Elf2: create a new linker from scratch

This iteration already has significantly better incremental support.

Closes #24110

CMakeLists.txt

@@ -583,6 +583,7 @@ set(ZIG_STAGE2_SOURCES
     src/link/Elf/relocatable.zig
     src/link/Elf/relocation.zig
     src/link/Elf/synthetic_sections.zig
+    src/link/Elf2.zig
     src/link/Goff.zig
     src/link/LdScript.zig
     src/link/Lld.zig
@@ -612,6 +613,7 @@ set(ZIG_STAGE2_SOURCES
     src/link/MachO/synthetic.zig
     src/link/MachO/Thunk.zig
     src/link/MachO/uuid.zig
+    src/link/MappedFile.zig
     src/link/Queue.zig
     src/link/StringTable.zig
     src/link/Wasm.zig

build.zig

@@ -202,6 +202,7 @@ pub fn build(b: *std.Build) !void {
     });
     exe.pie = pie;
     exe.entitlements = entitlements;
+    exe.use_new_linker = b.option(bool, "new-linker", "Use the new linker");
 
     const use_llvm = b.option(bool, "use-llvm", "Use the llvm backend");
     exe.use_llvm = use_llvm;

lib/compiler_rt/fma.zig

@@ -203,7 +203,7 @@ fn add_adjusted(a: f64, b: f64) f64 {
         if (uhii & 1 == 0) {
             // hibits += copysign(1.0, sum.hi, sum.lo)
             const uloi: u64 = @bitCast(sum.lo);
-            uhii += 1 - ((uhii ^ uloi) >> 62);
+            uhii = uhii + 1 - ((uhii ^ uloi) >> 62);
             sum.hi = @bitCast(uhii);
         }
     }
@@ -217,7 +217,7 @@ fn add_and_denorm(a: f64, b: f64, scale: i32) f64 {
         const bits_lost = -@as(i32, @intCast((uhii >> 52) & 0x7FF)) - scale + 1;
         if ((bits_lost != 1) == (uhii & 1 != 0)) {
             const uloi: u64 = @bitCast(sum.lo);
-            uhii += 1 - (((uhii ^ uloi) >> 62) & 2);
+            uhii = uhii + 1 - (((uhii ^ uloi) >> 62) & 2);
             sum.hi = @bitCast(uhii);
         }
     }
@@ -259,7 +259,7 @@ fn add_adjusted128(a: f128, b: f128) f128 {
         if (uhii & 1 == 0) {
             // hibits += copysign(1.0, sum.hi, sum.lo)
             const uloi: u128 = @bitCast(sum.lo);
-            uhii += 1 - ((uhii ^ uloi) >> 126);
+            uhii = uhii + 1 - ((uhii ^ uloi) >> 126);
             sum.hi = @bitCast(uhii);
         }
     }
@@ -284,7 +284,7 @@ fn add_and_denorm128(a: f128, b: f128, scale: i32) f128 {
         const bits_lost = -@as(i32, @intCast((uhii >> 112) & 0x7FFF)) - scale + 1;
         if ((bits_lost != 1) == (uhii & 1 != 0)) {
             const uloi: u128 = @bitCast(sum.lo);
-            uhii += 1 - (((uhii ^ uloi) >> 126) & 2);
+            uhii = uhii + 1 - (((uhii ^ uloi) >> 126) & 2);
             sum.hi = @bitCast(uhii);
         }
     }

lib/std/Build/Step/Compile.zig

@@ -192,6 +192,7 @@ want_lto: ?bool = null,
 
 use_llvm: ?bool,
 use_lld: ?bool,
+use_new_linker: ?bool,
 
 /// Corresponds to the `-fallow-so-scripts` / `-fno-allow-so-scripts` CLI
 /// flags, overriding the global user setting provided to the `zig build`
@@ -441,6 +442,7 @@ pub fn create(owner: *std.Build, options: Options) *Compile {
 
         .use_llvm = options.use_llvm,
         .use_lld = options.use_lld,
+        .use_new_linker = null,
 
         .zig_process = null,
     };
@@ -1096,6 +1098,7 @@ fn getZigArgs(compile: *Compile, fuzz: bool) ![][]const u8 {
 
     try addFlag(&zig_args, "llvm", compile.use_llvm);
     try addFlag(&zig_args, "lld", compile.use_lld);
+    try addFlag(&zig_args, "new-linker", compile.use_new_linker);
 
     if (compile.root_module.resolved_target.?.query.ofmt) |ofmt| {
         try zig_args.append(try std.fmt.allocPrint(arena, "-ofmt={s}", .{@tagName(ofmt)}));

lib/std/elf.zig

@@ -323,6 +323,8 @@ pub const PT_LOPROC = 0x70000000;
 /// End of processor-specific
 pub const PT_HIPROC = 0x7fffffff;
 
+pub const PN_XNUM = 0xffff;
+
 /// Section header table entry unused
 pub const SHT_NULL = 0;
 /// Program data
@@ -385,63 +387,149 @@ pub const SHT_HIUSER = 0xffffffff;
 // Note type for .note.gnu.build_id
 pub const NT_GNU_BUILD_ID = 3;
 
+/// Deprecated, use `@intFromEnum(std.elf.STB.LOCAL)`
+pub const STB_LOCAL = @intFromEnum(STB.LOCAL);
+/// Deprecated, use `@intFromEnum(std.elf.STB.GLOBAL)`
+pub const STB_GLOBAL = @intFromEnum(STB.GLOBAL);
+/// Deprecated, use `@intFromEnum(std.elf.STB.WEAK)`
+pub const STB_WEAK = @intFromEnum(STB.WEAK);
+/// Deprecated, use `std.elf.STB.NUM`
+pub const STB_NUM = STB.NUM;
+/// Deprecated, use `@intFromEnum(std.elf.STB.LOOS)`
+pub const STB_LOOS = @intFromEnum(STB.LOOS);
+/// Deprecated, use `@intFromEnum(std.elf.STB.GNU_UNIQUE)`
+pub const STB_GNU_UNIQUE = @intFromEnum(STB.GNU_UNIQUE);
+/// Deprecated, use `@intFromEnum(std.elf.STB.HIOS)`
+pub const STB_HIOS = @intFromEnum(STB.HIOS);
+/// Deprecated, use `@intFromEnum(std.elf.STB.LOPROC)`
+pub const STB_LOPROC = @intFromEnum(STB.LOPROC);
+/// Deprecated, use `@intFromEnum(std.elf.STB.HIPROC)`
+pub const STB_HIPROC = @intFromEnum(STB.HIPROC);
+
+/// Deprecated, use `@intFromEnum(std.elf.STB.MIPS_SPLIT_COMMON)`
+pub const STB_MIPS_SPLIT_COMMON = @intFromEnum(STB.MIBS_SPLIT_COMMON);
+
+/// Deprecated, use `@intFromEnum(std.elf.STT.NOTYPE)`
+pub const STT_NOTYPE = @intFromEnum(STT.NOTYPE);
+/// Deprecated, use `@intFromEnum(std.elf.STT.OBJECT)`
+pub const STT_OBJECT = @intFromEnum(STT.OBJECT);
+/// Deprecated, use `@intFromEnum(std.elf.STT.FUNC)`
+pub const STT_FUNC = @intFromEnum(STT.FUNC);
+/// Deprecated, use `@intFromEnum(std.elf.STT.SECTION)`
+pub const STT_SECTION = @intFromEnum(STT.SECTION);
+/// Deprecated, use `@intFromEnum(std.elf.STT.FILE)`
+pub const STT_FILE = @intFromEnum(STT.FILE);
+/// Deprecated, use `@intFromEnum(std.elf.STT.COMMON)`
+pub const STT_COMMON = @intFromEnum(STT.COMMON);
+/// Deprecated, use `@intFromEnum(std.elf.STT.TLS)`
+pub const STT_TLS = @intFromEnum(STT.TLS);
+/// Deprecated, use `std.elf.STT.NUM`
+pub const STT_NUM = STT.NUM;
+/// Deprecated, use `@intFromEnum(std.elf.STT.LOOS)`
+pub const STT_LOOS = @intFromEnum(STT.LOOS);
+/// Deprecated, use `@intFromEnum(std.elf.STT.GNU_IFUNC)`
+pub const STT_GNU_IFUNC = @intFromEnum(STT.GNU_IFUNC);
+/// Deprecated, use `@intFromEnum(std.elf.STT.HIOS)`
+pub const STT_HIOS = @intFromEnum(STT.HIOS);
+/// Deprecated, use `@intFromEnum(std.elf.STT.LOPROC)`
+pub const STT_LOPROC = @intFromEnum(STT.LOPROC);
+/// Deprecated, use `@intFromEnum(std.elf.STT.HIPROC)`
+pub const STT_HIPROC = @intFromEnum(STT.HIPROC);
+
+/// Deprecated, use `@intFromEnum(std.elf.STT.SPARC_REGISTER)`
+pub const STT_SPARC_REGISTER = @intFromEnum(STT.SPARC_REGISTER);
+
+/// Deprecated, use `@intFromEnum(std.elf.STT.PARISC_MILLICODE)`
+pub const STT_PARISC_MILLICODE = @intFromEnum(STT.PARISC_MILLICODE);
+
+/// Deprecated, use `@intFromEnum(std.elf.STT.HP_OPAQUE)`
+pub const STT_HP_OPAQUE = @intFromEnum(STT.HP_OPAQUE);
+/// Deprecated, use `@intFromEnum(std.elf.STT.HP_STUB)`
+pub const STT_HP_STUB = @intFromEnum(STT.HP_STUB);
+
+/// Deprecated, use `@intFromEnum(std.elf.STT.ARM_TFUNC)`
+pub const STT_ARM_TFUNC = @intFromEnum(STT.ARM_TFUNC);
+/// Deprecated, use `@intFromEnum(std.elf.STT.ARM_16BIT)`
+pub const STT_ARM_16BIT = @intFromEnum(STT.ARM_16BIT);
+
+pub const STB = enum(u4) {
     /// Local symbol
-pub const STB_LOCAL = 0;
+    LOCAL = 0,
     /// Global symbol
-pub const STB_GLOBAL = 1;
+    GLOBAL = 1,
     /// Weak symbol
-pub const STB_WEAK = 2;
+    WEAK = 2,
+    _,
+
     /// Number of defined types
-pub const STB_NUM = 3;
+    pub const NUM = @typeInfo(STB).@"enum".fields.len;
+
     /// Start of OS-specific
-pub const STB_LOOS = 10;
+    pub const LOOS: STB = @enumFromInt(10);
+    /// End of OS-specific
+    pub const HIOS: STB = @enumFromInt(12);
+
     /// Unique symbol
-pub const STB_GNU_UNIQUE = 10;
+    pub const GNU_UNIQUE: STB = @enumFromInt(@intFromEnum(LOOS) + 0);
-/// End of OS-specific
+
-pub const STB_HIOS = 12;
     /// Start of processor-specific
-pub const STB_LOPROC = 13;
+    pub const LOPROC: STB = @enumFromInt(13);
     /// End of processor-specific
-pub const STB_HIPROC = 15;
+    pub const HIPROC: STB = @enumFromInt(15);
 
