
+                }
+                // br x16
+                code[20] = 0x00;
+                code[21] = 0x02;
+                code[22] = 0x1f;
+                code[23] = 0xd6;
+                try self.file.?.pwriteAll(&code, stub_helper.offset);
+                break :blk stub_helper.offset + 6 * @sizeOf(u32);
+            },
+            else => unreachable,
+        }
+    };
+
+    for (self.lazy_imports.items()) |_, i| {
+        const index = @intCast(u32, i);
+        try self.writeLazySymbolPointer(index);
+        try self.writeStub(index);
+        try self.writeStubInStubHelper(index);
+    }
+}
+
+fn writeLazySymbolPointer(self: *Zld, index: u32) !void {
+    const text_segment = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    const stub_helper = text_segment.sections.items[self.stub_helper_section_index.?];
+    const data_segment = self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+    const la_symbol_ptr = data_segment.sections.items[self.la_symbol_ptr_section_index.?];
+
+    const stub_size: u4 = switch (self.arch.?) {
+        .x86_64 => 10,
+        .aarch64 => 3 * @sizeOf(u32),
+        else => unreachable,
+    };
+    const stub_off = self.stub_helper_stubs_start_off.? + index * stub_size;
+    const end = stub_helper.addr + stub_off - stub_helper.offset;
+    var buf: [@sizeOf(u64)]u8 = undefined;
+    mem.writeIntLittle(u64, &buf, end);
+    const off = la_symbol_ptr.offset + index * @sizeOf(u64);
+    log.debug("writing lazy symbol pointer entry 0x{x} at 0x{x}", .{ end, off });
+    try self.file.?.pwriteAll(&buf, off);
+}
+
+fn writeStub(self: *Zld, index: u32) !void {
+    const text_segment = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    const stubs = text_segment.sections.items[self.stubs_section_index.?];
+    const data_segment = self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+    const la_symbol_ptr = data_segment.sections.items[self.la_symbol_ptr_section_index.?];
+
+    const stub_off = stubs.offset + index * stubs.reserved2;
+    const stub_addr = stubs.addr + index * stubs.reserved2;
+    const la_ptr_addr = la_symbol_ptr.addr + index * @sizeOf(u64);
+    log.debug("writing stub at 0x{x}", .{stub_off});
+    var code = try self.allocator.alloc(u8, stubs.reserved2);
+    defer self.allocator.free(code);
+    switch (self.arch.?) {
+        .x86_64 => {
+            assert(la_ptr_addr >= stub_addr + stubs.reserved2);
+            const displacement = try math.cast(u32, la_ptr_addr - stub_addr - stubs.reserved2);
+            // jmp
+            code[0] = 0xff;
+            code[1] = 0x25;
+            mem.writeIntLittle(u32, code[2..][0..4], displacement);
+        },
+        .aarch64 => {
+            assert(la_ptr_addr >= stub_addr);
+            outer: {
+                const this_addr = stub_addr;
+                const target_addr = la_ptr_addr;
+                inner: {
+                    const displacement = math.divExact(u64, target_addr - this_addr, 4) catch |_| break :inner;
+                    const literal = math.cast(u18, displacement) catch |_| break :inner;
+                    // ldr x16, literal
+                    mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.ldr(.x16, .{
+                        .literal = literal,
+                    }).toU32());
+                    // nop
+                    mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.nop().toU32());
+                    break :outer;
+                }
+                inner: {
+                    const new_this_addr = this_addr + @sizeOf(u32);
+                    const displacement = math.divExact(u64, target_addr - new_this_addr, 4) catch |_| break :inner;
+                    const literal = math.cast(u18, displacement) catch |_| break :inner;
+                    // nop
+                    mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.nop().toU32());
+                    // ldr x16, literal
+                    mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.ldr(.x16, .{
+                        .literal = literal,
+                    }).toU32());
+                    break :outer;
+                }
+                // Use adrp followed by ldr(immediate).
+                const this_page = @intCast(i32, this_addr >> 12);
+                const target_page = @intCast(i32, target_addr >> 12);
+                const pages = @intCast(i21, target_page - this_page);
+                mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.adrp(.x16, pages).toU32());
+                const narrowed = @truncate(u12, target_addr);
+                const offset = try math.divExact(u12, narrowed, 8);
+                mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.ldr(.x16, .{
+                    .register = .{
+                        .rn = .x16,
+                        .offset = aarch64.Instruction.LoadStoreOffset.imm(offset),
+                    },
+                }).toU32());
+            }
+            // br x16
+            mem.writeIntLittle(u32, code[8..12], aarch64.Instruction.br(.x16).toU32());
+        },
+        else => unreachable,
+    }
+    try self.file.?.pwriteAll(code, stub_off);
+}
+
+fn writeStubInStubHelper(self: *Zld, index: u32) !void {
+    const text_segment = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    const stub_helper = text_segment.sections.items[self.stub_helper_section_index.?];
+
+    const stub_size: u4 = switch (self.arch.?) {
+        .x86_64 => 10,
+        .aarch64 => 3 * @sizeOf(u32),
+        else => unreachable,
+    };
+    const stub_off = self.stub_helper_stubs_start_off.? + index * stub_size;
+    var code = try self.allocator.alloc(u8, stub_size);
+    defer self.allocator.free(code);
+    switch (self.arch.?) {
+        .x86_64 => {
+            const displacement = try math.cast(
+                i32,
+                @intCast(i64, stub_helper.offset) - @intCast(i64, stub_off) - stub_size,
+            );
+            // pushq
+            code[0] = 0x68;
+            mem.writeIntLittle(u32, code[1..][0..4], 0x0); // Just a placeholder populated in `populateLazyBindOffsetsInStubHelper`.
+            // jmpq
+            code[5] = 0xe9;
+            mem.writeIntLittle(u32, code[6..][0..4], @bitCast(u32, displacement));
+        },
+        .aarch64 => {
+            const displacement = try math.cast(i28, @intCast(i64, stub_helper.offset) - @intCast(i64, stub_off) - 4);
+            const literal = @divExact(stub_size - @sizeOf(u32), 4);
+            // ldr w16, literal
+            mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.ldr(.w16, .{
+                .literal = literal,
+            }).toU32());
+            // b disp
+            mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.b(displacement).toU32());
+            mem.writeIntLittle(u32, code[8..12], 0x0); // Just a placeholder populated in `populateLazyBindOffsetsInStubHelper`.
+        },
+        else => unreachable,
+    }
+    try self.file.?.pwriteAll(code, stub_off);
+}
+
+fn resolveSymbols(self: *Zld) !void {
+    for (self.objects.items) |object, object_id| {
+        const seg = object.load_commands.items[object.segment_cmd_index.?].Segment;
+        log.debug("\n\n", .{});
+        log.debug("resolving symbols in {s}", .{object.name});
+
+        for (object.symtab.items) |sym| {
+            if (isImport(&sym)) continue;
+
+            const sym_name = object.getString(sym.n_strx);
+            const out_name = try self.allocator.dupe(u8, sym_name);
+            const locs = try self.locals.getOrPut(self.allocator, out_name);
+            defer {
+                if (locs.found_existing) self.allocator.free(out_name);
+            }
+
+            if (!locs.found_existing) {
+                locs.entry.value = .{};
+            }
+
+            const tt: Symbol.Type = blk: {
+                if (isLocal(&sym)) {
+                    break :blk .Local;
+                } else if (isWeakDef(&sym)) {
+                    break :blk .WeakGlobal;
+                } else {
+                    break :blk .Global;
+                }
+            };
+            if (tt == .Global) {
+                for (locs.entry.value.items) |ss| {
+                    if (ss.tt == .Global) {
+                        log.debug("symbol already defined '{s}'", .{sym_name});
+                        continue;
+                        // log.err("symbol '{s}' defined multiple times: {}", .{ sym_name, sym });
+                        // return error.MultipleSymbolDefinitions;
+                    }
+                }
+            }
+
+            const source_sect_id = sym.n_sect - 1;
+            const target_mapping = self.mappings.get(.{
+                .object_id = @intCast(u16, object_id),
+                .source_sect_id = source_sect_id,
+            }) orelse {
+                if (self.unhandled_sections.get(.{
+                    .object_id = @intCast(u16, object_id),
+                    .source_sect_id = source_sect_id,
+                }) != null) continue;
+
+                log.err("section not mapped for symbol '{s}': {}", .{ sym_name, sym });
+                return error.SectionNotMappedForSymbol;
+            };
+            const source_sect = seg.sections.items[source_sect_id];
+            const target_seg = self.load_commands.items[target_mapping.target_seg_id].Segment;
+            const target_sect = target_seg.sections.items[target_mapping.target_sect_id];
+            const target_addr = target_sect.addr + target_mapping.offset;
+            const n_value = sym.n_value - source_sect.addr + target_addr;
+
+            log.debug("resolving '{s}':{} as {s} symbol at 0x{x}", .{ sym_name, sym, tt, n_value });
+
+            // TODO there might be a more generic way of doing this.
+            var n_sect: u16 = 0;
+            for (self.load_commands.items) |cmd, cmd_id| {
+                if (cmd != .Segment) break;
+                if (cmd_id == target_mapping.target_seg_id) {
+                    n_sect += target_mapping.target_sect_id + 1;
+                    break;
+                }
+                n_sect += @intCast(u16, cmd.Segment.sections.items.len);
+            }
+
+            const n_strx = try self.makeString(sym_name);
+            try locs.entry.value.append(self.allocator, .{
+                .inner = .{
+                    .n_strx = n_strx,
+                    .n_value = n_value,
+                    .n_type = macho.N_SECT,
+                    .n_desc = sym.n_desc,
+                    .n_sect = @intCast(u8, n_sect),
+                },
+                .tt = tt,
+                .object_id = @intCast(u16, object_id),
+            });
+        }
+    }
+}
+
+fn doRelocs(self: *Zld) !void {
+    for (self.objects.items) |object, object_id| {
+        log.debug("\n\n", .{});
+        log.debug("relocating object {s}", .{object.name});
+
+        const seg = object.load_commands.items[object.segment_cmd_index.?].Segment;
+
+        for (seg.sections.items) |sect, source_sect_id| {
+            const segname = parseName(&sect.segname);
+            const sectname = parseName(&sect.sectname);
+
+            var code = try self.allocator.alloc(u8, sect.size);
+            _ = try object.file.preadAll(code, sect.offset);
+            defer self.allocator.free(code);
+
+            // Parse relocs (if any)
+            var raw_relocs = try self.allocator.alloc(u8, @sizeOf(macho.relocation_info) * sect.nreloc);
+            defer self.allocator.free(raw_relocs);
+            _ = try object.file.preadAll(raw_relocs, sect.reloff);
+            const relocs = mem.bytesAsSlice(macho.relocation_info, raw_relocs);
+
+            // Get mapping
+            const target_mapping = self.mappings.get(.{
+                .object_id = @intCast(u16, object_id),
+                .source_sect_id = @intCast(u16, source_sect_id),
+            }) orelse {
+                log.debug("no mapping for {s},{s}; skipping", .{ segname, sectname });
+                continue;
+            };
+            const target_seg = self.load_commands.items[target_mapping.target_seg_id].Segment;
+            const target_sect = target_seg.sections.items[target_mapping.target_sect_id];
+            const target_sect_addr = target_sect.addr + target_mapping.offset;
+            const target_sect_off = target_sect.offset + target_mapping.offset;
+
+            var addend: ?u64 = null;
+            var sub: ?i64 = null;
+
+            for (relocs) |rel| {
+                const off = @intCast(u32, rel.r_address);
+                const this_addr = target_sect_addr + off;
+
+                switch (self.arch.?) {
+                    .aarch64 => {
+                        const rel_type = @intToEnum(macho.reloc_type_arm64, rel.r_type);
+                        log.debug("{s}", .{rel_type});
+                        log.debug("    | source address 0x{x}", .{this_addr});
+                        log.debug("    | offset 0x{x}", .{off});
+
+                        if (rel_type == .ARM64_RELOC_ADDEND) {
+                            addend = rel.r_symbolnum;
+                            log.debug("    | calculated addend = 0x{x}", .{addend});
+                            // TODO followed by either PAGE21 or PAGEOFF12 only.