-pub const STB_MIPS_SPLIT_COMMON = 13;
+    pub const MIPS_SPLIT_COMMON: STB = @enumFromInt(@intFromEnum(LOPROC) + 0);
+};
 
+pub const STT = enum(u4) {
     /// Symbol type is unspecified
-pub const STT_NOTYPE = 0;
+    NOTYPE = 0,
     /// Symbol is a data object
-pub const STT_OBJECT = 1;
+    OBJECT = 1,
     /// Symbol is a code object
-pub const STT_FUNC = 2;
+    FUNC = 2,
     /// Symbol associated with a section
-pub const STT_SECTION = 3;
+    SECTION = 3,
     /// Symbol's name is file name
-pub const STT_FILE = 4;
+    FILE = 4,
     /// Symbol is a common data object
-pub const STT_COMMON = 5;
+    COMMON = 5,
     /// Symbol is thread-local data object
-pub const STT_TLS = 6;
+    TLS = 6,
+    _,
+
     /// Number of defined types
-pub const STT_NUM = 7;
+    pub const NUM = @typeInfo(STT).@"enum".fields.len;
+
     /// Start of OS-specific
-pub const STT_LOOS = 10;
+    pub const LOOS: STT = @enumFromInt(10);
-/// Symbol is indirect code object
-pub const STT_GNU_IFUNC = 10;
     /// End of OS-specific
-pub const STT_HIOS = 12;
+    pub const HIOS: STT = @enumFromInt(12);
+
+    /// Symbol is indirect code object
+    pub const GNU_IFUNC: STT = @enumFromInt(@intFromEnum(LOOS) + 0);
+
+    pub const HP_OPAQUE: STT = @enumFromInt(@intFromEnum(LOOS) + 1);
+    pub const HP_STUB: STT = @enumFromInt(@intFromEnum(LOOS) + 2);
+
     /// Start of processor-specific
-pub const STT_LOPROC = 13;
+    pub const LOPROC: STT = @enumFromInt(13);
     /// End of processor-specific
-pub const STT_HIPROC = 15;
+    pub const HIPROC: STT = @enumFromInt(15);
 
-pub const STT_SPARC_REGISTER = 13;
+    pub const SPARC_REGISTER: STT = @enumFromInt(@intFromEnum(LOPROC) + 0);
 
-pub const STT_PARISC_MILLICODE = 13;
+    pub const PARISC_MILLICODE: STT = @enumFromInt(@intFromEnum(LOPROC) + 0);
 
-pub const STT_HP_OPAQUE = (STT_LOOS + 0x1);
+    pub const ARM_TFUNC: STT = @enumFromInt(@intFromEnum(LOPROC) + 0);
-pub const STT_HP_STUB = (STT_LOOS + 0x2);
+    pub const ARM_16BIT: STT = @enumFromInt(@intFromEnum(HIPROC) + 2);
+};
 
-pub const STT_ARM_TFUNC = STT_LOPROC;
+pub const STV = enum(u3) {
-pub const STT_ARM_16BIT = STT_HIPROC;
+    DEFAULT = 0,
+    INTERNAL = 1,
+    HIDDEN = 2,
+    PROTECTED = 3,
+};
 
 pub const MAGIC = "\x7fELF";
 
@@ -534,15 +622,15 @@ pub const Header = struct {
         const buf = try r.peek(@sizeOf(Elf64_Ehdr));
 
         if (!mem.eql(u8, buf[0..4], MAGIC)) return error.InvalidElfMagic;
-        if (buf[EI_VERSION] != 1) return error.InvalidElfVersion;
+        if (buf[EI.VERSION] != 1) return error.InvalidElfVersion;
 
-        const endian: std.builtin.Endian = switch (buf[EI_DATA]) {
+        const endian: std.builtin.Endian = switch (buf[EI.DATA]) {
             ELFDATA2LSB => .little,
             ELFDATA2MSB => .big,
             else => return error.InvalidElfEndian,
         };
 
-        return switch (buf[EI_CLASS]) {
+        return switch (buf[EI.CLASS]) {
             ELFCLASS32 => .init(try r.takeStruct(Elf32_Ehdr, endian), endian),
             ELFCLASS64 => .init(try r.takeStruct(Elf64_Ehdr, endian), endian),
             else => return error.InvalidElfClass,
@@ -559,8 +647,8 @@ pub const Header = struct {
                 else => @compileError("bad type"),
             },
             .endian = endian,
-            .os_abi = @enumFromInt(hdr.e_ident[EI_OSABI]),
+            .os_abi = @enumFromInt(hdr.e_ident[EI.OSABI]),
-            .abi_version = hdr.e_ident[EI_ABIVERSION],
+            .abi_version = hdr.e_ident[EI.ABIVERSION],
             .type = hdr.e_type,
             .machine = hdr.e_machine,
             .entry = hdr.e_entry,
@@ -683,38 +771,200 @@ fn takeShdr(reader: *std.Io.Reader, elf_header: Header) !?Elf64_Shdr {
     };
 }
 
-pub const ELFCLASSNONE = 0;
+pub const EI = struct {
-pub const ELFCLASS32 = 1;
+    pub const CLASS = 4;
-pub const ELFCLASS64 = 2;
+    pub const DATA = 5;
-pub const ELFCLASSNUM = 3;
+    pub const VERSION = 6;
+    pub const OSABI = 7;
+    pub const ABIVERSION = 8;
+    pub const PAD = 9;
+    pub const NIDENT = 16;
+};
 
-pub const ELFDATANONE = 0;
+/// Deprecated, use `std.elf.EI.CLASS`
-pub const ELFDATA2LSB = 1;
+pub const EI_CLASS = EI.CLASS;
-pub const ELFDATA2MSB = 2;
+/// Deprecated, use `std.elf.EI.DATA`
-pub const ELFDATANUM = 3;
+pub const EI_DATA = EI.DATA;
-
+/// Deprecated, use `std.elf.EI.VERSION`
-pub const EI_CLASS = 4;
+pub const EI_VERSION = EI.VERSION;
-pub const EI_DATA = 5;
+/// Deprecated, use `std.elf.EI.OSABI`
-pub const EI_VERSION = 6;
+pub const EI_OSABI = EI.OSABI;
-pub const EI_OSABI = 7;
+/// Deprecated, use `std.elf.EI.ABIVERSION`
-pub const EI_ABIVERSION = 8;
+pub const EI_ABIVERSION = EI.ABIVERSION;
-pub const EI_PAD = 9;
+/// Deprecated, use `std.elf.EI.PAD`
-
+pub const EI_PAD = EI.PAD;
-pub const EI_NIDENT = 16;
+/// Deprecated, use `std.elf.EI.NIDENT`
+pub const EI_NIDENT = EI.NIDENT;
 
 pub const Half = u16;
 pub const Word = u32;
 pub const Sword = i32;
-pub const Elf32_Xword = u64;
+pub const Xword = u64;
-pub const Elf32_Sxword = i64;
+pub const Sxword = i64;
-pub const Elf64_Xword = u64;
+pub const Section = u16;
+pub const Elf32 = struct {
+    pub const Addr = u32;
+    pub const Off = u32;
+    pub const Ehdr = extern struct {
+        ident: [EI.NIDENT]u8,
+        type: ET,
+        machine: EM,
+        version: Word,
+        entry: Elf32.Addr,
+        phoff: Elf32.Off,
+        shoff: Elf32.Off,
+        flags: Word,
+        ehsize: Half,
+        phentsize: Half,
+        phnum: Half,
+        shentsize: Half,
+        shnum: Half,
+        shstrndx: Half,
+    };
+    pub const Phdr = extern struct {
+        type: Word,
+        offset: Elf32.Off,
+        vaddr: Elf32.Addr,
+        paddr: Elf32.Addr,
+        filesz: Word,
+        memsz: Word,
+        flags: PF,
+        @"align": Word,
+    };
+    pub const Shdr = extern struct {
+        name: Word,
+        type: Word,
+        flags: packed struct { shf: SHF },
+        addr: Elf32.Addr,
+        offset: Elf32.Off,
+        size: Word,
+        link: Word,
+        info: Word,
+        addralign: Word,
+        entsize: Word,
+    };
+    pub const Chdr = extern struct {
+        type: COMPRESS,
+        size: Word,
+        addralign: Word,
+    };
+    pub const Sym = extern struct {
+        name: Word,
+        value: Elf32.Addr,
+        size: Word,
+        info: Info,
+        other: Other,
+        shndx: Section,
+
+        pub const Info = packed struct(u8) {
+            type: STT,
+            bind: STB,
+        };
+
+        pub const Other = packed struct(u8) {
+            visibility: STV,
+            unused: u5 = 0,
+        };
+    };
+    comptime {
+        assert(@sizeOf(Elf32.Ehdr) == 52);
+        assert(@sizeOf(Elf32.Phdr) == 32);
+        assert(@sizeOf(Elf32.Shdr) == 40);
+        assert(@sizeOf(Elf32.Sym) == 16);
+    }
+};
+pub const Elf64 = struct {
+    pub const Addr = u64;
+    pub const Off = u64;
+    pub const Ehdr = extern struct {
+        ident: [EI.NIDENT]u8,
+        type: ET,
+        machine: EM,
+        version: Word,
+        entry: Elf64.Addr,
+        phoff: Elf64.Off,
+        shoff: Elf64.Off,
+        flags: Word,
+        ehsize: Half,
+        phentsize: Half,
+        phnum: Half,
+        shentsize: Half,
+        shnum: Half,
+        shstrndx: Half,
+    };
+    pub const Phdr = extern struct {
+        type: Word,
+        flags: PF,
+        offset: Elf64.Off,
+        vaddr: Elf64.Addr,
+        paddr: Elf64.Addr,
+        filesz: Xword,
+        memsz: Xword,
+        @"align": Xword,
+    };
+    pub const Shdr = extern struct {
+        name: Word,
+        type: Word,
+        flags: packed struct { shf: SHF, unused: Word = 0 },
+        addr: Elf64.Addr,
+        offset: Elf64.Off,
+        size: Xword,
+        link: Word,
+        info: Word,
+        addralign: Xword,
+        entsize: Xword,
+    };
+    pub const Chdr = extern struct {
+        type: COMPRESS,
+        reserved: Word = 0,
+        size: Xword,
+        addralign: Xword,
+    };
+    pub const Sym = extern struct {
+        name: Word,
+        info: Info,
+        other: Other,
+        shndx: Section,
+        value: Elf64.Addr,
+        size: Xword,
+
+        pub const Info = Elf32.Sym.Info;
+        pub const Other = Elf32.Sym.Other;
+    };
+    comptime {
+        assert(@sizeOf(Elf64.Ehdr) == 64);
+        assert(@sizeOf(Elf64.Phdr) == 56);
+        assert(@sizeOf(Elf64.Shdr) == 64);
+        assert(@sizeOf(Elf64.Sym) == 24);
+    }
+};
+pub const ElfN = switch (@sizeOf(usize)) {
+    4 => Elf32,
+    8 => Elf64,
+    else => @compileError("expected pointer size of 32 or 64"),
+};
+
+/// Deprecated, use `std.elf.Xword`
+pub const Elf32_Xword = Xword;
+/// Deprecated, use `std.elf.Sxword`
+pub const Elf32_Sxword = Sxword;
+/// Deprecated, use `std.elf.Xword`
+pub const Elf64_Xword = Xword;
+/// Deprecated, use `std.elf.Sxword`
 pub const Elf64_Sxword = i64;
+/// Deprecated, use `std.elf.Elf32.Addr`
 pub const Elf32_Addr = u32;
+/// Deprecated, use `std.elf.Elf64.Addr`
 pub const Elf64_Addr = u64;
+/// Deprecated, use `std.elf.Elf32.Off`
 pub const Elf32_Off = u32;
+/// Deprecated, use `std.elf.Elf64.Off`
 pub const Elf64_Off = u64;
+/// Deprecated, use `std.elf.Section`
 pub const Elf32_Section = u16;
+/// Deprecated, use `std.elf.Section`
 pub const Elf64_Section = u16;
+/// Deprecated, use `std.elf.Elf32.Ehdr`
 pub const Elf32_Ehdr = extern struct {
     e_ident: [EI_NIDENT]u8,
     e_type: ET,
@@ -731,8 +981,9 @@ pub const Elf32_Ehdr = extern struct {
     e_shnum: Half,
     e_shstrndx: Half,
 };
+/// Deprecated, use `std.elf.Elf64.Ehdr`
 pub const Elf64_Ehdr = extern struct {
-    e_ident: [EI_NIDENT]u8,
+    e_ident: [EI.NIDENT]u8,
     e_type: ET,
     e_machine: EM,
     e_version: Word,
@@ -747,6 +998,7 @@ pub const Elf64_Ehdr = extern struct {
     e_shnum: Half,
     e_shstrndx: Half,
 };
+/// Deprecated, use `std.elf.Elf32.Phdr`
 pub const Elf32_Phdr = extern struct {
     p_type: Word,
     p_offset: Elf32_Off,
@@ -757,6 +1009,7 @@ pub const Elf32_Phdr = extern struct {
     p_flags: Word,
     p_align: Word,
 };
+/// Deprecated, use `std.elf.Elf64.Phdr`
 pub const Elf64_Phdr = extern struct {
     p_type: Word,
     p_flags: Word,
@@ -767,6 +1020,7 @@ pub const Elf64_Phdr = extern struct {
     p_memsz: Elf64_Xword,
     p_align: Elf64_Xword,
 };
+/// Deprecated, use `std.elf.Elf32.Shdr`
 pub const Elf32_Shdr = extern struct {
     sh_name: Word,
     sh_type: Word,
@@ -779,6 +1033,7 @@ pub const Elf32_Shdr = extern struct {
     sh_addralign: Word,
     sh_entsize: Word,
 };
+/// Deprecated, use `std.elf.Elf64.Shdr`
 pub const Elf64_Shdr = extern struct {
     sh_name: Word,
     sh_type: Word,
@@ -791,17 +1046,20 @@ pub const Elf64_Shdr = extern struct {
     sh_addralign: Elf64_Xword,
     sh_entsize: Elf64_Xword,
 };
+/// Deprecated, use `std.elf.Elf32.Chdr`
 pub const Elf32_Chdr = extern struct {
     ch_type: COMPRESS,
     ch_size: Word,
     ch_addralign: Word,
 };
+/// Deprecated, use `std.elf.Elf64.Chdr`
 pub const Elf64_Chdr = extern struct {
     ch_type: COMPRESS,
     ch_reserved: Word = 0,
     ch_size: Elf64_Xword,
     ch_addralign: Elf64_Xword,
 };
+/// Deprecated, use `std.elf.Elf32.Sym`
 pub const Elf32_Sym = extern struct {
     st_name: Word,
     st_value: Elf32_Addr,
@@ -817,6 +1075,7 @@ pub const Elf32_Sym = extern struct {
         return @truncate(self.st_info >> 4);
     }
 };
+/// Deprecated, use `std.elf.Elf64.Sym`
 pub const Elf64_Sym = extern struct {
     st_name: Word,
     st_info: u8,
@@ -1020,27 +1279,18 @@ pub const Elf_MIPS_ABIFlags_v0 = extern struct {
     flags2: Word,
 };
 
-comptime {
-    assert(@sizeOf(Elf32_Ehdr) == 52);
-    assert(@sizeOf(Elf64_Ehdr) == 64);
-
-    assert(@sizeOf(Elf32_Phdr) == 32);
-    assert(@sizeOf(Elf64_Phdr) == 56);
-
-    assert(@sizeOf(Elf32_Shdr) == 40);
-    assert(@sizeOf(Elf64_Shdr) == 64);
-}
-
 pub const Auxv = switch (@sizeOf(usize)) {
     4 => Elf32_auxv_t,
     8 => Elf64_auxv_t,
     else => @compileError("expected pointer size of 32 or 64"),
 };
+/// Deprecated, use `std.elf.ElfN.Ehdr`
 pub const Ehdr = switch (@sizeOf(usize)) {
     4 => Elf32_Ehdr,
     8 => Elf64_Ehdr,
     else => @compileError("expected pointer size of 32 or 64"),
 };
+/// Deprecated, use `std.elf.ElfN.Phdr`
 pub const Phdr = switch (@sizeOf(usize)) {
     4 => Elf32_Phdr,
     8 => Elf64_Phdr,
@@ -1071,20 +1321,53 @@ pub const Shdr = switch (@sizeOf(usize)) {
     8 => Elf64_Shdr,
     else => @compileError("expected pointer size of 32 or 64"),
 };
+/// Deprecated, use `std.elf.ElfN.Chdr`
 pub const Chdr = switch (@sizeOf(usize)) {
     4 => Elf32_Chdr,
     8 => Elf64_Chdr,
     else => @compileError("expected pointer size of 32 or 64"),
 };
+/// Deprecated, use `std.elf.ElfN.Sym`
 pub const Sym = switch (@sizeOf(usize)) {
     4 => Elf32_Sym,
     8 => Elf64_Sym,
     else => @compileError("expected pointer size of 32 or 64"),
 };
-pub const Addr = switch (@sizeOf(usize)) {
+/// Deprecated, use `std.elf.ElfN.Addr`
-    4 => Elf32_Addr,
+pub const Addr = ElfN.Addr;
-    8 => Elf64_Addr,
+
-    else => @compileError("expected pointer size of 32 or 64"),
+/// Deprecated, use `@intFromEnum(std.elf.CLASS.NONE)`
+pub const ELFCLASSNONE = @intFromEnum(CLASS.NONE);
+/// Deprecated, use `@intFromEnum(std.elf.CLASS.@"32")`
+pub const ELFCLASS32 = @intFromEnum(CLASS.@"32");
+/// Deprecated, use `@intFromEnum(std.elf.CLASS.@"64")`
+pub const ELFCLASS64 = @intFromEnum(CLASS.@"64");
+/// Deprecated, use `@intFromEnum(std.elf.CLASS.NUM)`
+pub const ELFCLASSNUM = CLASS.NUM;
+pub const CLASS = enum(u8) {
+    NONE = 0,
+    @"32" = 1,
+    @"64" = 2,
+    _,
+
+    pub const NUM = @typeInfo(CLASS).@"enum".fields.len;
+};
+
+/// Deprecated, use `@intFromEnum(std.elf.DATA.NONE)`
+pub const ELFDATANONE = @intFromEnum(DATA.NONE);
+/// Deprecated, use `@intFromEnum(std.elf.DATA.@"2LSB")`
+pub const ELFDATA2LSB = @intFromEnum(DATA.@"2LSB");
+/// Deprecated, use `@intFromEnum(std.elf.DATA.@"2MSB")`
+pub const ELFDATA2MSB = @intFromEnum(DATA.@"2MSB");
+/// Deprecated, use `@intFromEnum(std.elf.DATA.NUM)`
+pub const ELFDATANUM = DATA.NUM;
+pub const DATA = enum(u8) {
+    NONE = 0,
+    @"2LSB" = 1,
+    @"2MSB" = 2,
+    _,
+
+    pub const NUM = @typeInfo(DATA).@"enum".fields.len;
 };
 
 pub const OSABI = enum(u8) {
@@ -1718,6 +2001,108 @@ pub const SHF_MIPS_STRING = 0x80000000;
 /// Make code section unreadable when in execute-only mode
 pub const SHF_ARM_PURECODE = 0x2000000;
 