+                            continue;
+                        }
+                    },
+                    .x86_64 => {
+                        const rel_type = @intToEnum(macho.reloc_type_x86_64, rel.r_type);
+                        log.debug("{s}", .{rel_type});
+                        log.debug("    | source address 0x{x}", .{this_addr});
+                        log.debug("    | offset 0x{x}", .{off});
+                    },
+                    else => {},
+                }
+
+                const target_addr = try self.relocTargetAddr(@intCast(u16, object_id), rel);
+                log.debug("    | target address 0x{x}", .{target_addr});
+                if (rel.r_extern == 1) {
+                    const target_symname = object.getString(object.symtab.items[rel.r_symbolnum].n_strx);
+                    log.debug("    | target symbol '{s}'", .{target_symname});
+                } else {
+                    const target_sectname = seg.sections.items[rel.r_symbolnum - 1].sectname;
+                    log.debug("    | target section '{s}'", .{parseName(&target_sectname)});
+                }
+
+                switch (self.arch.?) {
+                    .x86_64 => {
+                        const rel_type = @intToEnum(macho.reloc_type_x86_64, rel.r_type);
+
+                        switch (rel_type) {
+                            .X86_64_RELOC_BRANCH => {
+                                assert(rel.r_length == 2);
+                                const inst = code[off..][0..4];
+                                const displacement = @bitCast(u32, @intCast(i32, @intCast(i64, target_addr) - @intCast(i64, this_addr) - 4));
+                                mem.writeIntLittle(u32, inst, displacement);
+                            },
+                            .X86_64_RELOC_GOT_LOAD => {
+                                assert(rel.r_length == 2);
+                                const inst = code[off..][0..4];
+                                const displacement = @bitCast(u32, @intCast(i32, @intCast(i64, target_addr) - @intCast(i64, this_addr) - 4));
+
+                                blk: {
+                                    const data_const_seg = self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+                                    const got = data_const_seg.sections.items[self.got_section_index.?];
+                                    if (got.addr <= target_addr and target_addr < got.addr + got.size) break :blk;
+                                    log.debug("    | rewriting to leaq", .{});
+                                    code[off - 2] = 0x8d;
+                                }
+
+                                mem.writeIntLittle(u32, inst, displacement);
+                            },
+                            .X86_64_RELOC_GOT => {
+                                assert(rel.r_length == 2);
+                                // TODO Instead of referring to the target symbol directly, we refer to it
+                                // indirectly via GOT. Getting actual target address should be done in the
+                                // helper relocTargetAddr function rather than here.
+                                const sym = object.symtab.items[rel.r_symbolnum];
+                                const sym_name = try self.allocator.dupe(u8, object.getString(sym.n_strx));
+                                const res = try self.nonlazy_pointers.getOrPut(self.allocator, sym_name);
+                                defer if (res.found_existing) self.allocator.free(sym_name);
+
+                                const data_const_seg = self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+                                const got = data_const_seg.sections.items[self.got_section_index.?];
+
+                                if (!res.found_existing) {
+                                    const index = @intCast(u32, self.nonlazy_pointers.items().len) - 1;
+                                    assert(index < max_local_got_indirections); // TODO This is just a temp solution.
+                                    res.entry.value = .{
+                                        .index = index,
+                                        .target_addr = target_addr,
+                                    };
+                                    var buf: [@sizeOf(u64)]u8 = undefined;
+                                    mem.writeIntLittle(u64, &buf, target_addr);
+                                    const got_offset = got.offset + (index + self.nonlazy_imports.items().len) * @sizeOf(u64);
+
+                                    log.debug("    | GOT off 0x{x}", .{got.offset});
+                                    log.debug("    | writing GOT entry 0x{x} at 0x{x}", .{ target_addr, got_offset });
+
+                                    try self.file.?.pwriteAll(&buf, got_offset);
+                                }
+
+                                const index = res.entry.value.index + self.nonlazy_imports.items().len;
+                                const actual_target_addr = got.addr + index * @sizeOf(u64);
+
+                                log.debug("    | GOT addr 0x{x}", .{got.addr});
+                                log.debug("    | actual target address in GOT 0x{x}", .{actual_target_addr});
+
+                                const inst = code[off..][0..4];
+                                const displacement = @bitCast(u32, @intCast(i32, @intCast(i64, actual_target_addr) - @intCast(i64, this_addr) - 4));
+                                mem.writeIntLittle(u32, inst, displacement);
+                            },
+                            .X86_64_RELOC_TLV => {
+                                assert(rel.r_length == 2);
+                                // We need to rewrite the opcode from movq to leaq.
+                                code[off - 2] = 0x8d;
+                                // Add displacement.
+                                const inst = code[off..][0..4];
+                                const displacement = @bitCast(u32, @intCast(i32, @intCast(i64, target_addr) - @intCast(i64, this_addr) - 4));
+                                mem.writeIntLittle(u32, inst, displacement);
+                            },
+                            .X86_64_RELOC_SIGNED,
+                            .X86_64_RELOC_SIGNED_1,
+                            .X86_64_RELOC_SIGNED_2,
+                            .X86_64_RELOC_SIGNED_4,
+                            => {
+                                assert(rel.r_length == 2);
+                                const inst = code[off..][0..4];
+                                const offset = @intCast(i64, mem.readIntLittle(i32, inst));
+                                log.debug("    | calculated addend 0x{x}", .{offset});
+                                const actual_target_addr = blk: {
+                                    if (rel.r_extern == 1) {
+                                        break :blk @intCast(i64, target_addr) + offset;
+                                    } else {
+                                        const correction: i4 = switch (rel_type) {
+                                            .X86_64_RELOC_SIGNED => 0,
+                                            .X86_64_RELOC_SIGNED_1 => 1,
+                                            .X86_64_RELOC_SIGNED_2 => 2,
+                                            .X86_64_RELOC_SIGNED_4 => 4,
+                                            else => unreachable,
+                                        };
+                                        log.debug("    | calculated correction 0x{x}", .{correction});
+
+                                        // The value encoded in the instruction is a displacement - 4 - correction.
+                                        // To obtain the adjusted target address in the final binary, we need
+                                        // calculate the original target address within the object file, establish
+                                        // what the offset from the original target section was, and apply this
+                                        // offset to the resultant target section with this relocated binary.
+                                        const orig_sect_id = @intCast(u16, rel.r_symbolnum - 1);
+                                        const target_map = self.mappings.get(.{
+                                            .object_id = @intCast(u16, object_id),
+                                            .source_sect_id = orig_sect_id,
+                                        }) orelse unreachable;
+                                        const orig_seg = object.load_commands.items[object.segment_cmd_index.?].Segment;
+                                        const orig_sect = orig_seg.sections.items[orig_sect_id];
+                                        const orig_offset = off + offset + 4 + correction - @intCast(i64, orig_sect.addr);
+                                        log.debug("    | original offset 0x{x}", .{orig_offset});
+                                        const adjusted = @intCast(i64, target_addr) + orig_offset;
+                                        log.debug("    | adjusted target address 0x{x}", .{adjusted});
+                                        break :blk adjusted - correction;
+                                    }
+                                };
+                                const result = actual_target_addr - @intCast(i64, this_addr) - 4;
+                                const displacement = @bitCast(u32, @intCast(i32, result));
+                                mem.writeIntLittle(u32, inst, displacement);
+                            },
+                            .X86_64_RELOC_SUBTRACTOR => {
+                                sub = @intCast(i64, target_addr);
+                            },
+                            .X86_64_RELOC_UNSIGNED => {
+                                switch (rel.r_length) {
+                                    3 => {
+                                        const inst = code[off..][0..8];
+                                        const offset = mem.readIntLittle(i64, inst);
+
+                                        const result = outer: {
+                                            if (rel.r_extern == 1) {
+                                                log.debug("    | calculated addend 0x{x}", .{offset});
+                                                if (sub) |s| {
+                                                    break :outer @intCast(i64, target_addr) - s + offset;
+                                                } else {
+                                                    break :outer @intCast(i64, target_addr) + offset;
+                                                }
+                                            } else {
+                                                // The value encoded in the instruction is an absolute offset
+                                                // from the start of MachO header to the target address in the
+                                                // object file. To extract the address, we calculate the offset from
+                                                // the beginning of the source section to the address, and apply it to
+                                                // the target address value.
+                                                const orig_sect_id = @intCast(u16, rel.r_symbolnum - 1);
+                                                const target_map = self.mappings.get(.{
+                                                    .object_id = @intCast(u16, object_id),
+                                                    .source_sect_id = orig_sect_id,
+                                                }) orelse unreachable;
+                                                const orig_seg = object.load_commands.items[object.segment_cmd_index.?].Segment;
+                                                const orig_sect = orig_seg.sections.items[orig_sect_id];
+                                                const orig_offset = offset - @intCast(i64, orig_sect.addr);
+                                                const actual_target_addr = inner: {
+                                                    if (sub) |s| {
+                                                        break :inner @intCast(i64, target_addr) - s + orig_offset;
+                                                    } else {
+                                                        break :inner @intCast(i64, target_addr) + orig_offset;
+                                                    }
+                                                };
+                                                log.debug("    | adjusted target address 0x{x}", .{actual_target_addr});
+                                                break :outer actual_target_addr;
+                                            }
+                                        };
+                                        mem.writeIntLittle(u64, inst, @bitCast(u64, result));
+                                        sub = null;
+
+                                        rebases: {
+                                            var hit: bool = false;
+                                            if (target_mapping.target_seg_id == self.data_segment_cmd_index.?) {
+                                                if (self.data_section_index) |index| {
+                                                    if (index == target_mapping.target_sect_id) hit = true;
+                                                }
+                                            }
+                                            if (target_mapping.target_seg_id == self.data_const_segment_cmd_index.?) {
+                                                if (self.data_const_section_index) |index| {
+                                                    if (index == target_mapping.target_sect_id) hit = true;
+                                                }
+                                            }
+
+                                            if (!hit) break :rebases;
+
+                                            try self.local_rebases.append(self.allocator, .{
+                                                .offset = this_addr - target_seg.inner.vmaddr,
+                                                .segment_id = target_mapping.target_seg_id,
+                                            });
+                                        }
+                                        // TLV is handled via a separate offset mechanism.
+                                        // Calculate the offset to the initializer.
+                                        if (target_sect.flags == macho.S_THREAD_LOCAL_VARIABLES) tlv: {
+                                            assert(rel.r_extern == 1);
+                                            const sym = object.symtab.items[rel.r_symbolnum];
+                                            if (isImport(&sym)) break :tlv;
+
+                                            const base_addr = blk: {
+                                                if (self.tlv_data_section_index) |index| {
+                                                    const tlv_data = target_seg.sections.items[index];
+                                                    break :blk tlv_data.addr;
+                                                } else {
+                                                    const tlv_bss = target_seg.sections.items[self.tlv_bss_section_index.?];
+                                                    break :blk tlv_bss.addr;
+                                                }
+                                            };
+                                            // Since we require TLV data to always preceed TLV bss section, we calculate
+                                            // offsets wrt to the former if it is defined; otherwise, wrt to the latter.
+                                            try self.threadlocal_offsets.append(self.allocator, target_addr - base_addr);
+                                        }
+                                    },
+                                    2 => {
+                                        const inst = code[off..][0..4];
+                                        const offset = mem.readIntLittle(i32, inst);
+                                        log.debug("    | calculated addend 0x{x}", .{offset});
+                                        const result = if (sub) |s|
+                                            @intCast(i64, target_addr) - s + offset
+                                        else
+                                            @intCast(i64, target_addr) + offset;
+                                        mem.writeIntLittle(u32, inst, @truncate(u32, @bitCast(u64, result)));
+                                        sub = null;
+                                    },
+                                    else => |len| {
+                                        log.err("unexpected relocation length 0x{x}", .{len});
+                                        return error.UnexpectedRelocationLength;
+                                    },
+                                }
+                            },
+                        }
+                    },
+                    .aarch64 => {
+                        const rel_type = @intToEnum(macho.reloc_type_arm64, rel.r_type);
+
+                        switch (rel_type) {
+                            .ARM64_RELOC_BRANCH26 => {
+                                assert(rel.r_length == 2);
+                                const inst = code[off..][0..4];
+                                const displacement = @intCast(
+                                    i28,
+                                    @intCast(i64, target_addr) - @intCast(i64, this_addr),
+                                );
+                                var parsed = mem.bytesAsValue(
+                                    meta.TagPayload(
+                                        aarch64.Instruction,
+                                        aarch64.Instruction.UnconditionalBranchImmediate,
+                                    ),
+                                    inst,
+                                );
+                                parsed.imm26 = @truncate(u26, @bitCast(u28, displacement) >> 2);
+                            },
+                            .ARM64_RELOC_PAGE21,
+                            .ARM64_RELOC_GOT_LOAD_PAGE21,
+                            .ARM64_RELOC_TLVP_LOAD_PAGE21,
+                            => {
+                                assert(rel.r_length == 2);
+                                const inst = code[off..][0..4];
+                                const ta = if (addend) |a| target_addr + a else target_addr;
+                                const this_page = @intCast(i32, this_addr >> 12);
+                                const target_page = @intCast(i32, ta >> 12);
+                                const pages = @bitCast(u21, @intCast(i21, target_page - this_page));
+                                log.debug("    | moving by {} pages", .{pages});
+                                var parsed = mem.bytesAsValue(
+                                    meta.TagPayload(
+                                        aarch64.Instruction,
+                                        aarch64.Instruction.PCRelativeAddress,
+                                    ),
+                                    inst,
+                                );
+                                parsed.immhi = @truncate(u19, pages >> 2);
+                                parsed.immlo = @truncate(u2, pages);
+                                addend = null;
+                            },
+                            .ARM64_RELOC_PAGEOFF12,
+                            .ARM64_RELOC_GOT_LOAD_PAGEOFF12,
+                            => {
+                                const inst = code[off..][0..4];
+                                if (aarch64IsArithmetic(inst)) {
+                                    log.debug("    | detected ADD opcode", .{});
+                                    // add
+                                    var parsed = mem.bytesAsValue(
+                                        meta.TagPayload(
+                                            aarch64.Instruction,
+                                            aarch64.Instruction.AddSubtractImmediate,
+                                        ),
+                                        inst,
+                                    );
+                                    const ta = if (addend) |a| target_addr + a else target_addr;
+                                    const narrowed = @truncate(u12, ta);
+                                    parsed.imm12 = narrowed;
+                                } else {
+                                    log.debug("    | detected LDR/STR opcode", .{});
+                                    // ldr/str
+                                    var parsed = mem.bytesAsValue(
+                                        meta.TagPayload(
+                                            aarch64.Instruction,
+                                            aarch64.Instruction.LoadStoreRegister,
+                                        ),
+                                        inst,
+                                    );
+
+                                    const ta = if (addend) |a| target_addr + a else target_addr;
+                                    const narrowed = @truncate(u12, ta);
+                                    log.debug("    | narrowed 0x{x}", .{narrowed});
+                                    log.debug("    | parsed.size 0x{x}", .{parsed.size});
+
+                                    if (rel_type == .ARM64_RELOC_GOT_LOAD_PAGEOFF12) blk: {
+                                        const data_const_seg = self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+                                        const got = data_const_seg.sections.items[self.got_section_index.?];
+                                        if (got.addr <= target_addr and target_addr < got.addr + got.size) break :blk;
+
+                                        log.debug("    | rewriting to add", .{});
+                                        mem.writeIntLittle(u32, inst, aarch64.Instruction.add(
+                                            @intToEnum(aarch64.Register, parsed.rt),
+                                            @intToEnum(aarch64.Register, parsed.rn),
+                                            narrowed,
+                                            false,
+                                        ).toU32());
+                                        addend = null;
+                                        continue;
+                                    }
+
+                                    const offset: u12 = blk: {
+                                        if (parsed.size == 0) {
+                                            if (parsed.v == 1) {
+                                                // 128-bit SIMD is scaled by 16.
+                                                break :blk try math.divExact(u12, narrowed, 16);
+                                            }
+                                            // Otherwise, 8-bit SIMD or ldrb.
+                                            break :blk narrowed;
+                                        } else {
+                                            const denom: u4 = try math.powi(u4, 2, parsed.size);
+                                            break :blk try math.divExact(u12, narrowed, denom);
+                                        }
+                                    };
+                                    parsed.offset = offset;
+                                }
+                                addend = null;
+                            },
+                            .ARM64_RELOC_TLVP_LOAD_PAGEOFF12 => {
+                                const RegInfo = struct {
+                                    rd: u5,
+                                    rn: u5,
+                                    size: u1,
+                                };
+                                const inst = code[off..][0..4];
+                                const parsed: RegInfo = blk: {
+                                    if (aarch64IsArithmetic(inst)) {
+                                        const curr = mem.bytesAsValue(
+                                            meta.TagPayload(
+                                                aarch64.Instruction,
+                                                aarch64.Instruction.AddSubtractImmediate,
+                                            ),
+                                            inst,
+                                        );
+                                        break :blk .{ .rd = curr.rd, .rn = curr.rn, .size = curr.sf };
+                                    } else {
+                                        const curr = mem.bytesAsValue(
+                                            meta.TagPayload(
+                                                aarch64.Instruction,
+                                                aarch64.Instruction.LoadStoreRegister,
+                                            ),
+                                            inst,
+                                        );
+                                        break :blk .{ .rd = curr.rt, .rn = curr.rn, .size = @truncate(u1, curr.size) };
+                                    }
+                                };
+                                const ta = if (addend) |a| target_addr + a else target_addr;
+                                const narrowed = @truncate(u12, ta);
+                                log.debug("    | rewriting TLV access to ADD opcode", .{});
+                                // For TLV, we always generate an add instruction.
+                                mem.writeIntLittle(u32, inst, aarch64.Instruction.add(
+                                    @intToEnum(aarch64.Register, parsed.rd),
+                                    @intToEnum(aarch64.Register, parsed.rn),
+                                    narrowed,
+                                    false,
+                                ).toU32());
+                            },
+                            .ARM64_RELOC_SUBTRACTOR => {
+                                sub = @intCast(i64, target_addr);
+                            },
+                            .ARM64_RELOC_UNSIGNED => {
+                                switch (rel.r_length) {
+                                    3 => {
+                                        const inst = code[off..][0..8];
+                                        const offset = mem.readIntLittle(i64, inst);
+                                        log.debug("    | calculated addend 0x{x}", .{offset});
+                                        const result = if (sub) |s|
+                                            @intCast(i64, target_addr) - s + offset
+                                        else
+                                            @intCast(i64, target_addr) + offset;
+                                        mem.writeIntLittle(u64, inst, @bitCast(u64, result));
+                                        sub = null;
+
+                                        rebases: {
+                                            var hit: bool = false;
+                                            if (target_mapping.target_seg_id == self.data_segment_cmd_index.?) {
+                                                if (self.data_section_index) |index| {
+                                                    if (index == target_mapping.target_sect_id) hit = true;
+                                                }
+                                            }
+                                            if (target_mapping.target_seg_id == self.data_const_segment_cmd_index.?) {
+                                                if (self.data_const_section_index) |index| {
+                                                    if (index == target_mapping.target_sect_id) hit = true;
+                                                }
+                                            }
+
+                                            if (!hit) break :rebases;
+
+                                            try self.local_rebases.append(self.allocator, .{
+                                                .offset = this_addr - target_seg.inner.vmaddr,
+                                                .segment_id = target_mapping.target_seg_id,
+                                            });
+                                        }
+                                        // TLV is handled via a separate offset mechanism.
+                                        // Calculate the offset to the initializer.
+                                        if (target_sect.flags == macho.S_THREAD_LOCAL_VARIABLES) tlv: {
+                                            assert(rel.r_extern == 1);
+                                            const sym = object.symtab.items[rel.r_symbolnum];
+                                            if (isImport(&sym)) break :tlv;
+
+                                            const base_addr = blk: {
+                                                if (self.tlv_data_section_index) |index| {
+                                                    const tlv_data = target_seg.sections.items[index];
+                                                    break :blk tlv_data.addr;
+                                                } else {
+                                                    const tlv_bss = target_seg.sections.items[self.tlv_bss_section_index.?];
+                                                    break :blk tlv_bss.addr;
+                                                }
+                                            };
+                                            // Since we require TLV data to always preceed TLV bss section, we calculate
+                                            // offsets wrt to the former if it is defined; otherwise, wrt to the latter.
+                                            try self.threadlocal_offsets.append(self.allocator, target_addr - base_addr);
+                                        }
+                                    },
+                                    2 => {
+                                        const inst = code[off..][0..4];
+                                        const offset = mem.readIntLittle(i32, inst);
+                                        log.debug("    | calculated addend 0x{x}", .{offset});
+                                        const result = if (sub) |s|
+                                            @intCast(i64, target_addr) - s + offset
+                                        else
+                                            @intCast(i64, target_addr) + offset;
+                                        mem.writeIntLittle(u32, inst, @truncate(u32, @bitCast(u64, result)));
+                                        sub = null;
+                                    },
+                                    else => |len| {
+                                        log.err("unexpected relocation length 0x{x}", .{len});
+                                        return error.UnexpectedRelocationLength;
+                                    },
+                                }
+                            },
+                            .ARM64_RELOC_POINTER_TO_GOT => return error.TODOArm64RelocPointerToGot,
+                            else => unreachable,
+                        }
+                    },
+                    else => unreachable,
+                }
+            }
+
+            log.debug("writing contents of '{s},{s}' section from '{s}' from 0x{x} to 0x{x}", .{
+                segname,
+                sectname,
+                object.name,
+                target_sect_off,
+                target_sect_off + code.len,
+            });
+
+            if (target_sect.flags == macho.S_ZEROFILL or
+                target_sect.flags == macho.S_THREAD_LOCAL_ZEROFILL or
+                target_sect.flags == macho.S_THREAD_LOCAL_VARIABLES)
+            {
+                log.debug("zeroing out '{s},{s}' from 0x{x} to 0x{x}", .{
+                    parseName(&target_sect.segname),
+                    parseName(&target_sect.sectname),
+                    target_sect_off,
+                    target_sect_off + code.len,
+                });
+                // Zero-out the space
+                var zeroes = try self.allocator.alloc(u8, code.len);
+                defer self.allocator.free(zeroes);
+                mem.set(u8, zeroes, 0);
+                try self.file.?.pwriteAll(zeroes, target_sect_off);
+            } else {
+                try self.file.?.pwriteAll(code, target_sect_off);
+            }
+        }
+    }
+}
+
+fn relocTargetAddr(self: *Zld, object_id: u16, rel: macho.relocation_info) !u64 {
+    const object = self.objects.items[object_id];
+    const seg = object.load_commands.items[object.segment_cmd_index.?].Segment;
+    const target_addr = blk: {
+        if (rel.r_extern == 1) {
+            const sym = object.symtab.items[rel.r_symbolnum];
+            if (isLocal(&sym) or isExport(&sym)) {
+                // Relocate using section offsets only.
+                const target_mapping = self.mappings.get(.{
+                    .object_id = object_id,
+                    .source_sect_id = sym.n_sect - 1,
+                }) orelse unreachable;
+                const source_sect = seg.sections.items[target_mapping.source_sect_id];
+                const target_seg = self.load_commands.items[target_mapping.target_seg_id].Segment;
+                const target_sect = target_seg.sections.items[target_mapping.target_sect_id];
+                const target_sect_addr = target_sect.addr + target_mapping.offset;
+                log.debug("    | symbol local to object", .{});
+                break :blk target_sect_addr + sym.n_value - source_sect.addr;
+            } else if (isImport(&sym)) {
+                // Relocate to either the artifact's local symbol, or an import from
+                // shared library.
+                const sym_name = object.getString(sym.n_strx);
+                if (self.locals.get(sym_name)) |locs| {
+                    var n_value: ?u64 = null;
+                    for (locs.items) |loc| {
+                        switch (loc.tt) {
+                            .Global => {
+                                n_value = loc.inner.n_value;
+                                break;
+                            },
+                            .WeakGlobal => {
+                                n_value = loc.inner.n_value;
+                            },
+                            .Local => {},
+                        }
+                    }
+                    if (n_value) |v| {
+                        break :blk v;
+                    }
+                    log.err("local symbol export '{s}' not found", .{sym_name});
+                    return error.LocalSymbolExportNotFound;
+                } else if (self.lazy_imports.get(sym_name)) |ext| {
+                    const segment = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+                    const stubs = segment.sections.items[self.stubs_section_index.?];
+                    break :blk stubs.addr + ext.index * stubs.reserved2;
+                } else if (self.nonlazy_imports.get(sym_name)) |ext| {
+                    const segment = self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+                    const got = segment.sections.items[self.got_section_index.?];
+                    break :blk got.addr + ext.index * @sizeOf(u64);
+                } else if (mem.eql(u8, sym_name, "__tlv_bootstrap")) {
+                    const segment = self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+                    const tlv = segment.sections.items[self.tlv_section_index.?];
+                    break :blk tlv.addr + self.tlv_bootstrap.?.index * @sizeOf(u64);
+                } else {
+                    log.err("failed to resolve symbol '{s}' as a relocation target", .{sym_name});
+                    return error.FailedToResolveRelocationTarget;
+                }
+            } else {
+                log.err("unexpected symbol {}, {s}", .{ sym, object.getString(sym.n_strx) });
+                return error.UnexpectedSymbolWhenRelocating;
+            }
+        } else {
+            // TODO I think we need to reparse the relocation_info as scattered_relocation_info
+            // here to get the actual section plus offset into that section of the relocated
+            // symbol. Unless the fine-grained location is encoded within the cell in the code
+            // buffer?