+pub const SHF = packed struct(Word) {
+    /// Section data should be writable during execution.
+    WRITE: bool = false,
+    /// Section occupies memory during program execution.
+    ALLOC: bool = false,
+    /// Section contains executable machine instructions.
+    EXECINSTR: bool = false,
+    unused3: u1 = 0,
+    /// The data in this section may be merged.
+    MERGE: bool = false,
+    /// The data in this section is null-terminated strings.
+    STRINGS: bool = false,
+    /// A field in this section holds a section header table index.
+    INFO_LINK: bool = false,
+    /// Adds special ordering requirements for link editors.
+    LINK_ORDER: bool = false,
+    /// This section requires special OS-specific processing to avoid incorrect behavior.
+    OS_NONCONFORMING: bool = false,
+    /// This section is a member of a section group.
+    GROUP: bool = false,
+    /// This section holds Thread-Local Storage.
+    TLS: bool = false,
+    /// Identifies a section containing compressed data.
+    COMPRESSED: bool = false,
+    unused12: u8 = 0,
+    OS: packed union {
+        MASK: u8,
+        GNU: packed struct(u8) {
+            unused0: u1 = 0,
+            /// Not to be GCed by the linker
+            RETAIN: bool = false,
+            unused2: u6 = 0,
+        },
+        MIPS: packed struct(u8) {
+            unused0: u4 = 0,
+            /// Section contains text/data which may be replicated in other sections.
+            /// Linker must retain only one copy.
+            NODUPES: bool = false,
+            /// Linker must generate implicit hidden weak names.
+            NAMES: bool = false,
+            /// Section data local to process.
+            LOCAL: bool = false,
+            /// Do not strip this section.
+            NOSTRIP: bool = false,
+        },
+        ARM: packed struct(u8) {
+            unused0: u5 = 0,
+            /// Make code section unreadable when in execute-only mode
+            PURECODE: bool = false,
+            unused6: u2 = 0,
+        },
+    } = .{ .MASK = 0 },
+    PROC: packed union {
+        MASK: u4,
+        XCORE: packed struct(u4) {
+            /// All sections with the "d" flag are grouped together by the linker to form
+            /// the data section and the dp register is set to the start of the section by
+            /// the boot code.
+            DP_SECTION: bool = false,
+            /// All sections with the "c" flag are grouped together by the linker to form
+            /// the constant pool and the cp register is set to the start of the constant
+            /// pool by the boot code.
+            CP_SECTION: bool = false,
+            unused2: u1 = 0,
+            /// This section is excluded from the final executable or shared library.
+            EXCLUDE: bool = false,
+        },
+        X86_64: packed struct(u4) {
+            /// If an object file section does not have this flag set, then it may not hold
+            /// more than 2GB and can be freely referred to in objects using smaller code
+            /// models. Otherwise, only objects using larger code models can refer to them.
+            /// For example, a medium code model object can refer to data in a section that
+            /// sets this flag besides being able to refer to data in a section that does
+            /// not set it; likewise, a small code model object can refer only to code in a
+            /// section that does not set this flag.
+            LARGE: bool = false,
+            unused1: u2 = 0,
+            /// This section is excluded from the final executable or shared library.
+            EXCLUDE: bool = false,
+        },
+        HEX: packed struct(u4) {
+            /// All sections with the GPREL flag are grouped into a global data area
+            /// for faster accesses
+            GPREL: bool = false,
+            unused1: u2 = 0,
+            /// This section is excluded from the final executable or shared library.
+            EXCLUDE: bool = false,
+        },
+        MIPS: packed struct(u4) {
+            /// All sections with the GPREL flag are grouped into a global data area
+            /// for faster accesses
+            GPREL: bool = false,
+            /// This section should be merged.
+            MERGE: bool = false,
+            /// Address size to be inferred from section entry size.
+            ADDR: bool = false,
+            /// Section data is string data by default.
+            STRING: bool = false,
+        },
+    } = .{ .MASK = 0 },
+};
+
 /// Execute
 pub const PF_X = 1;
 
@@ -1733,6 +2118,19 @@ pub const PF_MASKOS = 0x0ff00000;
 /// Bits for processor-specific semantics.
 pub const PF_MASKPROC = 0xf0000000;
 
+pub const PF = packed struct(Word) {
+    X: bool = false,
+    W: bool = false,
+    R: bool = false,
+    unused3: u17 = 0,
+    OS: packed union {
+        MASK: u8,
+    } = .{ .MASK = 0 },
+    PROC: packed union {
+        MASK: u4,
+    } = .{ .MASK = 0 },
+};
+
 /// Undefined section
 pub const SHN_UNDEF = 0;
 /// Start of reserved indices
@@ -2303,13 +2701,6 @@ pub const R_PPC64 = enum(u32) {
     _,
 };
 
-pub const STV = enum(u3) {
-    DEFAULT = 0,
-    INTERNAL = 1,
-    HIDDEN = 2,
-    PROTECTED = 3,
-};
-
 pub const ar_hdr = extern struct {
     /// Member file name, sometimes / terminated.
     ar_name: [16]u8,

lib/std/zig/system.zig

@@ -516,15 +516,15 @@ pub fn abiAndDynamicLinkerFromFile(
     const hdr32: *elf.Elf32_Ehdr = @ptrCast(&hdr_buf);
     const hdr64: *elf.Elf64_Ehdr = @ptrCast(&hdr_buf);
     if (!mem.eql(u8, hdr32.e_ident[0..4], elf.MAGIC)) return error.InvalidElfMagic;
-    const elf_endian: std.builtin.Endian = switch (hdr32.e_ident[elf.EI_DATA]) {
+    const elf_endian: std.builtin.Endian = switch (hdr32.e_ident[elf.EI.DATA]) {
         elf.ELFDATA2LSB => .little,
         elf.ELFDATA2MSB => .big,
         else => return error.InvalidElfEndian,
     };
     const need_bswap = elf_endian != native_endian;
-    if (hdr32.e_ident[elf.EI_VERSION] != 1) return error.InvalidElfVersion;
+    if (hdr32.e_ident[elf.EI.VERSION] != 1) return error.InvalidElfVersion;
 
-    const is_64 = switch (hdr32.e_ident[elf.EI_CLASS]) {
+    const is_64 = switch (hdr32.e_ident[elf.EI.CLASS]) {
         elf.ELFCLASS32 => false,
         elf.ELFCLASS64 => true,
         else => return error.InvalidElfClass,
@@ -920,15 +920,15 @@ fn glibcVerFromSoFile(file: fs.File) !std.SemanticVersion {
     const hdr32: *elf.Elf32_Ehdr = @ptrCast(&hdr_buf);
     const hdr64: *elf.Elf64_Ehdr = @ptrCast(&hdr_buf);
     if (!mem.eql(u8, hdr32.e_ident[0..4], elf.MAGIC)) return error.InvalidElfMagic;
-    const elf_endian: std.builtin.Endian = switch (hdr32.e_ident[elf.EI_DATA]) {
+    const elf_endian: std.builtin.Endian = switch (hdr32.e_ident[elf.EI.DATA]) {
         elf.ELFDATA2LSB => .little,
         elf.ELFDATA2MSB => .big,
         else => return error.InvalidElfEndian,
     };
     const need_bswap = elf_endian != native_endian;
-    if (hdr32.e_ident[elf.EI_VERSION] != 1) return error.InvalidElfVersion;
+    if (hdr32.e_ident[elf.EI.VERSION] != 1) return error.InvalidElfVersion;
 
-    const is_64 = switch (hdr32.e_ident[elf.EI_CLASS]) {
+    const is_64 = switch (hdr32.e_ident[elf.EI.CLASS]) {
         elf.ELFCLASS32 => false,
         elf.ELFCLASS64 => true,
         else => return error.InvalidElfClass,

src/Compilation.zig

@@ -177,7 +177,6 @@ debug_compiler_runtime_libs: bool,
 debug_compile_errors: bool,
 /// Do not check this field directly. Instead, use the `debugIncremental` wrapper function.
 debug_incremental: bool,
-incremental: bool,
 alloc_failure_occurred: bool = false,
 last_update_was_cache_hit: bool = false,
 