+            const target_mapping = self.mappings.get(.{
+                .object_id = object_id,
+                .source_sect_id = @intCast(u16, rel.r_symbolnum - 1),
+            }) orelse unreachable;
+            const target_seg = self.load_commands.items[target_mapping.target_seg_id].Segment;
+            const target_sect = target_seg.sections.items[target_mapping.target_sect_id];
+            break :blk target_sect.addr + target_mapping.offset;
+        }
+    };
+    return target_addr;
+}
+
+fn populateMetadata(self: *Zld) !void {
+    if (self.pagezero_segment_cmd_index == null) {
+        self.pagezero_segment_cmd_index = @intCast(u16, self.load_commands.items.len);
+        try self.load_commands.append(self.allocator, .{
+            .Segment = SegmentCommand.empty(.{
+                .cmd = macho.LC_SEGMENT_64,
+                .cmdsize = @sizeOf(macho.segment_command_64),
+                .segname = makeStaticString("__PAGEZERO"),
+                .vmaddr = 0,
+                .vmsize = 0x100000000, // size always set to 4GB
+                .fileoff = 0,
+                .filesize = 0,
+                .maxprot = 0,
+                .initprot = 0,
+                .nsects = 0,
+                .flags = 0,
+            }),
+        });
+    }
+
+    if (self.text_segment_cmd_index == null) {
+        self.text_segment_cmd_index = @intCast(u16, self.load_commands.items.len);
+        try self.load_commands.append(self.allocator, .{
+            .Segment = SegmentCommand.empty(.{
+                .cmd = macho.LC_SEGMENT_64,
+                .cmdsize = @sizeOf(macho.segment_command_64),
+                .segname = makeStaticString("__TEXT"),
+                .vmaddr = 0x100000000, // always starts at 4GB
+                .vmsize = 0,
+                .fileoff = 0,
+                .filesize = 0,
+                .maxprot = macho.VM_PROT_READ | macho.VM_PROT_EXECUTE,
+                .initprot = macho.VM_PROT_READ | macho.VM_PROT_EXECUTE,
+                .nsects = 0,
+                .flags = 0,
+            }),
+        });
+    }
+
+    if (self.text_section_index == null) {
+        const text_seg = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+        self.text_section_index = @intCast(u16, text_seg.sections.items.len);
+        const alignment: u2 = switch (self.arch.?) {
+            .x86_64 => 0,
+            .aarch64 => 2,
+            else => unreachable, // unhandled architecture type
+        };
+        try text_seg.addSection(self.allocator, .{
+            .sectname = makeStaticString("__text"),
+            .segname = makeStaticString("__TEXT"),
+            .addr = 0,
+            .size = 0,
+            .offset = 0,
+            .@"align" = alignment,
+            .reloff = 0,
+            .nreloc = 0,
+            .flags = macho.S_REGULAR | macho.S_ATTR_PURE_INSTRUCTIONS | macho.S_ATTR_SOME_INSTRUCTIONS,
+            .reserved1 = 0,
+            .reserved2 = 0,
+            .reserved3 = 0,
+        });
+    }
+
+    if (self.stubs_section_index == null) {
+        const text_seg = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+        self.stubs_section_index = @intCast(u16, text_seg.sections.items.len);
+        const alignment: u2 = switch (self.arch.?) {
+            .x86_64 => 0,
+            .aarch64 => 2,
+            else => unreachable, // unhandled architecture type
+        };
+        const stub_size: u4 = switch (self.arch.?) {
+            .x86_64 => 6,
+            .aarch64 => 3 * @sizeOf(u32),
+            else => unreachable, // unhandled architecture type
+        };
+        try text_seg.addSection(self.allocator, .{
+            .sectname = makeStaticString("__stubs"),
+            .segname = makeStaticString("__TEXT"),
+            .addr = 0,
+            .size = 0,
+            .offset = 0,
+            .@"align" = alignment,
+            .reloff = 0,
+            .nreloc = 0,
+            .flags = macho.S_SYMBOL_STUBS | macho.S_ATTR_PURE_INSTRUCTIONS | macho.S_ATTR_SOME_INSTRUCTIONS,
+            .reserved1 = 0,
+            .reserved2 = stub_size,
+            .reserved3 = 0,
+        });
+    }
+
+    if (self.stub_helper_section_index == null) {
+        const text_seg = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+        self.stub_helper_section_index = @intCast(u16, text_seg.sections.items.len);
+        const alignment: u2 = switch (self.arch.?) {
+            .x86_64 => 0,
+            .aarch64 => 2,
+            else => unreachable, // unhandled architecture type
+        };
+        const stub_helper_size: u6 = switch (self.arch.?) {
+            .x86_64 => 15,
+            .aarch64 => 6 * @sizeOf(u32),
+            else => unreachable,
+        };
+        try text_seg.addSection(self.allocator, .{
+            .sectname = makeStaticString("__stub_helper"),
+            .segname = makeStaticString("__TEXT"),
+            .addr = 0,
+            .size = stub_helper_size,
+            .offset = 0,
+            .@"align" = alignment,
+            .reloff = 0,
+            .nreloc = 0,
+            .flags = macho.S_REGULAR | macho.S_ATTR_PURE_INSTRUCTIONS | macho.S_ATTR_SOME_INSTRUCTIONS,
+            .reserved1 = 0,
+            .reserved2 = 0,
+            .reserved3 = 0,
+        });
+    }
+
+    if (self.data_const_segment_cmd_index == null) {
+        self.data_const_segment_cmd_index = @intCast(u16, self.load_commands.items.len);
+        try self.load_commands.append(self.allocator, .{
+            .Segment = SegmentCommand.empty(.{
+                .cmd = macho.LC_SEGMENT_64,
+                .cmdsize = @sizeOf(macho.segment_command_64),
+                .segname = makeStaticString("__DATA_CONST"),
+                .vmaddr = 0,
+                .vmsize = 0,
+                .fileoff = 0,
+                .filesize = 0,
+                .maxprot = macho.VM_PROT_READ | macho.VM_PROT_WRITE,
+                .initprot = macho.VM_PROT_READ | macho.VM_PROT_WRITE,
+                .nsects = 0,
+                .flags = 0,
+            }),
+        });
+    }
+
+    if (self.got_section_index == null) {
+        const data_const_seg = &self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+        self.got_section_index = @intCast(u16, data_const_seg.sections.items.len);
+        try data_const_seg.addSection(self.allocator, .{
+            .sectname = makeStaticString("__got"),
+            .segname = makeStaticString("__DATA_CONST"),
+            .addr = 0,
+            .size = 0,
+            .offset = 0,
+            .@"align" = 3, // 2^3 = @sizeOf(u64)
+            .reloff = 0,
+            .nreloc = 0,
+            .flags = macho.S_NON_LAZY_SYMBOL_POINTERS,
+            .reserved1 = 0,
+            .reserved2 = 0,
+            .reserved3 = 0,
+        });
+    }
+
+    if (self.data_segment_cmd_index == null) {
+        self.data_segment_cmd_index = @intCast(u16, self.load_commands.items.len);
+        try self.load_commands.append(self.allocator, .{
+            .Segment = SegmentCommand.empty(.{
+                .cmd = macho.LC_SEGMENT_64,
+                .cmdsize = @sizeOf(macho.segment_command_64),
+                .segname = makeStaticString("__DATA"),
+                .vmaddr = 0,
+                .vmsize = 0,
+                .fileoff = 0,
+                .filesize = 0,
+                .maxprot = macho.VM_PROT_READ | macho.VM_PROT_WRITE,
+                .initprot = macho.VM_PROT_READ | macho.VM_PROT_WRITE,
+                .nsects = 0,
+                .flags = 0,
+            }),
+        });
+    }
+
+    if (self.la_symbol_ptr_section_index == null) {
+        const data_seg = &self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+        self.la_symbol_ptr_section_index = @intCast(u16, data_seg.sections.items.len);
+        try data_seg.addSection(self.allocator, .{
+            .sectname = makeStaticString("__la_symbol_ptr"),
+            .segname = makeStaticString("__DATA"),
+            .addr = 0,
+            .size = 0,
+            .offset = 0,
+            .@"align" = 3, // 2^3 = @sizeOf(u64)
+            .reloff = 0,
+            .nreloc = 0,
+            .flags = macho.S_LAZY_SYMBOL_POINTERS,
+            .reserved1 = 0,
+            .reserved2 = 0,
+            .reserved3 = 0,
+        });
+    }
+
+    if (self.data_section_index == null) {
+        const data_seg = &self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+        self.data_section_index = @intCast(u16, data_seg.sections.items.len);
+        try data_seg.addSection(self.allocator, .{
+            .sectname = makeStaticString("__data"),
+            .segname = makeStaticString("__DATA"),
+            .addr = 0,
+            .size = 0,
+            .offset = 0,
+            .@"align" = 3, // 2^3 = @sizeOf(u64)
+            .reloff = 0,
+            .nreloc = 0,
+            .flags = macho.S_REGULAR,
+            .reserved1 = 0,
+            .reserved2 = 0,
+            .reserved3 = 0,
+        });
+    }
+
+    if (self.linkedit_segment_cmd_index == null) {
+        self.linkedit_segment_cmd_index = @intCast(u16, self.load_commands.items.len);
+        try self.load_commands.append(self.allocator, .{
+            .Segment = SegmentCommand.empty(.{
+                .cmd = macho.LC_SEGMENT_64,
+                .cmdsize = @sizeOf(macho.segment_command_64),
+                .segname = makeStaticString("__LINKEDIT"),
+                .vmaddr = 0,
+                .vmsize = 0,
+                .fileoff = 0,
+                .filesize = 0,
+                .maxprot = macho.VM_PROT_READ,
+                .initprot = macho.VM_PROT_READ,
+                .nsects = 0,
+                .flags = 0,
+            }),
+        });
+    }
+
+    if (self.dyld_info_cmd_index == null) {
+        self.dyld_info_cmd_index = @intCast(u16, self.load_commands.items.len);
+        try self.load_commands.append(self.allocator, .{
+            .DyldInfoOnly = .{
+                .cmd = macho.LC_DYLD_INFO_ONLY,
+                .cmdsize = @sizeOf(macho.dyld_info_command),
+                .rebase_off = 0,
+                .rebase_size = 0,
+                .bind_off = 0,
+                .bind_size = 0,
+                .weak_bind_off = 0,
+                .weak_bind_size = 0,
+                .lazy_bind_off = 0,
+                .lazy_bind_size = 0,
+                .export_off = 0,
+                .export_size = 0,
+            },
+        });
+    }
+
+    if (self.symtab_cmd_index == null) {
+        self.symtab_cmd_index = @intCast(u16, self.load_commands.items.len);
+        try self.load_commands.append(self.allocator, .{
+            .Symtab = .{
+                .cmd = macho.LC_SYMTAB,
+                .cmdsize = @sizeOf(macho.symtab_command),
+                .symoff = 0,
+                .nsyms = 0,
+                .stroff = 0,
+                .strsize = 0,
+            },
+        });
+        try self.strtab.append(self.allocator, 0);
+    }
+
+    if (self.dysymtab_cmd_index == null) {
+        self.dysymtab_cmd_index = @intCast(u16, self.load_commands.items.len);
+        try self.load_commands.append(self.allocator, .{
+            .Dysymtab = .{
+                .cmd = macho.LC_DYSYMTAB,
+                .cmdsize = @sizeOf(macho.dysymtab_command),
+                .ilocalsym = 0,
+                .nlocalsym = 0,
+                .iextdefsym = 0,
+                .nextdefsym = 0,
+                .iundefsym = 0,
+                .nundefsym = 0,
+                .tocoff = 0,
+                .ntoc = 0,
+                .modtaboff = 0,
+                .nmodtab = 0,
+                .extrefsymoff = 0,
+                .nextrefsyms = 0,
+                .indirectsymoff = 0,
+                .nindirectsyms = 0,
+                .extreloff = 0,
+                .nextrel = 0,
+                .locreloff = 0,
+                .nlocrel = 0,
+            },
+        });
+    }
+
+    if (self.dylinker_cmd_index == null) {
+        self.dylinker_cmd_index = @intCast(u16, self.load_commands.items.len);
+        const cmdsize = @intCast(u32, mem.alignForwardGeneric(
+            u64,
+            @sizeOf(macho.dylinker_command) + mem.lenZ(DEFAULT_DYLD_PATH),
+            @sizeOf(u64),
+        ));
+        var dylinker_cmd = emptyGenericCommandWithData(macho.dylinker_command{
+            .cmd = macho.LC_LOAD_DYLINKER,
+            .cmdsize = cmdsize,
+            .name = @sizeOf(macho.dylinker_command),
+        });
+        dylinker_cmd.data = try self.allocator.alloc(u8, cmdsize - dylinker_cmd.inner.name);
+        mem.set(u8, dylinker_cmd.data, 0);
+        mem.copy(u8, dylinker_cmd.data, mem.spanZ(DEFAULT_DYLD_PATH));
+        try self.load_commands.append(self.allocator, .{ .Dylinker = dylinker_cmd });
+    }
+
+    if (self.libsystem_cmd_index == null) {
+        self.libsystem_cmd_index = @intCast(u16, self.load_commands.items.len);
+        const cmdsize = @intCast(u32, mem.alignForwardGeneric(
+            u64,
+            @sizeOf(macho.dylib_command) + mem.lenZ(LIB_SYSTEM_PATH),
+            @sizeOf(u64),
+        ));
+        // TODO Find a way to work out runtime version from the OS version triple stored in std.Target.
+        // In the meantime, we're gonna hardcode to the minimum compatibility version of 0.0.0.
+        const min_version = 0x0;
+        var dylib_cmd = emptyGenericCommandWithData(macho.dylib_command{
+            .cmd = macho.LC_LOAD_DYLIB,
+            .cmdsize = cmdsize,
+            .dylib = .{
+                .name = @sizeOf(macho.dylib_command),
+                .timestamp = 2, // not sure why not simply 0; this is reverse engineered from Mach-O files
+                .current_version = min_version,
+                .compatibility_version = min_version,
+            },
+        });
+        dylib_cmd.data = try self.allocator.alloc(u8, cmdsize - dylib_cmd.inner.dylib.name);
+        mem.set(u8, dylib_cmd.data, 0);
+        mem.copy(u8, dylib_cmd.data, mem.spanZ(LIB_SYSTEM_PATH));
+        try self.load_commands.append(self.allocator, .{ .Dylib = dylib_cmd });
+    }
+
+    if (self.main_cmd_index == null) {
+        self.main_cmd_index = @intCast(u16, self.load_commands.items.len);
+        try self.load_commands.append(self.allocator, .{
+            .Main = .{
+                .cmd = macho.LC_MAIN,
+                .cmdsize = @sizeOf(macho.entry_point_command),
+                .entryoff = 0x0,
+                .stacksize = 0,
+            },
+        });
+    }
+
+    if (self.source_version_cmd_index == null) {
+        self.source_version_cmd_index = @intCast(u16, self.load_commands.items.len);
+        try self.load_commands.append(self.allocator, .{
+            .SourceVersion = .{
+                .cmd = macho.LC_SOURCE_VERSION,
+                .cmdsize = @sizeOf(macho.source_version_command),
+                .version = 0x0,
+            },
+        });
+    }
+
+    if (self.uuid_cmd_index == null) {
+        self.uuid_cmd_index = @intCast(u16, self.load_commands.items.len);
+        var uuid_cmd: macho.uuid_command = .{
+            .cmd = macho.LC_UUID,
+            .cmdsize = @sizeOf(macho.uuid_command),
+            .uuid = undefined,
+        };
+        std.crypto.random.bytes(&uuid_cmd.uuid);
+        try self.load_commands.append(self.allocator, .{ .Uuid = uuid_cmd });
+    }
+
+    if (self.code_signature_cmd_index == null and self.arch.? == .aarch64) {
+        self.code_signature_cmd_index = @intCast(u16, self.load_commands.items.len);
+        try self.load_commands.append(self.allocator, .{
+            .LinkeditData = .{
+                .cmd = macho.LC_CODE_SIGNATURE,
+                .cmdsize = @sizeOf(macho.linkedit_data_command),
+                .dataoff = 0,
+                .datasize = 0,
+            },
+        });
+    }
+
+    if (self.data_in_code_cmd_index == null and self.arch.? == .x86_64) {
+        self.data_in_code_cmd_index = @intCast(u16, self.load_commands.items.len);
+        try self.load_commands.append(self.allocator, .{
+            .LinkeditData = .{
+                .cmd = macho.LC_DATA_IN_CODE,
+                .cmdsize = @sizeOf(macho.linkedit_data_command),
+                .dataoff = 0,
+                .datasize = 0,
+            },
+        });
+    }
+}
+
+fn flush(self: *Zld) !void {
+    if (self.bss_section_index) |index| {
+        const seg = &self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+        const sect = &seg.sections.items[index];
+        sect.offset = 0;
+    }
+
+    if (self.tlv_bss_section_index) |index| {
+        const seg = &self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+        const sect = &seg.sections.items[index];
+        sect.offset = 0;
+    }
+
+    if (self.tlv_section_index) |index| {
+        const seg = self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+        const sect = &seg.sections.items[index];
+
+        var buffer = try self.allocator.alloc(u8, sect.size);
+        defer self.allocator.free(buffer);
+        _ = try self.file.?.preadAll(buffer, sect.offset);
+
+        var stream = std.io.fixedBufferStream(buffer);
+        var writer = stream.writer();
+
+        const seek_amt = 2 * @sizeOf(u64);
+        while (self.threadlocal_offsets.popOrNull()) |offset| {
+            try writer.context.seekBy(seek_amt);
+            try writer.writeIntLittle(u64, offset);
+        }
+
+        try self.file.?.pwriteAll(buffer, sect.offset);
+    }
+
+    try self.setEntryPoint();
+    try self.writeRebaseInfoTable();
+    try self.writeBindInfoTable();
+    try self.writeLazyBindInfoTable();
+    try self.writeExportInfo();
+    if (self.arch.? == .x86_64) {
+        try self.writeDataInCode();
+    }
+
+    {
+        const seg = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+        const symtab = &self.load_commands.items[self.symtab_cmd_index.?].Symtab;
+        symtab.symoff = @intCast(u32, seg.inner.fileoff + seg.inner.filesize);
+    }
+
+    try self.writeDebugInfo();
+    try self.writeSymbolTable();
+    try self.writeDynamicSymbolTable();
+    try self.writeStringTable();
+
+    {
+        // Seal __LINKEDIT size
+        const seg = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+        seg.inner.vmsize = mem.alignForwardGeneric(u64, seg.inner.filesize, self.page_size.?);
+    }
+
+    if (self.arch.? == .aarch64) {
+        try self.writeCodeSignaturePadding();
+    }
+
+    try self.writeLoadCommands();
+    try self.writeHeader();
+
+    if (self.arch.? == .aarch64) {
+        try self.writeCodeSignature();
+    }
+
+    if (comptime std.Target.current.isDarwin() and std.Target.current.cpu.arch == .aarch64) {
+        try fs.cwd().copyFile(self.out_path.?, fs.cwd(), self.out_path.?, .{});
+    }
+}
+
+fn setEntryPoint(self: *Zld) !void {
+    // TODO we should respect the -entry flag passed in by the user to set a custom
+    // entrypoint. For now, assume default of `_main`.
+    const seg = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    const text = seg.sections.items[self.text_section_index.?];
+    const entry_syms = self.locals.get("_main") orelse return error.MissingMainEntrypoint;
+
+    var entry_sym: ?macho.nlist_64 = null;
+    for (entry_syms.items) |es| {
+        switch (es.tt) {
+            .Global => {
+                entry_sym = es.inner;
+                break;
+            },
+            .WeakGlobal => {
+                entry_sym = es.inner;
+            },
+            .Local => {},
+        }
+    }
+    if (entry_sym == null) {
+        log.err("no (weak) global definition of _main found", .{});
+        return error.MissingMainEntrypoint;
+    }
+
+    const name = try self.allocator.dupe(u8, "_main");
+    try self.exports.putNoClobber(self.allocator, name, .{
+        .n_strx = entry_sym.?.n_strx,
+        .n_value = entry_sym.?.n_value,
+        .n_type = macho.N_SECT | macho.N_EXT,
+        .n_desc = entry_sym.?.n_desc,
+        .n_sect = entry_sym.?.n_sect,
+    });
+
+    const ec = &self.load_commands.items[self.main_cmd_index.?].Main;
+    ec.entryoff = @intCast(u32, entry_sym.?.n_value - seg.inner.vmaddr);
+}
+
+fn writeRebaseInfoTable(self: *Zld) !void {
+    var pointers = std.ArrayList(Pointer).init(self.allocator);
+    defer pointers.deinit();
+
+    try pointers.ensureCapacity(pointers.items.len + self.local_rebases.items.len);
+    pointers.appendSliceAssumeCapacity(self.local_rebases.items);
+
+    if (self.got_section_index) |idx| {
+        // TODO this should be cleaned up!
+        try pointers.ensureCapacity(pointers.items.len + self.nonlazy_pointers.items().len);
+        const seg = self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+        const sect = seg.sections.items[idx];
+        const base_offset = sect.addr - seg.inner.vmaddr;
+        const segment_id = @intCast(u16, self.data_const_segment_cmd_index.?);
+        const index_offset = @intCast(u32, self.nonlazy_imports.items().len);
+        for (self.nonlazy_pointers.items()) |entry| {
+            const index = index_offset + entry.value.index;
+            pointers.appendAssumeCapacity(.{
+                .offset = base_offset + index * @sizeOf(u64),
+                .segment_id = segment_id,
+            });
+        }
+    }
+
+    if (self.mod_init_func_section_index) |idx| {
+        // TODO audit and investigate this.
+        const seg = self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+        const sect = seg.sections.items[idx];
+        const npointers = sect.size * @sizeOf(u64);
+        const base_offset = sect.addr - seg.inner.vmaddr;
+        const segment_id = @intCast(u16, self.data_const_segment_cmd_index.?);
+
+        try pointers.ensureCapacity(pointers.items.len + npointers);
+        var i: usize = 0;
+        while (i < npointers) : (i += 1) {
+            pointers.appendAssumeCapacity(.{
+                .offset = base_offset + i * @sizeOf(u64),
+                .segment_id = segment_id,
+            });
+        }
+    }
+
+    if (self.mod_term_func_section_index) |idx| {
+        // TODO audit and investigate this.
+        const seg = self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+        const sect = seg.sections.items[idx];
+        const npointers = sect.size * @sizeOf(u64);
+        const base_offset = sect.addr - seg.inner.vmaddr;
+        const segment_id = @intCast(u16, self.data_const_segment_cmd_index.?);
+
+        try pointers.ensureCapacity(pointers.items.len + npointers);
+        var i: usize = 0;
+        while (i < npointers) : (i += 1) {
+            pointers.appendAssumeCapacity(.{
+                .offset = base_offset + i * @sizeOf(u64),
+                .segment_id = segment_id,
+            });
+        }
+    }
+
+    if (self.la_symbol_ptr_section_index) |idx| {
+        try pointers.ensureCapacity(pointers.items.len + self.lazy_imports.items().len);
+        const seg = self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+        const sect = seg.sections.items[idx];
+        const base_offset = sect.addr - seg.inner.vmaddr;
+        const segment_id = @intCast(u16, self.data_segment_cmd_index.?);
+        for (self.lazy_imports.items()) |entry| {
+            pointers.appendAssumeCapacity(.{
+                .offset = base_offset + entry.value.index * @sizeOf(u64),
+                .segment_id = segment_id,
+            });
+        }
+    }
+
+    std.sort.sort(Pointer, pointers.items, {}, pointerCmp);
+
+    const size = try rebaseInfoSize(pointers.items);
+    var buffer = try self.allocator.alloc(u8, @intCast(usize, size));
+    defer self.allocator.free(buffer);
+
+    var stream = std.io.fixedBufferStream(buffer);
+    try writeRebaseInfo(pointers.items, stream.writer());
+
+    const seg = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+    const dyld_info = &self.load_commands.items[self.dyld_info_cmd_index.?].DyldInfoOnly;
+    dyld_info.rebase_off = @intCast(u32, seg.inner.fileoff);
+    dyld_info.rebase_size = @intCast(u32, mem.alignForwardGeneric(u64, buffer.len, @sizeOf(u64)));
+    seg.inner.filesize += dyld_info.rebase_size;
+
+    log.debug("writing rebase info from 0x{x} to 0x{x}", .{ dyld_info.rebase_off, dyld_info.rebase_off + dyld_info.rebase_size });
+
+    try self.file.?.pwriteAll(buffer, dyld_info.rebase_off);
+}
+
+fn writeBindInfoTable(self: *Zld) !void {
+    var pointers = std.ArrayList(Pointer).init(self.allocator);
+    defer pointers.deinit();
+
+    if (self.got_section_index) |idx| {
+        try pointers.ensureCapacity(pointers.items.len + self.nonlazy_imports.items().len);
+        const seg = self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+        const sect = seg.sections.items[idx];
+        const base_offset = sect.addr - seg.inner.vmaddr;
+        const segment_id = @intCast(u16, self.data_const_segment_cmd_index.?);
+        for (self.nonlazy_imports.items()) |entry| {
+            pointers.appendAssumeCapacity(.{
+                .offset = base_offset + entry.value.index * @sizeOf(u64),
+                .segment_id = segment_id,
+                .dylib_ordinal = entry.value.dylib_ordinal,
+                .name = entry.key,
+            });
+        }
+    }
+
+    if (self.tlv_section_index) |idx| {
+        const seg = self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+        const sect = seg.sections.items[idx];
+        const base_offset = sect.addr - seg.inner.vmaddr;
+        const segment_id = @intCast(u16, self.data_segment_cmd_index.?);
+        try pointers.append(.{
+            .offset = base_offset + self.tlv_bootstrap.?.index * @sizeOf(u64),
+            .segment_id = segment_id,
+            .dylib_ordinal = self.tlv_bootstrap.?.dylib_ordinal,
+            .name = "__tlv_bootstrap",
+        });
+    }
+
+    const size = try bindInfoSize(pointers.items);
+    var buffer = try self.allocator.alloc(u8, @intCast(usize, size));
+    defer self.allocator.free(buffer);
+
+    var stream = std.io.fixedBufferStream(buffer);
+    try writeBindInfo(pointers.items, stream.writer());
+
+    const seg = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+    const dyld_info = &self.load_commands.items[self.dyld_info_cmd_index.?].DyldInfoOnly;
+    dyld_info.bind_off = @intCast(u32, seg.inner.fileoff + seg.inner.filesize);
+    dyld_info.bind_size = @intCast(u32, mem.alignForwardGeneric(u64, buffer.len, @alignOf(u64)));
+    seg.inner.filesize += dyld_info.bind_size;
+
+    log.debug("writing binding info from 0x{x} to 0x{x}", .{ dyld_info.bind_off, dyld_info.bind_off + dyld_info.bind_size });
+
+    try self.file.?.pwriteAll(buffer, dyld_info.bind_off);
+}
+
+fn writeLazyBindInfoTable(self: *Zld) !void {
+    var pointers = std.ArrayList(Pointer).init(self.allocator);
+    defer pointers.deinit();
+    try pointers.ensureCapacity(self.lazy_imports.items().len);
+
+    if (self.la_symbol_ptr_section_index) |idx| {
+        const seg = self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+        const sect = seg.sections.items[idx];
+        const base_offset = sect.addr - seg.inner.vmaddr;
+        const segment_id = @intCast(u16, self.data_segment_cmd_index.?);
+        for (self.lazy_imports.items()) |entry| {
+            pointers.appendAssumeCapacity(.{
+                .offset = base_offset + entry.value.index * @sizeOf(u64),
+                .segment_id = segment_id,
+                .dylib_ordinal = entry.value.dylib_ordinal,
+                .name = entry.key,