@@ -256,7 +255,9 @@ mutex: if (builtin.single_threaded) struct {
 test_filters: []const []const u8,
 
 link_task_wait_group: WaitGroup = .{},
-link_prog_node: std.Progress.Node = std.Progress.Node.none,
+link_prog_node: std.Progress.Node = .none,
+link_uav_prog_node: std.Progress.Node = .none,
+link_lazy_prog_node: std.Progress.Node = .none,
 
 llvm_opt_bisect_limit: c_int,
 
@@ -1746,7 +1747,6 @@ pub const CreateOptions = struct {
     debug_compiler_runtime_libs: bool = false,
     debug_compile_errors: bool = false,
     debug_incremental: bool = false,
-    incremental: bool = false,
     /// Normally when you create a `Compilation`, Zig will automatically build
     /// and link in required dependencies, such as compiler-rt and libc. When
     /// building such dependencies themselves, this flag must be set to avoid
@@ -1982,6 +1982,7 @@ pub fn create(gpa: Allocator, arena: Allocator, diag: *CreateDiagnostic, options
             };
             if (have_zcu and (!need_llvm or use_llvm)) {
                 if (output_mode == .Obj) break :s .zcu;
+                if (options.config.use_new_linker) break :s .zcu;
                 switch (target_util.zigBackend(target, use_llvm)) {
                     else => {},
                     .stage2_aarch64, .stage2_x86_64 => if (target.ofmt == .coff) {
@@ -2188,8 +2189,8 @@ pub fn create(gpa: Allocator, arena: Allocator, diag: *CreateDiagnostic, options
                 .inherited = .{},
                 .global = options.config,
                 .parent = options.root_mod,
-            }) catch |err| return switch (err) {
+            }) catch |err| switch (err) {
-                error.OutOfMemory => |e| return e,
+                error.OutOfMemory => return error.OutOfMemory,
                 // None of these are possible because the configuration matches the root module
                 // which already passed these checks.
                 error.ValgrindUnsupportedOnTarget => unreachable,
@@ -2266,7 +2267,6 @@ pub fn create(gpa: Allocator, arena: Allocator, diag: *CreateDiagnostic, options
             .debug_compiler_runtime_libs = options.debug_compiler_runtime_libs,
             .debug_compile_errors = options.debug_compile_errors,
             .debug_incremental = options.debug_incremental,
-            .incremental = options.incremental,
             .root_name = root_name,
             .sysroot = sysroot,
             .windows_libs = .empty,
@@ -2409,6 +2409,8 @@ pub fn create(gpa: Allocator, arena: Allocator, diag: *CreateDiagnostic, options
                 // Synchronize with other matching comments: ZigOnlyHashStuff
                 hash.add(use_llvm);
                 hash.add(options.config.use_lib_llvm);
+                hash.add(options.config.use_lld);
+                hash.add(options.config.use_new_linker);
                 hash.add(options.config.dll_export_fns);
                 hash.add(options.config.is_test);
                 hash.addListOfBytes(options.test_filters);
@@ -3075,14 +3077,29 @@ pub fn update(comp: *Compilation, main_progress_node: std.Progress.Node) UpdateE
 
     // The linker progress node is set up here instead of in `performAllTheWork`, because
     // we also want it around during `flush`.
-    const have_link_node = comp.bin_file != null;
+    if (comp.bin_file) |lf| {
-    if (have_link_node) {
         comp.link_prog_node = main_progress_node.start("Linking", 0);
+        if (lf.cast(.elf2)) |elf| {
+            comp.link_prog_node.increaseEstimatedTotalItems(3);
+            comp.link_uav_prog_node = comp.link_prog_node.start("Constants", 0);
+            comp.link_lazy_prog_node = comp.link_prog_node.start("Synthetics", 0);
+            elf.mf.update_prog_node = comp.link_prog_node.start("Relocations", elf.mf.updates.items.len);
         }
-    defer if (have_link_node) {
+    }
+    defer {
         comp.link_prog_node.end();
         comp.link_prog_node = .none;
-    };
+        comp.link_uav_prog_node.end();
+        comp.link_uav_prog_node = .none;
+        comp.link_lazy_prog_node.end();
+        comp.link_lazy_prog_node = .none;
+        if (comp.bin_file) |lf| {
+            if (lf.cast(.elf2)) |elf| {
+                elf.mf.update_prog_node.end();
+                elf.mf.update_prog_node = .none;
+            }
+        }
+    }
 
     try comp.performAllTheWork(main_progress_node);
 
@@ -3100,6 +3117,8 @@ pub fn update(comp: *Compilation, main_progress_node: std.Progress.Node) UpdateE
                 try pt.populateTestFunctions();
             }
 
+            link.updateErrorData(pt);
+
             try pt.processExports();
         }
 
@@ -3474,6 +3493,8 @@ fn addNonIncrementalStuffToCacheManifest(
 
     man.hash.add(comp.config.use_llvm);
     man.hash.add(comp.config.use_lib_llvm);
+    man.hash.add(comp.config.use_lld);
+    man.hash.add(comp.config.use_new_linker);
     man.hash.add(comp.config.is_test);
     man.hash.add(comp.config.import_memory);
     man.hash.add(comp.config.export_memory);
@@ -4073,7 +4094,7 @@ pub fn getAllErrorsAlloc(comp: *Compilation) error{OutOfMemory}!ErrorBundle {
         defer sorted_failed_analysis.deinit(gpa);
         var added_any_analysis_error = false;
         for (sorted_failed_analysis.items(.key), sorted_failed_analysis.items(.value)) |anal_unit, error_msg| {
-            if (comp.incremental) {
+            if (comp.config.incremental) {
                 const refs = try zcu.resolveReferences();
                 if (!refs.contains(anal_unit)) continue;
             }
@@ -4240,7 +4261,7 @@ pub fn getAllErrorsAlloc(comp: *Compilation) error{OutOfMemory}!ErrorBundle {
 
     // TODO: eventually, this should be behind `std.debug.runtime_safety`. But right now, this is a
     // very common way for incremental compilation bugs to manifest, so let's always check it.
-    if (comp.zcu) |zcu| if (comp.incremental and bundle.root_list.items.len == 0) {
+    if (comp.zcu) |zcu| if (comp.config.incremental and bundle.root_list.items.len == 0) {
         for (zcu.transitive_failed_analysis.keys()) |failed_unit| {
             const refs = try zcu.resolveReferences();
             var ref = refs.get(failed_unit) orelse continue;
@@ -4949,7 +4970,7 @@ fn performAllTheWork(
             tr.stats.n_reachable_files = @intCast(zcu.alive_files.count());
         }
 
-        if (comp.incremental) {
+        if (comp.config.incremental) {
             const update_zir_refs_node = main_progress_node.start("Update ZIR References", 0);
             defer update_zir_refs_node.end();
             try pt.updateZirRefs();

src/Compilation/Config.zig

@@ -49,6 +49,8 @@ use_lib_llvm: bool,
 use_lld: bool,
 c_frontend: CFrontend,
 lto: std.zig.LtoMode,
+use_new_linker: bool,
+incremental: bool,
 /// WASI-only. Type of WASI execution model ("command" or "reactor").
 /// Always set to `command` for non-WASI targets.
 wasi_exec_model: std.builtin.WasiExecModel,
@@ -104,6 +106,8 @@ pub const Options = struct {
     use_lld: ?bool = null,
     use_clang: ?bool = null,
     lto: ?std.zig.LtoMode = null,
+    use_new_linker: ?bool = null,
+    incremental: bool = false,
     /// WASI-only. Type of WASI execution model ("command" or "reactor").
     wasi_exec_model: ?std.builtin.WasiExecModel = null,
     import_memory: ?bool = null,
@@ -147,6 +151,8 @@ pub const ResolveError = error{
     LldUnavailable,
     ClangUnavailable,
     DllExportFnsRequiresWindows,
+    NewLinkerIncompatibleWithLld,
+    NewLinkerIncompatibleObjectFormat,
 };
 
 pub fn resolve(options: Options) ResolveError!Config {
@@ -458,6 +464,22 @@ pub fn resolve(options: Options) ResolveError!Config {
         break :b .none;
     };
 
+    const use_new_linker = b: {
+        if (use_lld) {
+            if (options.use_new_linker == true) return error.NewLinkerIncompatibleWithLld;
+            break :b false;
+        }
+
+        if (!target_util.hasNewLinkerSupport(target.ofmt)) {
+            if (options.use_new_linker == true) return error.NewLinkerIncompatibleObjectFormat;
+            break :b false;
+        }
+
+        if (options.use_new_linker) |x| break :b x;
+
+        break :b options.incremental;
+    };
+
     const root_strip = b: {
         if (options.root_strip) |x| break :b x;
         if (root_optimize_mode == .ReleaseSmall) break :b true;
@@ -531,6 +553,8 @@ pub fn resolve(options: Options) ResolveError!Config {
         .root_error_tracing = root_error_tracing,
         .pie = pie,
         .lto = lto,
+        .use_new_linker = use_new_linker,
+        .incremental = options.incremental,
         .import_memory = import_memory,
         .export_memory = export_memory,
         .shared_memory = shared_memory,

src/InternPool.zig

@@ -6424,14 +6424,25 @@ pub const Alignment = enum(u6) {
         return n + 1;
     }
 
-    const LlvmBuilderAlignment = std.zig.llvm.Builder.Alignment;
+    pub fn toStdMem(a: Alignment) std.mem.Alignment {
-
+        assert(a != .none);
-    pub fn toLlvm(this: @This()) LlvmBuilderAlignment {
+        return @enumFromInt(@intFromEnum(a));
-        return @enumFromInt(@intFromEnum(this));
     }
 
-    pub fn fromLlvm(other: LlvmBuilderAlignment) @This() {
+    pub fn fromStdMem(a: std.mem.Alignment) Alignment {
-        return @enumFromInt(@intFromEnum(other));
+        const r: Alignment = @enumFromInt(@intFromEnum(a));
+        assert(r != .none);
+        return r;
+    }
+
+    const LlvmBuilderAlignment = std.zig.llvm.Builder.Alignment;
+
+    pub fn toLlvm(a: Alignment) LlvmBuilderAlignment {
+        return @enumFromInt(@intFromEnum(a));
+    }
+
+    pub fn fromLlvm(a: LlvmBuilderAlignment) Alignment {
+        return @enumFromInt(@intFromEnum(a));
     }
 };
 

src/Sema.zig

@@ -3032,7 +3032,7 @@ fn zirStructDecl(
     });
     errdefer pt.destroyNamespace(new_namespace_index);
 
-    if (pt.zcu.comp.incremental) {
+    if (pt.zcu.comp.config.incremental) {
         try pt.addDependency(.wrap(.{ .type = wip_ty.index }), .{ .src_hash = tracked_inst });
     }
 
@@ -3430,7 +3430,7 @@ fn zirUnionDecl(
     });
     errdefer pt.destroyNamespace(new_namespace_index);
 