+            });
+        }
+    }
+
+    const size = try lazyBindInfoSize(pointers.items);
+    var buffer = try self.allocator.alloc(u8, @intCast(usize, size));
+    defer self.allocator.free(buffer);
+
+    var stream = std.io.fixedBufferStream(buffer);
+    try writeLazyBindInfo(pointers.items, stream.writer());
+
+    const seg = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+    const dyld_info = &self.load_commands.items[self.dyld_info_cmd_index.?].DyldInfoOnly;
+    dyld_info.lazy_bind_off = @intCast(u32, seg.inner.fileoff + seg.inner.filesize);
+    dyld_info.lazy_bind_size = @intCast(u32, mem.alignForwardGeneric(u64, buffer.len, @alignOf(u64)));
+    seg.inner.filesize += dyld_info.lazy_bind_size;
+
+    log.debug("writing lazy binding info from 0x{x} to 0x{x}", .{ dyld_info.lazy_bind_off, dyld_info.lazy_bind_off + dyld_info.lazy_bind_size });
+
+    try self.file.?.pwriteAll(buffer, dyld_info.lazy_bind_off);
+    try self.populateLazyBindOffsetsInStubHelper(buffer);
+}
+
+fn populateLazyBindOffsetsInStubHelper(self: *Zld, buffer: []const u8) !void {
+    var stream = std.io.fixedBufferStream(buffer);
+    var reader = stream.reader();
+    var offsets = std.ArrayList(u32).init(self.allocator);
+    try offsets.append(0);
+    defer offsets.deinit();
+    var valid_block = false;
+
+    while (true) {
+        const inst = reader.readByte() catch |err| switch (err) {
+            error.EndOfStream => break,
+            else => return err,
+        };
+        const imm: u8 = inst & macho.BIND_IMMEDIATE_MASK;
+        const opcode: u8 = inst & macho.BIND_OPCODE_MASK;
+
+        switch (opcode) {
+            macho.BIND_OPCODE_DO_BIND => {
+                valid_block = true;
+            },
+            macho.BIND_OPCODE_DONE => {
+                if (valid_block) {
+                    const offset = try stream.getPos();
+                    try offsets.append(@intCast(u32, offset));
+                }
+                valid_block = false;
+            },
+            macho.BIND_OPCODE_SET_SYMBOL_TRAILING_FLAGS_IMM => {
+                var next = try reader.readByte();
+                while (next != @as(u8, 0)) {
+                    next = try reader.readByte();
+                }
+            },
+            macho.BIND_OPCODE_SET_SEGMENT_AND_OFFSET_ULEB => {
+                _ = try leb.readULEB128(u64, reader);
+            },
+            macho.BIND_OPCODE_SET_DYLIB_ORDINAL_ULEB => {
+                _ = try leb.readULEB128(u64, reader);
+            },
+            macho.BIND_OPCODE_SET_ADDEND_SLEB => {
+                _ = try leb.readILEB128(i64, reader);
+            },
+            else => {},
+        }
+    }
+    assert(self.lazy_imports.items().len <= offsets.items.len);
+
+    const stub_size: u4 = switch (self.arch.?) {
+        .x86_64 => 10,
+        .aarch64 => 3 * @sizeOf(u32),
+        else => unreachable,
+    };
+    const off: u4 = switch (self.arch.?) {
+        .x86_64 => 1,
+        .aarch64 => 2 * @sizeOf(u32),
+        else => unreachable,
+    };
+    var buf: [@sizeOf(u32)]u8 = undefined;
+    for (self.lazy_imports.items()) |entry| {
+        const symbol = entry.value;
+        const placeholder_off = self.stub_helper_stubs_start_off.? + symbol.index * stub_size + off;
+        mem.writeIntLittle(u32, &buf, offsets.items[symbol.index]);
+        try self.file.?.pwriteAll(&buf, placeholder_off);
+    }
+}
+
+fn writeExportInfo(self: *Zld) !void {
+    var trie = Trie.init(self.allocator);
+    defer trie.deinit();
+
+    const text_segment = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    for (self.exports.items()) |entry| {
+        const name = entry.key;
+        const symbol = entry.value;
+        // TODO figure out if we should put all exports into the export trie
+        assert(symbol.n_value >= text_segment.inner.vmaddr);
+        try trie.put(.{
+            .name = name,
+            .vmaddr_offset = symbol.n_value - text_segment.inner.vmaddr,
+            .export_flags = macho.EXPORT_SYMBOL_FLAGS_KIND_REGULAR,
+        });
+    }
+
+    try trie.finalize();
+    var buffer = try self.allocator.alloc(u8, @intCast(usize, trie.size));
+    defer self.allocator.free(buffer);
+    var stream = std.io.fixedBufferStream(buffer);
+    const nwritten = try trie.write(stream.writer());
+    assert(nwritten == trie.size);
+
+    const seg = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+    const dyld_info = &self.load_commands.items[self.dyld_info_cmd_index.?].DyldInfoOnly;
+    dyld_info.export_off = @intCast(u32, seg.inner.fileoff + seg.inner.filesize);
+    dyld_info.export_size = @intCast(u32, mem.alignForwardGeneric(u64, buffer.len, @alignOf(u64)));
+    seg.inner.filesize += dyld_info.export_size;
+
+    log.debug("writing export info from 0x{x} to 0x{x}", .{ dyld_info.export_off, dyld_info.export_off + dyld_info.export_size });
+
+    try self.file.?.pwriteAll(buffer, dyld_info.export_off);
+}
+
+fn writeDebugInfo(self: *Zld) !void {
+    var stabs = std.ArrayList(macho.nlist_64).init(self.allocator);
+    defer stabs.deinit();
+
+    for (self.objects.items) |object, object_id| {
+        var debug_info = blk: {
+            var di = try DebugInfo.parseFromObject(self.allocator, object);
+            break :blk di orelse continue;
+        };
+        defer debug_info.deinit(self.allocator);
+
+        // We assume there is only one CU.
+        const compile_unit = debug_info.inner.findCompileUnit(0x0) catch |err| switch (err) {
+            error.MissingDebugInfo => {
+                // TODO audit cases with missing debug info and audit our dwarf.zig module.
+                log.debug("invalid or missing debug info in {s}; skipping", .{object.name});
+                continue;
+            },
+            else => |e| return e,
+        };
+        const name = try compile_unit.die.getAttrString(&debug_info.inner, dwarf.AT_name);
+        const comp_dir = try compile_unit.die.getAttrString(&debug_info.inner, dwarf.AT_comp_dir);
+
+        {
+            const tu_path = try std.fs.path.join(self.allocator, &[_][]const u8{ comp_dir, name });
+            defer self.allocator.free(tu_path);
+            const dirname = std.fs.path.dirname(tu_path) orelse "./";
+            // Current dir
+            try stabs.append(.{
+                .n_strx = try self.makeString(tu_path[0 .. dirname.len + 1]),
+                .n_type = macho.N_SO,
+                .n_sect = 0,
+                .n_desc = 0,
+                .n_value = 0,
+            });
+            // Artifact name
+            try stabs.append(.{
+                .n_strx = try self.makeString(tu_path[dirname.len + 1 ..]),
+                .n_type = macho.N_SO,
+                .n_sect = 0,
+                .n_desc = 0,
+                .n_value = 0,
+            });
+            // Path to object file with debug info
+            var buffer: [std.fs.MAX_PATH_BYTES]u8 = undefined;
+            const full_path = blk: {
+                if (object.ar_name) |prefix| {
+                    const path = try std.os.realpath(prefix, &buffer);
+                    break :blk try std.fmt.allocPrint(self.allocator, "{s}({s})", .{ path, object.name });
+                } else {
+                    const path = try std.os.realpath(object.name, &buffer);
+                    break :blk try mem.dupe(self.allocator, u8, path);
+                }
+            };
+            defer self.allocator.free(full_path);
+            const stat = try object.file.stat();
+            const mtime = @intCast(u64, @divFloor(stat.mtime, 1_000_000_000));
+            try stabs.append(.{
+                .n_strx = try self.makeString(full_path),
+                .n_type = macho.N_OSO,
+                .n_sect = 0,
+                .n_desc = 1,
+                .n_value = mtime,
+            });
+        }
+        log.debug("analyzing debug info in '{s}'", .{object.name});
+
+        for (object.symtab.items) |source_sym| {
+            const symname = object.getString(source_sym.n_strx);
+            const source_addr = source_sym.n_value;
+            const target_syms = self.locals.get(symname) orelse continue;
+            const target_sym: Symbol = blk: {
+                for (target_syms.items) |ts| {
+                    if (ts.object_id == @intCast(u16, object_id)) break :blk ts;
+                } else continue;
+            };
+
+            const maybe_size = blk: for (debug_info.inner.func_list.items) |func| {
+                if (func.pc_range) |range| {
+                    if (source_addr >= range.start and source_addr < range.end) {
+                        break :blk range.end - range.start;
+                    }
+                }
+            } else null;
+
+            if (maybe_size) |size| {
+                try stabs.append(.{
+                    .n_strx = 0,
+                    .n_type = macho.N_BNSYM,
+                    .n_sect = target_sym.inner.n_sect,
+                    .n_desc = 0,
+                    .n_value = target_sym.inner.n_value,
+                });
+                try stabs.append(.{
+                    .n_strx = target_sym.inner.n_strx,
+                    .n_type = macho.N_FUN,
+                    .n_sect = target_sym.inner.n_sect,
+                    .n_desc = 0,
+                    .n_value = target_sym.inner.n_value,
+                });
+                try stabs.append(.{
+                    .n_strx = 0,
+                    .n_type = macho.N_FUN,
+                    .n_sect = 0,
+                    .n_desc = 0,
+                    .n_value = size,
+                });
+                try stabs.append(.{
+                    .n_strx = 0,
+                    .n_type = macho.N_ENSYM,
+                    .n_sect = target_sym.inner.n_sect,
+                    .n_desc = 0,
+                    .n_value = size,
+                });
+            } else {
+                // TODO need a way to differentiate symbols: global, static, local, etc.
+                try stabs.append(.{
+                    .n_strx = target_sym.inner.n_strx,
+                    .n_type = macho.N_STSYM,
+                    .n_sect = target_sym.inner.n_sect,
+                    .n_desc = 0,
+                    .n_value = target_sym.inner.n_value,
+                });
+            }
+        }
+
+        // Close the source file!
+        try stabs.append(.{
+            .n_strx = 0,
+            .n_type = macho.N_SO,
+            .n_sect = 0,
+            .n_desc = 0,
+            .n_value = 0,
+        });
+    }
+
+    if (stabs.items.len == 0) return;
+
+    // Write stabs into the symbol table
+    const linkedit = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+    const symtab = &self.load_commands.items[self.symtab_cmd_index.?].Symtab;
+
+    symtab.nsyms = @intCast(u32, stabs.items.len);
+
+    const stabs_off = symtab.symoff;
+    const stabs_size = symtab.nsyms * @sizeOf(macho.nlist_64);
+    log.debug("writing symbol stabs from 0x{x} to 0x{x}", .{ stabs_off, stabs_size + stabs_off });
+    try self.file.?.pwriteAll(mem.sliceAsBytes(stabs.items), stabs_off);
+
+    linkedit.inner.filesize += stabs_size;
+
+    // Update dynamic symbol table.
+    const dysymtab = &self.load_commands.items[self.dysymtab_cmd_index.?].Dysymtab;
+    dysymtab.nlocalsym = symtab.nsyms;
+}
+
+fn writeSymbolTable(self: *Zld) !void {
+    const seg = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+    const symtab = &self.load_commands.items[self.symtab_cmd_index.?].Symtab;
+
+    var locals = std.ArrayList(macho.nlist_64).init(self.allocator);
+    defer locals.deinit();
+
+    for (self.locals.items()) |entries| {
+        log.debug("'{s}': {} entries", .{ entries.key, entries.value.items.len });
+        // var symbol: ?macho.nlist_64 = null;
+        for (entries.value.items) |entry| {
+            log.debug("    | {}", .{entry.inner});
+            log.debug("    | {}", .{entry.tt});
+            log.debug("    | {s}", .{self.objects.items[entry.object_id].name});
+            try locals.append(entry.inner);
+        }
+    }
+    const nlocals = locals.items.len;
+
+    const nexports = self.exports.items().len;
+    var exports = std.ArrayList(macho.nlist_64).init(self.allocator);
+    defer exports.deinit();
+
+    try exports.ensureCapacity(nexports);
+    for (self.exports.items()) |entry| {
+        exports.appendAssumeCapacity(entry.value);
+    }
+
+    const has_tlv: bool = self.tlv_bootstrap != null;
+
+    var nundefs = self.lazy_imports.items().len + self.nonlazy_imports.items().len;
+    if (has_tlv) nundefs += 1;
+
+    var undefs = std.ArrayList(macho.nlist_64).init(self.allocator);
+    defer undefs.deinit();
+
+    try undefs.ensureCapacity(nundefs);
+    for (self.lazy_imports.items()) |entry| {
+        undefs.appendAssumeCapacity(entry.value.symbol);
+    }
+    for (self.nonlazy_imports.items()) |entry| {
+        undefs.appendAssumeCapacity(entry.value.symbol);
+    }
+    if (has_tlv) {
+        undefs.appendAssumeCapacity(self.tlv_bootstrap.?.symbol);
+    }
+
+    const locals_off = symtab.symoff + symtab.nsyms * @sizeOf(macho.nlist_64);
+    const locals_size = nlocals * @sizeOf(macho.nlist_64);
+    log.debug("writing local symbols from 0x{x} to 0x{x}", .{ locals_off, locals_size + locals_off });
+    try self.file.?.pwriteAll(mem.sliceAsBytes(locals.items), locals_off);
+
+    const exports_off = locals_off + locals_size;
+    const exports_size = nexports * @sizeOf(macho.nlist_64);
+    log.debug("writing exported symbols from 0x{x} to 0x{x}", .{ exports_off, exports_size + exports_off });
+    try self.file.?.pwriteAll(mem.sliceAsBytes(exports.items), exports_off);
+
+    const undefs_off = exports_off + exports_size;
+    const undefs_size = nundefs * @sizeOf(macho.nlist_64);
+    log.debug("writing undefined symbols from 0x{x} to 0x{x}", .{ undefs_off, undefs_size + undefs_off });
+    try self.file.?.pwriteAll(mem.sliceAsBytes(undefs.items), undefs_off);
+
+    symtab.nsyms += @intCast(u32, nlocals + nexports + nundefs);
+    seg.inner.filesize += locals_size + exports_size + undefs_size;
+
+    // Update dynamic symbol table.