-    if (pt.zcu.comp.incremental) {
+    if (pt.zcu.comp.config.incremental) {
         try pt.addDependency(.wrap(.{ .type = wip_ty.index }), .{ .src_hash = tracked_inst });
     }
 
@@ -6217,7 +6217,7 @@ fn zirExport(sema: *Sema, block: *Block, inst: Zir.Inst.Index) CompileError!void
             if (ptr_info.byte_offset != 0) {
                 return sema.fail(block, ptr_src, "TODO: export pointer in middle of value", .{});
             }
-            if (options.linkage == .internal) return;
+            if (zcu.llvm_object != null and options.linkage == .internal) return;
             const export_ty = Value.fromInterned(uav.val).typeOf(zcu);
             if (!try sema.validateExternType(export_ty, .other)) {
                 return sema.failWithOwnedErrorMsg(block, msg: {
@@ -6256,7 +6256,7 @@ pub fn analyzeExport(
     const zcu = pt.zcu;
     const ip = &zcu.intern_pool;
 
-    if (options.linkage == .internal)
+    if (zcu.llvm_object != null and options.linkage == .internal)
         return;
 
     try sema.ensureNavResolved(block, src, orig_nav_index, .fully);
@@ -7709,7 +7709,7 @@ fn analyzeCall(
         // TODO: comptime call memoization is currently not supported under incremental compilation
         // since dependencies are not marked on callers. If we want to keep this around (we should
         // check that it's worthwhile first!), each memoized call needs an `AnalUnit`.
-        if (zcu.comp.incremental) break :m false;
+        if (zcu.comp.config.incremental) break :m false;
         if (!block.isComptime()) break :m false;
         for (args) |a| {
             const val = (try sema.resolveValue(a)).?;
@@ -31208,7 +31208,7 @@ fn addReferenceEntry(
         .func => |f| assert(ip.unwrapCoercedFunc(f) == f), // for `.{ .func = f }`, `f` must be uncoerced
         else => {},
     }
-    if (!zcu.comp.incremental and zcu.comp.reference_trace == 0) return;
+    if (!zcu.comp.config.incremental and zcu.comp.reference_trace == 0) return;
     const gop = try sema.references.getOrPut(sema.gpa, referenced_unit);
     if (gop.found_existing) return;
     try zcu.addUnitReference(sema.owner, referenced_unit, src, inline_frame: {
@@ -31225,7 +31225,7 @@ pub fn addTypeReferenceEntry(
     referenced_type: InternPool.Index,
 ) !void {
     const zcu = sema.pt.zcu;
-    if (!zcu.comp.incremental and zcu.comp.reference_trace == 0) return;
+    if (!zcu.comp.config.incremental and zcu.comp.reference_trace == 0) return;
     const gop = try sema.type_references.getOrPut(sema.gpa, referenced_type);
     if (gop.found_existing) return;
     try zcu.addTypeReference(sema.owner, referenced_type, src);
@@ -36875,7 +36875,7 @@ fn isKnownZigType(sema: *Sema, ref: Air.Inst.Ref, tag: std.builtin.TypeId) bool
 
 pub fn declareDependency(sema: *Sema, dependee: InternPool.Dependee) !void {
     const pt = sema.pt;
-    if (!pt.zcu.comp.incremental) return;
+    if (!pt.zcu.comp.config.incremental) return;
 
     const gop = try sema.dependencies.getOrPut(sema.gpa, dependee);
     if (gop.found_existing) return;

src/Value.zig

@@ -23,7 +23,7 @@ pub fn format(val: Value, writer: *std.Io.Writer) !void {
 
 /// This is a debug function. In order to print values in a meaningful way
 /// we also need access to the type.
-pub fn dump(start_val: Value, w: std.Io.Writer) std.Io.Writer.Error!void {
+pub fn dump(start_val: Value, w: *std.Io.Writer) std.Io.Writer.Error!void {
     try w.print("(interned: {})", .{start_val.toIntern()});
 }
 

src/Zcu.zig

@@ -3166,7 +3166,7 @@ fn markTransitiveDependersPotentiallyOutdated(zcu: *Zcu, maybe_outdated: AnalUni
 }
 
 pub fn findOutdatedToAnalyze(zcu: *Zcu) Allocator.Error!?AnalUnit {
-    if (!zcu.comp.incremental) return null;
+    if (!zcu.comp.config.incremental) return null;
 
     if (zcu.outdated.count() == 0) {
         // Any units in `potentially_outdated` must just be stuck in loops with one another: none of those

src/Zcu/PerThread.zig

@@ -1815,7 +1815,7 @@ fn createFileRootStruct(
     wip_ty.setName(ip, try file.internFullyQualifiedName(pt), .none);
     ip.namespacePtr(namespace_index).owner_type = wip_ty.index;
 
-    if (zcu.comp.incremental) {
+    if (zcu.comp.config.incremental) {
         try pt.addDependency(.wrap(.{ .type = wip_ty.index }), .{ .src_hash = tracked_inst });
     }
 

src/arch/riscv64/CodeGen.zig

@@ -858,9 +858,11 @@ pub fn generateLazy(
     pt: Zcu.PerThread,
     src_loc: Zcu.LazySrcLoc,
     lazy_sym: link.File.LazySymbol,
-    code: *std.ArrayListUnmanaged(u8),
+    atom_index: u32,
+    w: *std.Io.Writer,
     debug_output: link.File.DebugInfoOutput,
-) CodeGenError!void {
+) (CodeGenError || std.Io.Writer.Error)!void {
+    _ = atom_index;
     const comp = bin_file.comp;
     const gpa = comp.gpa;
     const mod = comp.root_mod;
@@ -914,7 +916,7 @@ pub fn generateLazy(
         },
         .bin_file = bin_file,
         .debug_output = debug_output,
-        .code = code,
+        .w = w,
         .prev_di_pc = undefined, // no debug info yet
         .prev_di_line = undefined, // no debug info yet
         .prev_di_column = undefined, // no debug info yet

src/arch/riscv64/Emit.zig

@@ -3,7 +3,7 @@
 bin_file: *link.File,
 lower: Lower,
 debug_output: link.File.DebugInfoOutput,
-code: *std.ArrayListUnmanaged(u8),
+w: *std.Io.Writer,
 
 prev_di_line: u32,
 prev_di_column: u32,
@@ -13,7 +13,7 @@ prev_di_pc: usize,
 code_offset_mapping: std.AutoHashMapUnmanaged(Mir.Inst.Index, usize) = .empty,
 relocs: std.ArrayListUnmanaged(Reloc) = .empty,
 
-pub const Error = Lower.Error || error{
+pub const Error = Lower.Error || std.Io.Writer.Error || error{
     EmitFail,
 };
 
@@ -25,13 +25,13 @@ pub fn emitMir(emit: *Emit) Error!void {
         try emit.code_offset_mapping.putNoClobber(
             emit.lower.allocator,
             mir_index,
-            @intCast(emit.code.items.len),
+            @intCast(emit.w.end),
         );
         const lowered = try emit.lower.lowerMir(mir_index, .{ .allow_frame_locs = true });
         var lowered_relocs = lowered.relocs;
         for (lowered.insts, 0..) |lowered_inst, lowered_index| {
-            const start_offset: u32 = @intCast(emit.code.items.len);
+            const start_offset: u32 = @intCast(emit.w.end);
-            std.mem.writeInt(u32, try emit.code.addManyAsArray(gpa, 4), lowered_inst.toU32(), .little);
+            try emit.w.writeInt(u32, lowered_inst.toU32(), .little);
 
             while (lowered_relocs.len > 0 and
                 lowered_relocs[0].lowered_inst_index == lowered_index) : ({
@@ -175,7 +175,7 @@ fn fixupRelocs(emit: *Emit) Error!void {
             return emit.fail("relocation target not found!", .{});
 
         const disp = @as(i32, @intCast(target)) - @as(i32, @intCast(reloc.source));
-        const code: *[4]u8 = emit.code.items[reloc.source + reloc.offset ..][0..4];
+        const code = emit.w.buffered()[reloc.source + reloc.offset ..][0..4];
 
         switch (reloc.fmt) {
             .J => riscv_util.writeInstJ(code, @bitCast(disp)),
@@ -187,7 +187,7 @@ fn fixupRelocs(emit: *Emit) Error!void {
 
 fn dbgAdvancePCAndLine(emit: *Emit, line: u32, column: u32) Error!void {
     const delta_line = @as(i33, line) - @as(i33, emit.prev_di_line);
-    const delta_pc: usize = emit.code.items.len - emit.prev_di_pc;
+    const delta_pc: usize = emit.w.end - emit.prev_di_pc;
     log.debug("  (advance pc={d} and line={d})", .{ delta_pc, delta_line });
     switch (emit.debug_output) {
         .dwarf => |dw| {
@@ -196,7 +196,7 @@ fn dbgAdvancePCAndLine(emit: *Emit, line: u32, column: u32) Error!void {
             try dw.advancePCAndLine(delta_line, delta_pc);
             emit.prev_di_line = line;
             emit.prev_di_column = column;
-            emit.prev_di_pc = emit.code.items.len;
+            emit.prev_di_pc = emit.w.end;
         },
         .none => {},
     }

src/arch/riscv64/Mir.zig

@@ -109,9 +109,11 @@ pub fn emit(
     pt: Zcu.PerThread,
     src_loc: Zcu.LazySrcLoc,
     func_index: InternPool.Index,
-    code: *std.ArrayListUnmanaged(u8),
+    atom_index: u32,
+    w: *std.Io.Writer,
     debug_output: link.File.DebugInfoOutput,
-) codegen.CodeGenError!void {
+) (codegen.CodeGenError || std.Io.Writer.Error)!void {
+    _ = atom_index;
     const zcu = pt.zcu;
     const comp = zcu.comp;
     const gpa = comp.gpa;
@@ -132,7 +134,7 @@ pub fn emit(
         },
         .bin_file = lf,
         .debug_output = debug_output,
-        .code = code,
+        .w = w,
         .prev_di_pc = 0,
         .prev_di_line = func.lbrace_line,
         .prev_di_column = func.lbrace_column,

src/arch/sparc64/Emit.zig

@@ -21,7 +21,7 @@ debug_output: link.File.DebugInfoOutput,
 target: *const std.Target,
 err_msg: ?*ErrorMsg = null,
 src_loc: Zcu.LazySrcLoc,
-code: *std.ArrayListUnmanaged(u8),
+w: *std.Io.Writer,
 
 prev_di_line: u32,
 prev_di_column: u32,
@@ -40,7 +40,7 @@ branch_forward_origins: std.AutoHashMapUnmanaged(Mir.Inst.Index, std.ArrayListUn
 /// instruction
 code_offset_mapping: std.AutoHashMapUnmanaged(Mir.Inst.Index, usize) = .empty,
 