+    const dysymtab = &self.load_commands.items[self.dysymtab_cmd_index.?].Dysymtab;
+    dysymtab.nlocalsym += @intCast(u32, nlocals);
+    dysymtab.iextdefsym = dysymtab.nlocalsym;
+    dysymtab.nextdefsym = @intCast(u32, nexports);
+    dysymtab.iundefsym = dysymtab.nlocalsym + dysymtab.nextdefsym;
+    dysymtab.nundefsym = @intCast(u32, nundefs);
+}
+
+fn writeDynamicSymbolTable(self: *Zld) !void {
+    const seg = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+    const text_segment = &self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    const stubs = &text_segment.sections.items[self.stubs_section_index.?];
+    const data_const_segment = &self.load_commands.items[self.data_const_segment_cmd_index.?].Segment;
+    const got = &data_const_segment.sections.items[self.got_section_index.?];
+    const data_segment = &self.load_commands.items[self.data_segment_cmd_index.?].Segment;
+    const la_symbol_ptr = &data_segment.sections.items[self.la_symbol_ptr_section_index.?];
+    const dysymtab = &self.load_commands.items[self.dysymtab_cmd_index.?].Dysymtab;
+
+    const lazy = self.lazy_imports.items();
+    const nonlazy = self.nonlazy_imports.items();
+    const got_locals = self.nonlazy_pointers.items();
+    dysymtab.indirectsymoff = @intCast(u32, seg.inner.fileoff + seg.inner.filesize);
+    dysymtab.nindirectsyms = @intCast(u32, lazy.len * 2 + nonlazy.len + got_locals.len);
+    const needed_size = dysymtab.nindirectsyms * @sizeOf(u32);
+    seg.inner.filesize += needed_size;
+
+    log.debug("writing indirect symbol table from 0x{x} to 0x{x}", .{
+        dysymtab.indirectsymoff,
+        dysymtab.indirectsymoff + needed_size,
+    });
+
+    var buf = try self.allocator.alloc(u8, needed_size);
+    defer self.allocator.free(buf);
+    var stream = std.io.fixedBufferStream(buf);
+    var writer = stream.writer();
+
+    stubs.reserved1 = 0;
+    for (lazy) |_, i| {
+        const symtab_idx = @intCast(u32, dysymtab.iundefsym + i);
+        try writer.writeIntLittle(u32, symtab_idx);
+    }
+
+    const base_id = @intCast(u32, lazy.len);
+    got.reserved1 = base_id;
+    for (nonlazy) |_, i| {
+        const symtab_idx = @intCast(u32, dysymtab.iundefsym + i + base_id);
+        try writer.writeIntLittle(u32, symtab_idx);
+    }
+    // TODO there should be one common set of GOT entries.
+    for (got_locals) |_| {
+        try writer.writeIntLittle(u32, macho.INDIRECT_SYMBOL_LOCAL);
+    }
+
+    la_symbol_ptr.reserved1 = got.reserved1 + @intCast(u32, nonlazy.len) + @intCast(u32, got_locals.len);
+    for (lazy) |_, i| {
+        const symtab_idx = @intCast(u32, dysymtab.iundefsym + i);
+        try writer.writeIntLittle(u32, symtab_idx);
+    }
+
+    try self.file.?.pwriteAll(buf, dysymtab.indirectsymoff);
+}
+
+fn writeStringTable(self: *Zld) !void {
+    const seg = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+    const symtab = &self.load_commands.items[self.symtab_cmd_index.?].Symtab;
+    symtab.stroff = @intCast(u32, seg.inner.fileoff + seg.inner.filesize);
+    symtab.strsize = @intCast(u32, mem.alignForwardGeneric(u64, self.strtab.items.len, @alignOf(u64)));
+    seg.inner.filesize += symtab.strsize;
+
+    log.debug("writing string table from 0x{x} to 0x{x}", .{ symtab.stroff, symtab.stroff + symtab.strsize });
+
+    try self.file.?.pwriteAll(self.strtab.items, symtab.stroff);
+
+    if (symtab.strsize > self.strtab.items.len and self.arch.? == .x86_64) {
+        // This is the last section, so we need to pad it out.
+        try self.file.?.pwriteAll(&[_]u8{0}, seg.inner.fileoff + seg.inner.filesize - 1);
+    }
+}
+
+fn writeDataInCode(self: *Zld) !void {
+    const seg = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+    const dice_cmd = &self.load_commands.items[self.data_in_code_cmd_index.?].LinkeditData;
+    const fileoff = seg.inner.fileoff + seg.inner.filesize;
+
+    var buf = std.ArrayList(u8).init(self.allocator);
+    defer buf.deinit();
+
+    const text_seg = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    const text_sect = text_seg.sections.items[self.text_section_index.?];
+    for (self.objects.items) |object, object_id| {
+        const source_seg = object.load_commands.items[object.segment_cmd_index.?].Segment;
+        const source_sect = source_seg.sections.items[object.text_section_index.?];
+        const target_mapping = self.mappings.get(.{
+            .object_id = @intCast(u16, object_id),
+            .source_sect_id = object.text_section_index.?,
+        }) orelse continue;
+
+        try buf.ensureCapacity(
+            buf.items.len + object.data_in_code_entries.items.len * @sizeOf(macho.data_in_code_entry),
+        );
+        for (object.data_in_code_entries.items) |dice| {
+            const new_dice: macho.data_in_code_entry = .{
+                .offset = text_sect.offset + target_mapping.offset + dice.offset,
+                .length = dice.length,
+                .kind = dice.kind,
+            };
+            buf.appendSliceAssumeCapacity(mem.asBytes(&new_dice));
+        }
+    }
+    const datasize = @intCast(u32, buf.items.len);
+
+    dice_cmd.dataoff = @intCast(u32, fileoff);
+    dice_cmd.datasize = datasize;
+    seg.inner.filesize += datasize;
+
+    log.debug("writing data-in-code from 0x{x} to 0x{x}", .{ fileoff, fileoff + datasize });
+
+    try self.file.?.pwriteAll(buf.items, fileoff);
+}
+
+fn writeCodeSignaturePadding(self: *Zld) !void {
+    const seg = &self.load_commands.items[self.linkedit_segment_cmd_index.?].Segment;
+    const code_sig_cmd = &self.load_commands.items[self.code_signature_cmd_index.?].LinkeditData;
+    const fileoff = seg.inner.fileoff + seg.inner.filesize;
+    const needed_size = CodeSignature.calcCodeSignaturePaddingSize(
+        self.out_path.?,
+        fileoff,
+        self.page_size.?,
+    );
+    code_sig_cmd.dataoff = @intCast(u32, fileoff);
+    code_sig_cmd.datasize = needed_size;
+
+    // Advance size of __LINKEDIT segment
+    seg.inner.filesize += needed_size;
+    seg.inner.vmsize = mem.alignForwardGeneric(u64, seg.inner.filesize, self.page_size.?);
+
+    log.debug("writing code signature padding from 0x{x} to 0x{x}", .{ fileoff, fileoff + needed_size });
+
+    // Pad out the space. We need to do this to calculate valid hashes for everything in the file
+    // except for code signature data.
+    try self.file.?.pwriteAll(&[_]u8{0}, fileoff + needed_size - 1);
+}
+
+fn writeCodeSignature(self: *Zld) !void {
+    const text_seg = self.load_commands.items[self.text_segment_cmd_index.?].Segment;
+    const code_sig_cmd = self.load_commands.items[self.code_signature_cmd_index.?].LinkeditData;
+
+    var code_sig = CodeSignature.init(self.allocator, self.page_size.?);
+    defer code_sig.deinit();
+    try code_sig.calcAdhocSignature(
+        self.file.?,
+        self.out_path.?,
+        text_seg.inner,
+        code_sig_cmd,
+        .Exe,
+    );
+
+    var buffer = try self.allocator.alloc(u8, code_sig.size());
+    defer self.allocator.free(buffer);
+    var stream = std.io.fixedBufferStream(buffer);
+    try code_sig.write(stream.writer());
+
+    log.debug("writing code signature from 0x{x} to 0x{x}", .{ code_sig_cmd.dataoff, code_sig_cmd.dataoff + buffer.len });
+
+    try self.file.?.pwriteAll(buffer, code_sig_cmd.dataoff);
+}
+
+fn writeLoadCommands(self: *Zld) !void {
+    var sizeofcmds: u32 = 0;
+    for (self.load_commands.items) |lc| {
+        sizeofcmds += lc.cmdsize();
+    }
+
+    var buffer = try self.allocator.alloc(u8, sizeofcmds);
+    defer self.allocator.free(buffer);
+    var writer = std.io.fixedBufferStream(buffer).writer();
+    for (self.load_commands.items) |lc| {
+        try lc.write(writer);
+    }
+
+    const off = @sizeOf(macho.mach_header_64);
+    log.debug("writing {} load commands from 0x{x} to 0x{x}", .{ self.load_commands.items.len, off, off + sizeofcmds });
+    try self.file.?.pwriteAll(buffer, off);
+}
+
+fn writeHeader(self: *Zld) !void {
+    var header: macho.mach_header_64 = undefined;
+    header.magic = macho.MH_MAGIC_64;
+
+    const CpuInfo = struct {
+        cpu_type: macho.cpu_type_t,
+        cpu_subtype: macho.cpu_subtype_t,
+    };
+
+    const cpu_info: CpuInfo = switch (self.arch.?) {
+        .aarch64 => .{
+            .cpu_type = macho.CPU_TYPE_ARM64,
+            .cpu_subtype = macho.CPU_SUBTYPE_ARM_ALL,
+        },
+        .x86_64 => .{
+            .cpu_type = macho.CPU_TYPE_X86_64,
+            .cpu_subtype = macho.CPU_SUBTYPE_X86_64_ALL,
+        },
+        else => return error.UnsupportedCpuArchitecture,
+    };
+    header.cputype = cpu_info.cpu_type;
+    header.cpusubtype = cpu_info.cpu_subtype;
+    header.filetype = macho.MH_EXECUTE;
+    header.flags = macho.MH_NOUNDEFS | macho.MH_DYLDLINK | macho.MH_PIE | macho.MH_TWOLEVEL;
+    header.reserved = 0;
+
+    if (self.tlv_section_index) |_|
+        header.flags |= macho.MH_HAS_TLV_DESCRIPTORS;
+
+    header.ncmds = @intCast(u32, self.load_commands.items.len);
+    header.sizeofcmds = 0;
+    for (self.load_commands.items) |cmd| {
+        header.sizeofcmds += cmd.cmdsize();
+    }
+    log.debug("writing Mach-O header {}", .{header});
+    try self.file.?.pwriteAll(mem.asBytes(&header), 0);
+}
+
+pub fn makeStaticString(bytes: []const u8) [16]u8 {
+    var buf = [_]u8{0} ** 16;
+    assert(bytes.len <= buf.len);
+    mem.copy(u8, &buf, bytes);
+    return buf;
+}
+
+fn makeString(self: *Zld, bytes: []const u8) !u32 {
+    try self.strtab.ensureCapacity(self.allocator, self.strtab.items.len + bytes.len + 1);
+    const offset = @intCast(u32, self.strtab.items.len);
+    log.debug("writing new string '{s}' into string table at offset 0x{x}", .{ bytes, offset });
+    self.strtab.appendSliceAssumeCapacity(bytes);
+    self.strtab.appendAssumeCapacity(0);
+    return offset;
+}
+
+fn getString(self: *const Zld, str_off: u32) []const u8 {
+    assert(str_off < self.strtab.items.len);
+    return mem.spanZ(@ptrCast([*:0]const u8, self.strtab.items.ptr + str_off));
+}
+
+pub fn parseName(name: *const [16]u8) []const u8 {
+    const len = mem.indexOfScalar(u8, name, @as(u8, 0)) orelse name.len;
+    return name[0..len];
+}
+
+fn isLocal(sym: *const macho.nlist_64) callconv(.Inline) bool {
+    if (isExtern(sym)) return false;
+    const tt = macho.N_TYPE & sym.n_type;
+    return tt == macho.N_SECT;
+}
+
+fn isExport(sym: *const macho.nlist_64) callconv(.Inline) bool {
+    if (!isExtern(sym)) return false;
+    const tt = macho.N_TYPE & sym.n_type;
+    return tt == macho.N_SECT;
+}
+
+fn isImport(sym: *const macho.nlist_64) callconv(.Inline) bool {
+    if (!isExtern(sym)) return false;
+    const tt = macho.N_TYPE & sym.n_type;
+    return tt == macho.N_UNDF;
+}
+
+fn isExtern(sym: *const macho.nlist_64) callconv(.Inline) bool {
+    if ((sym.n_type & macho.N_EXT) == 0) return false;
+    return (sym.n_type & macho.N_PEXT) == 0;
+}
+
+fn isWeakDef(sym: *const macho.nlist_64) callconv(.Inline) bool {
+    return (sym.n_desc & macho.N_WEAK_DEF) != 0;