-const InnerError = error{
+const InnerError = std.Io.Writer.Error || error{
     OutOfMemory,
     EmitFail,
 };
@@ -292,7 +292,7 @@ fn mirConditionalBranch(emit: *Emit, inst: Mir.Inst.Index) !void {
         .bpcc => switch (tag) {
             .bpcc => {
                 const branch_predict_int = emit.mir.instructions.items(.data)[inst].branch_predict_int;
-                const offset = @as(i64, @intCast(emit.code_offset_mapping.get(branch_predict_int.inst).?)) - @as(i64, @intCast(emit.code.items.len));
+                const offset = @as(i64, @intCast(emit.code_offset_mapping.get(branch_predict_int.inst).?)) - @as(i64, @intCast(emit.w.end));
                 log.debug("mirConditionalBranch: {} offset={}", .{ inst, offset });
 
                 try emit.writeInstruction(
@@ -310,7 +310,7 @@ fn mirConditionalBranch(emit: *Emit, inst: Mir.Inst.Index) !void {
         .bpr => switch (tag) {
             .bpr => {
                 const branch_predict_reg = emit.mir.instructions.items(.data)[inst].branch_predict_reg;
-                const offset = @as(i64, @intCast(emit.code_offset_mapping.get(branch_predict_reg.inst).?)) - @as(i64, @intCast(emit.code.items.len));
+                const offset = @as(i64, @intCast(emit.code_offset_mapping.get(branch_predict_reg.inst).?)) - @as(i64, @intCast(emit.w.end));
                 log.debug("mirConditionalBranch: {} offset={}", .{ inst, offset });
 
                 try emit.writeInstruction(
@@ -494,13 +494,13 @@ fn branchTarget(emit: *Emit, inst: Mir.Inst.Index) Mir.Inst.Index {
 
 fn dbgAdvancePCAndLine(emit: *Emit, line: u32, column: u32) !void {
     const delta_line = @as(i32, @intCast(line)) - @as(i32, @intCast(emit.prev_di_line));
-    const delta_pc: usize = emit.code.items.len - emit.prev_di_pc;
+    const delta_pc: usize = emit.w.end - emit.prev_di_pc;
     switch (emit.debug_output) {
         .dwarf => |dbg_out| {
             try dbg_out.advancePCAndLine(delta_line, delta_pc);
             emit.prev_di_line = line;
             emit.prev_di_column = column;
-            emit.prev_di_pc = emit.code.items.len;
+            emit.prev_di_pc = emit.w.end;
         },
         else => {},
     }
@@ -675,13 +675,8 @@ fn optimalBranchType(emit: *Emit, tag: Mir.Inst.Tag, offset: i64) !BranchType {
 }
 
 fn writeInstruction(emit: *Emit, instruction: Instruction) !void {
-    const comp = emit.bin_file.comp;
-    const gpa = comp.gpa;
-
     // SPARCv9 instructions are always arranged in BE regardless of the
     // endianness mode the CPU is running in (Section 3.1 of the ISA specification).
     // This is to ease porting in case someone wants to do a LE SPARCv9 backend.
-    const endian: Endian = .big;
+    try emit.w.writeInt(u32, instruction.toU32(), .big);
-
-    std.mem.writeInt(u32, try emit.code.addManyAsArray(gpa, 4), instruction.toU32(), endian);
 }

src/arch/sparc64/Mir.zig

@@ -380,9 +380,11 @@ pub fn emit(
     pt: Zcu.PerThread,
     src_loc: Zcu.LazySrcLoc,
     func_index: InternPool.Index,
-    code: *std.ArrayListUnmanaged(u8),
+    atom_index: u32,
+    w: *std.Io.Writer,
     debug_output: link.File.DebugInfoOutput,
-) codegen.CodeGenError!void {
+) (codegen.CodeGenError || std.Io.Writer.Error)!void {
+    _ = atom_index;
     const zcu = pt.zcu;
     const func = zcu.funcInfo(func_index);
     const nav = func.owner_nav;
@@ -393,7 +395,7 @@ pub fn emit(
         .debug_output = debug_output,
         .target = &mod.resolved_target.result,
         .src_loc = src_loc,
-        .code = code,
+        .w = w,
         .prev_di_pc = 0,
         .prev_di_line = func.lbrace_line,
         .prev_di_column = func.lbrace_column,

src/arch/x86_64/CodeGen.zig

@@ -550,9 +550,9 @@ pub const MCValue = union(enum) {
                 @tagName(pl.reg),
             }),
             .indirect => |pl| try w.print("[{s} + 0x{x}]", .{ @tagName(pl.reg), pl.off }),
-            .indirect_load_frame => |pl| try w.print("[[{} + 0x{x}]]", .{ pl.index, pl.off }),
+            .indirect_load_frame => |pl| try w.print("[[{f} + 0x{x}]]", .{ pl.index, pl.off }),
-            .load_frame => |pl| try w.print("[{} + 0x{x}]", .{ pl.index, pl.off }),
+            .load_frame => |pl| try w.print("[{f} + 0x{x}]", .{ pl.index, pl.off }),
-            .lea_frame => |pl| try w.print("{} + 0x{x}", .{ pl.index, pl.off }),
+            .lea_frame => |pl| try w.print("{f} + 0x{x}", .{ pl.index, pl.off }),
             .load_nav => |pl| try w.print("[nav:{d}]", .{@intFromEnum(pl)}),
             .lea_nav => |pl| try w.print("nav:{d}", .{@intFromEnum(pl)}),
             .load_uav => |pl| try w.print("[uav:{d}]", .{@intFromEnum(pl.val)}),
@@ -561,10 +561,10 @@ pub const MCValue = union(enum) {
             .lea_lazy_sym => |pl| try w.print("lazy:{s}:{d}", .{ @tagName(pl.kind), @intFromEnum(pl.ty) }),
             .load_extern_func => |pl| try w.print("[extern:{d}]", .{@intFromEnum(pl)}),
             .lea_extern_func => |pl| try w.print("extern:{d}", .{@intFromEnum(pl)}),
-            .elementwise_args => |pl| try w.print("elementwise:{d}:[{} + 0x{x}]", .{
+            .elementwise_args => |pl| try w.print("elementwise:{d}:[{f} + 0x{x}]", .{
                 pl.regs, pl.frame_index, pl.frame_off,
             }),
-            .reserved_frame => |pl| try w.print("(dead:{})", .{pl}),
+            .reserved_frame => |pl| try w.print("(dead:{f})", .{pl}),
             .air_ref => |pl| try w.print("(air:0x{x})", .{@intFromEnum(pl)}),
         }
     }
@@ -1038,7 +1038,8 @@ pub fn generateLazy(
     pt: Zcu.PerThread,
     src_loc: Zcu.LazySrcLoc,
     lazy_sym: link.File.LazySymbol,
-    code: *std.ArrayListUnmanaged(u8),
+    atom_index: u32,
+    w: *std.Io.Writer,
     debug_output: link.File.DebugInfoOutput,
 ) codegen.CodeGenError!void {
     const gpa = pt.zcu.gpa;
@@ -1081,7 +1082,7 @@ pub fn generateLazy(
         else => |e| return e,
     };
 
-    try function.getTmpMir().emitLazy(bin_file, pt, src_loc, lazy_sym, code, debug_output);
+    try function.getTmpMir().emitLazy(bin_file, pt, src_loc, lazy_sym, atom_index, w, debug_output);
 }
 
 const FormatNavData = struct {
@@ -2022,7 +2023,7 @@ fn gen(
                     .{},
                 );
                 self.ret_mcv.long = .{ .load_frame = .{ .index = frame_index } };
-                tracking_log.debug("spill {f} to {}", .{ self.ret_mcv.long, frame_index });
+                tracking_log.debug("spill {f} to {f}", .{ self.ret_mcv.long, frame_index });
             },
             else => unreachable,
         }

src/arch/x86_64/Emit.zig

@@ -6,7 +6,7 @@ pt: Zcu.PerThread,
 pic: bool,
 atom_index: u32,
 debug_output: link.File.DebugInfoOutput,
-code: *std.ArrayListUnmanaged(u8),
+w: *std.Io.Writer,
 
 prev_di_loc: Loc,
 /// Relative to the beginning of `code`.
@@ -18,7 +18,8 @@ table_relocs: std.ArrayListUnmanaged(TableReloc),
 
 pub const Error = Lower.Error || error{
     EmitFail,
-} || link.File.UpdateDebugInfoError;
+    NotFile,
+} || std.posix.MMapError || std.posix.MRemapError || link.File.UpdateDebugInfoError;
 