+}
+
+fn aarch64IsArithmetic(inst: *const [4]u8) callconv(.Inline) bool {
+    const group_decode = @truncate(u5, inst[3]);
+    return ((group_decode >> 2) == 4);
+}
diff --git a/src/link/MachO/bind.zig b/src/link/MachO/bind.zig
new file mode 100644
index 0000000000..d234fa8242
--- /dev/null
+++ b/src/link/MachO/bind.zig
@@ -0,0 +1,145 @@
+const std = @import("std");
+const leb = std.leb;
+const macho = std.macho;
+
+pub const Pointer = struct {
+    offset: u64,
+    segment_id: u16,
+    dylib_ordinal: ?i64 = null,
+    name: ?[]const u8 = null,
+};
+
+pub fn pointerCmp(context: void, a: Pointer, b: Pointer) bool {
+    if (a.segment_id < b.segment_id) return true;
+    if (a.segment_id == b.segment_id) {
+        return a.offset < b.offset;
+    }
+    return false;
+}
+
+pub fn rebaseInfoSize(pointers: []const Pointer) !u64 {
+    var stream = std.io.countingWriter(std.io.null_writer);
+    var writer = stream.writer();
+    var size: u64 = 0;
+
+    for (pointers) |pointer| {
+        size += 2;
+        try leb.writeILEB128(writer, pointer.offset);
+        size += 1;
+    }
+
+    size += 1 + stream.bytes_written;
+    return size;
+}
+
+pub fn writeRebaseInfo(pointers: []const Pointer, writer: anytype) !void {
+    for (pointers) |pointer| {
+        try writer.writeByte(macho.REBASE_OPCODE_SET_TYPE_IMM | @truncate(u4, macho.REBASE_TYPE_POINTER));
+        try writer.writeByte(macho.REBASE_OPCODE_SET_SEGMENT_AND_OFFSET_ULEB | @truncate(u4, pointer.segment_id));
+
+        try leb.writeILEB128(writer, pointer.offset);
+        try writer.writeByte(macho.REBASE_OPCODE_DO_REBASE_IMM_TIMES | @truncate(u4, 1));
+    }
+    try writer.writeByte(macho.REBASE_OPCODE_DONE);
+}
+
+pub fn bindInfoSize(pointers: []const Pointer) !u64 {
+    var stream = std.io.countingWriter(std.io.null_writer);
+    var writer = stream.writer();
+    var size: u64 = 0;
+
+    for (pointers) |pointer| {
+        size += 1;
+        if (pointer.dylib_ordinal.? > 15) {
+            try leb.writeULEB128(writer, @bitCast(u64, pointer.dylib_ordinal.?));
+        }
+        size += 1;
+
+        size += 1;
+        size += pointer.name.?.len;
+        size += 1;
+
+        size += 1;
+
+        try leb.writeILEB128(writer, pointer.offset);
+        size += 1;
+    }
+
+    size += stream.bytes_written + 1;
+    return size;
+}
+
+pub fn writeBindInfo(pointers: []const Pointer, writer: anytype) !void {
+    for (pointers) |pointer| {
+        if (pointer.dylib_ordinal.? > 15) {
+            try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_ORDINAL_ULEB);
+            try leb.writeULEB128(writer, @bitCast(u64, pointer.dylib_ordinal.?));
+        } else if (pointer.dylib_ordinal.? > 0) {
+            try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_ORDINAL_IMM | @truncate(u4, @bitCast(u64, pointer.dylib_ordinal.?)));
+        } else {
+            try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_SPECIAL_IMM | @truncate(u4, @bitCast(u64, pointer.dylib_ordinal.?)));
+        }
+        try writer.writeByte(macho.BIND_OPCODE_SET_TYPE_IMM | @truncate(u4, macho.BIND_TYPE_POINTER));
+
+        try writer.writeByte(macho.BIND_OPCODE_SET_SYMBOL_TRAILING_FLAGS_IMM); // TODO Sometimes we might want to add flags.
+        try writer.writeAll(pointer.name.?);
+        try writer.writeByte(0);
+
+        try writer.writeByte(macho.BIND_OPCODE_SET_SEGMENT_AND_OFFSET_ULEB | @truncate(u4, pointer.segment_id));
+
+        try leb.writeILEB128(writer, pointer.offset);
+        try writer.writeByte(macho.BIND_OPCODE_DO_BIND);
+    }
+
+    try writer.writeByte(macho.BIND_OPCODE_DONE);
+}
+
+pub fn lazyBindInfoSize(pointers: []const Pointer) !u64 {
+    var stream = std.io.countingWriter(std.io.null_writer);
+    var writer = stream.writer();
+    var size: u64 = 0;
+
+    for (pointers) |pointer| {
+        size += 1;
+
+        try leb.writeILEB128(writer, pointer.offset);
+
+        size += 1;
+        if (pointer.dylib_ordinal.? > 15) {
+            try leb.writeULEB128(writer, @bitCast(u64, pointer.dylib_ordinal.?));
+        }
+
+        size += 1;
+        size += pointer.name.?.len;
+        size += 1;
+
+        size += 2;
+    }
+
+    size += stream.bytes_written;
+    return size;
+}
+
+pub fn writeLazyBindInfo(pointers: []const Pointer, writer: anytype) !void {
+    for (pointers) |pointer| {
+        try writer.writeByte(macho.BIND_OPCODE_SET_SEGMENT_AND_OFFSET_ULEB | @truncate(u4, pointer.segment_id));
+
+        try leb.writeILEB128(writer, pointer.offset);
+
+        if (pointer.dylib_ordinal.? > 15) {
+            try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_ORDINAL_ULEB);
+            try leb.writeULEB128(writer, @bitCast(u64, pointer.dylib_ordinal.?));
+        } else if (pointer.dylib_ordinal.? > 0) {
+            try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_ORDINAL_IMM | @truncate(u4, @bitCast(u64, pointer.dylib_ordinal.?)));
+        } else {
+            try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_SPECIAL_IMM | @truncate(u4, @bitCast(u64, pointer.dylib_ordinal.?)));
+        }
+
+        try writer.writeByte(macho.BIND_OPCODE_SET_SYMBOL_TRAILING_FLAGS_IMM); // TODO Sometimes we might want to add flags.
+        try writer.writeAll(pointer.name.?);
+        try writer.writeByte(0);
+
+        try writer.writeByte(macho.BIND_OPCODE_DO_BIND);
+        try writer.writeByte(macho.BIND_OPCODE_DONE);
+    }
+}
diff --git a/src/link/MachO/imports.zig b/src/link/MachO/imports.zig
deleted file mode 100644
index 2bc34d14c5..0000000000
--- a/src/link/MachO/imports.zig
+++ /dev/null
@@ -1,152 +0,0 @@
-const std = @import("std");
-const leb = std.leb;
-const macho = std.macho;
-const mem = std.mem;
-
-const assert = std.debug.assert;
-const Allocator = mem.Allocator;
-
-pub const ExternSymbol = struct {
-    /// MachO symbol table entry.
-    inner: macho.nlist_64,
-
-    /// Id of the dynamic library where the specified entries can be found.
-    /// Id of 0 means self.
-    /// TODO this should really be an id into the table of all defined
-    /// dylibs.
-    dylib_ordinal: i64 = 0,
-
-    /// Id of the segment where this symbol is defined (will have its address
-    /// resolved).
-    segment: u16 = 0,
-
-    /// Offset relative to the start address of the `segment`.
-    offset: u32 = 0,
-};
-
-pub fn rebaseInfoSize(symbols: anytype) !u64 {
-    var stream = std.io.countingWriter(std.io.null_writer);
-    var writer = stream.writer();
-    var size: u64 = 0;
-
-    for (symbols) |entry| {
-        size += 2;
-        try leb.writeILEB128(writer, entry.value.offset);
-        size += 1;
-    }
-
-    size += 1 + stream.bytes_written;
-    return size;
-}
-
-pub fn writeRebaseInfo(symbols: anytype, writer: anytype) !void {
-    for (symbols) |entry| {
-        const symbol = entry.value;
-        try writer.writeByte(macho.REBASE_OPCODE_SET_TYPE_IMM | @truncate(u4, macho.REBASE_TYPE_POINTER));
-        try writer.writeByte(macho.REBASE_OPCODE_SET_SEGMENT_AND_OFFSET_ULEB | @truncate(u4, symbol.segment));
-        try leb.writeILEB128(writer, symbol.offset);
-        try writer.writeByte(macho.REBASE_OPCODE_DO_REBASE_IMM_TIMES | @truncate(u4, 1));
-    }
-    try writer.writeByte(macho.REBASE_OPCODE_DONE);
-}
-
-pub fn bindInfoSize(symbols: anytype) !u64 {
-    var stream = std.io.countingWriter(std.io.null_writer);
-    var writer = stream.writer();
-    var size: u64 = 0;
-
-    for (symbols) |entry| {
-        const symbol = entry.value;
-
-        size += 1;
-        if (symbol.dylib_ordinal > 15) {
-            try leb.writeULEB128(writer, @bitCast(u64, symbol.dylib_ordinal));
-        }
-        size += 1;
-
-        size += 1;
-        size += entry.key.len;
-        size += 1;
-
-        size += 1;
-        try leb.writeILEB128(writer, symbol.offset);
-        size += 2;
-    }
-
-    size += stream.bytes_written;
-    return size;
-}
-
-pub fn writeBindInfo(symbols: anytype, writer: anytype) !void {
-    for (symbols) |entry| {
-        const symbol = entry.value;
-
-        if (symbol.dylib_ordinal > 15) {
-            try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_ORDINAL_ULEB);
-            try leb.writeULEB128(writer, @bitCast(u64, symbol.dylib_ordinal));
-        } else if (symbol.dylib_ordinal > 0) {
-            try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_ORDINAL_IMM | @truncate(u4, @bitCast(u64, symbol.dylib_ordinal)));
-        } else {
-            try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_SPECIAL_IMM | @truncate(u4, @bitCast(u64, symbol.dylib_ordinal)));
-        }
-        try writer.writeByte(macho.BIND_OPCODE_SET_TYPE_IMM | @truncate(u4, macho.BIND_TYPE_POINTER));
-
-        try writer.writeByte(macho.BIND_OPCODE_SET_SYMBOL_TRAILING_FLAGS_IMM); // TODO Sometimes we might want to add flags.
-        try writer.writeAll(entry.key);
-        try writer.writeByte(0);
-
-        try writer.writeByte(macho.BIND_OPCODE_SET_SEGMENT_AND_OFFSET_ULEB | @truncate(u4, symbol.segment));
-        try leb.writeILEB128(writer, symbol.offset);
-        try writer.writeByte(macho.BIND_OPCODE_DO_BIND);
-        try writer.writeByte(macho.BIND_OPCODE_DONE);
-    }
-}
-
-pub fn lazyBindInfoSize(symbols: anytype) !u64 {
-    var stream = std.io.countingWriter(std.io.null_writer);
-    var writer = stream.writer();
-    var size: u64 = 0;
-
-    for (symbols) |entry| {
-        const symbol = entry.value;
-        size += 1;
-        try leb.writeILEB128(writer, symbol.offset);
-        size += 1;
-        if (symbol.dylib_ordinal > 15) {
-            try leb.writeULEB128(writer, @bitCast(u64, symbol.dylib_ordinal));
-        }
-
-        size += 1;
-        size += entry.key.len;
-        size += 1;
-
-        size += 2;
-    }
-
-    size += stream.bytes_written;
-    return size;
-}
-
-pub fn writeLazyBindInfo(symbols: anytype, writer: anytype) !void {
-    for (symbols) |entry| {
-        const symbol = entry.value;
-        try writer.writeByte(macho.BIND_OPCODE_SET_SEGMENT_AND_OFFSET_ULEB | @truncate(u4, symbol.segment));
-        try leb.writeILEB128(writer, symbol.offset);
-
-        if (symbol.dylib_ordinal > 15) {
-            try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_ORDINAL_ULEB);
-            try leb.writeULEB128(writer, @bitCast(u64, symbol.dylib_ordinal));
-        } else if (symbol.dylib_ordinal > 0) {
-            try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_ORDINAL_IMM | @truncate(u4, @bitCast(u64, symbol.dylib_ordinal)));
-        } else {
-            try writer.writeByte(macho.BIND_OPCODE_SET_DYLIB_SPECIAL_IMM | @truncate(u4, @bitCast(u64, symbol.dylib_ordinal)));
-        }
-
-        try writer.writeByte(macho.BIND_OPCODE_SET_SYMBOL_TRAILING_FLAGS_IMM); // TODO Sometimes we might want to add flags.
-        try writer.writeAll(entry.key);
-        try writer.writeByte(0);
-
-        try writer.writeByte(macho.BIND_OPCODE_DO_BIND);
-        try writer.writeByte(macho.BIND_OPCODE_DONE);
-    }
-}