 pub fn emitMir(emit: *Emit) Error!void {
     const comp = emit.bin_file.comp;
@@ -29,12 +30,12 @@ pub fn emitMir(emit: *Emit) Error!void {
     var local_index: usize = 0;
     for (0..emit.lower.mir.instructions.len) |mir_i| {
         const mir_index: Mir.Inst.Index = @intCast(mir_i);
-        emit.code_offset_mapping.items[mir_index] = @intCast(emit.code.items.len);
+        emit.code_offset_mapping.items[mir_index] = @intCast(emit.w.end);
         const lowered = try emit.lower.lowerMir(mir_index);
         var lowered_relocs = lowered.relocs;
         lowered_inst: for (lowered.insts, 0..) |lowered_inst, lowered_index| {
             if (lowered_inst.prefix == .directive) {
-                const start_offset: u32 = @intCast(emit.code.items.len);
+                const start_offset: u32 = @intCast(emit.w.end);
                 switch (emit.debug_output) {
                     .dwarf => |dwarf| switch (lowered_inst.encoding.mnemonic) {
                         .@".cfi_def_cfa" => try dwarf.genDebugFrame(start_offset, .{ .def_cfa = .{
@@ -164,6 +165,8 @@ pub fn emitMir(emit: *Emit) Error!void {
                         .index = if (emit.bin_file.cast(.elf)) |elf_file|
                             elf_file.zigObjectPtr().?.getOrCreateMetadataForLazySymbol(elf_file, emit.pt, lazy_sym) catch |err|
                                 return emit.fail("{s} creating lazy symbol", .{@errorName(err)})
+                        else if (emit.bin_file.cast(.elf2)) |elf|
+                            @intFromEnum(try elf.lazySymbol(lazy_sym))
                         else if (emit.bin_file.cast(.macho)) |macho_file|
                             macho_file.getZigObject().?.getOrCreateMetadataForLazySymbol(macho_file, emit.pt, lazy_sym) catch |err|
                                 return emit.fail("{s} creating lazy symbol", .{@errorName(err)})
@@ -180,12 +183,15 @@ pub fn emitMir(emit: *Emit) Error!void {
                     .extern_func => |extern_func| .{
                         .index = if (emit.bin_file.cast(.elf)) |elf_file|
                             try elf_file.getGlobalSymbol(extern_func.toSlice(&emit.lower.mir).?, null)
-                        else if (emit.bin_file.cast(.macho)) |macho_file|
+                        else if (emit.bin_file.cast(.elf2)) |elf| @intFromEnum(try elf.globalSymbol(.{
+                            .name = extern_func.toSlice(&emit.lower.mir).?,
+                            .type = .FUNC,
+                        })) else if (emit.bin_file.cast(.macho)) |macho_file|
                             try macho_file.getGlobalSymbol(extern_func.toSlice(&emit.lower.mir).?, null)
                         else if (emit.bin_file.cast(.coff)) |coff_file|
                             try coff_file.getGlobalSymbol(extern_func.toSlice(&emit.lower.mir).?, "compiler_rt")
                         else
-                            return emit.fail("external symbols unimplemented for {s}", .{@tagName(emit.bin_file.tag)}),
+                            return emit.fail("external symbol unimplemented for {s}", .{@tagName(emit.bin_file.tag)}),
                         .is_extern = true,
                         .type = .symbol,
                     },
@@ -205,7 +211,7 @@ pub fn emitMir(emit: *Emit) Error!void {
                         },
                         else => {},
                     }
-                    if (emit.bin_file.cast(.elf)) |_| {
+                    if (emit.bin_file.cast(.elf) != null or emit.bin_file.cast(.elf2) != null) {
                         if (!emit.pic) switch (lowered_inst.encoding.mnemonic) {
                             .lea => try emit.encodeInst(try .new(.none, .mov, &.{
                                 lowered_inst.ops[0],
@@ -315,7 +321,7 @@ pub fn emitMir(emit: *Emit) Error!void {
                 },
                 .branch, .tls => unreachable,
                 .tlv => {
-                    if (emit.bin_file.cast(.elf)) |elf_file| {
+                    if (emit.bin_file.cast(.elf) != null or emit.bin_file.cast(.elf2) != null) {
                         // TODO handle extern TLS vars, i.e., emit GD model
                         if (emit.pic) switch (lowered_inst.encoding.mnemonic) {
                             .lea, .mov => {
@@ -337,7 +343,12 @@ pub fn emitMir(emit: *Emit) Error!void {
                                 }, emit.lower.target), &.{.{
                                     .op_index = 0,
                                     .target = .{
-                                        .index = try elf_file.getGlobalSymbol("__tls_get_addr", null),
+                                        .index = if (emit.bin_file.cast(.elf)) |elf_file|
+                                            try elf_file.getGlobalSymbol("__tls_get_addr", null)
+                                        else if (emit.bin_file.cast(.elf2)) |elf| @intFromEnum(try elf.globalSymbol(.{
+                                            .name = "__tls_get_addr",
+                                            .type = .FUNC,
+                                        })) else unreachable,
                                         .is_extern = true,
                                         .type = .branch,
                                     },
@@ -441,7 +452,7 @@ pub fn emitMir(emit: *Emit) Error!void {
                             log.debug("mirDbgEnterBlock (line={d}, col={d})", .{
                                 emit.prev_di_loc.line, emit.prev_di_loc.column,
                             });
-                            try dwarf.enterBlock(emit.code.items.len);
+                            try dwarf.enterBlock(emit.w.end);
                         },
                         .none => {},
                     },
@@ -450,7 +461,7 @@ pub fn emitMir(emit: *Emit) Error!void {
                             log.debug("mirDbgLeaveBlock (line={d}, col={d})", .{
                                 emit.prev_di_loc.line, emit.prev_di_loc.column,
                             });
-                            try dwarf.leaveBlock(emit.code.items.len);
+                            try dwarf.leaveBlock(emit.w.end);
                         },
                         .none => {},
                     },
@@ -459,7 +470,7 @@ pub fn emitMir(emit: *Emit) Error!void {
                             log.debug("mirDbgEnterInline (line={d}, col={d})", .{
                                 emit.prev_di_loc.line, emit.prev_di_loc.column,
                             });
-                            try dwarf.enterInlineFunc(mir_inst.data.ip_index, emit.code.items.len, emit.prev_di_loc.line, emit.prev_di_loc.column);
+                            try dwarf.enterInlineFunc(mir_inst.data.ip_index, emit.w.end, emit.prev_di_loc.line, emit.prev_di_loc.column);
                         },
                         .none => {},
                     },
@@ -468,7 +479,7 @@ pub fn emitMir(emit: *Emit) Error!void {
                             log.debug("mirDbgLeaveInline (line={d}, col={d})", .{
                                 emit.prev_di_loc.line, emit.prev_di_loc.column,
                             });
-                            try dwarf.leaveInlineFunc(mir_inst.data.ip_index, emit.code.items.len);
+                            try dwarf.leaveInlineFunc(mir_inst.data.ip_index, emit.w.end);
                         },
                         .none => {},
                     },
@@ -634,7 +645,7 @@ pub fn emitMir(emit: *Emit) Error!void {
     for (emit.relocs.items) |reloc| {
         const target = emit.code_offset_mapping.items[reloc.target];
         const disp = @as(i64, @intCast(target)) - @as(i64, @intCast(reloc.inst_offset + reloc.inst_length)) + reloc.target_offset;
-        const inst_bytes = emit.code.items[reloc.inst_offset..][0..reloc.inst_length];
+        const inst_bytes = emit.w.buffered()[reloc.inst_offset..][0..reloc.inst_length];
         switch (reloc.source_length) {
             else => unreachable,
             inline 1, 4 => |source_length| std.mem.writeInt(
@@ -646,12 +657,12 @@ pub fn emitMir(emit: *Emit) Error!void {
         }
     }
     if (emit.lower.mir.table.len > 0) {
+        const ptr_size = @divExact(emit.lower.target.ptrBitWidth(), 8);
+        var table_offset = std.mem.alignForward(u32, @intCast(emit.w.end), ptr_size);
         if (emit.bin_file.cast(.elf)) |elf_file| {
             const zo = elf_file.zigObjectPtr().?;
             const atom = zo.symbol(emit.atom_index).atom(elf_file).?;
 
-            const ptr_size = @divExact(emit.lower.target.ptrBitWidth(), 8);
-            var table_offset = std.mem.alignForward(u32, @intCast(emit.code.items.len), ptr_size);
             for (emit.table_relocs.items) |table_reloc| try atom.addReloc(gpa, .{
                 .r_offset = table_reloc.source_offset,
                 .r_info = @as(u64, emit.atom_index) << 32 | @intFromEnum(std.elf.R_X86_64.@"32"),
@@ -665,7 +676,26 @@ pub fn emitMir(emit: *Emit) Error!void {
                 }, zo);
                 table_offset += ptr_size;
             }
-            try emit.code.appendNTimes(gpa, 0, table_offset - emit.code.items.len);
+            try emit.w.splatByteAll(0, table_offset - emit.w.end);
+        } else if (emit.bin_file.cast(.elf2)) |elf| {
+            for (emit.table_relocs.items) |table_reloc| try elf.addReloc(
+                @enumFromInt(emit.atom_index),
+                table_reloc.source_offset,
+                @enumFromInt(emit.atom_index),
+                @as(i64, table_offset) + table_reloc.target_offset,
+                .{ .x86_64 = .@"32" },
+            );
+            for (emit.lower.mir.table) |entry| {
+                try elf.addReloc(
+                    @enumFromInt(emit.atom_index),
+                    table_offset,
+                    @enumFromInt(emit.atom_index),
+                    emit.code_offset_mapping.items[entry],
+                    .{ .x86_64 = .@"64" },
+                );
+                table_offset += ptr_size;
+            }
+            try emit.w.splatByteAll(0, table_offset - emit.w.end);
         } else unreachable;
     }
 }
@@ -696,16 +726,12 @@ const RelocInfo = struct {
 fn encodeInst(emit: *Emit, lowered_inst: Instruction, reloc_info: []const RelocInfo) Error!void {
     const comp = emit.bin_file.comp;
     const gpa = comp.gpa;
-    const start_offset: u32 = @intCast(emit.code.items.len);
+    const start_offset: u32 = @intCast(emit.w.end);
-    {
+    lowered_inst.encode(emit.w, .{}) catch |err| switch (err) {
-        var aw: std.Io.Writer.Allocating = .fromArrayList(gpa, emit.code);
-        defer emit.code.* = aw.toArrayList();
-        lowered_inst.encode(&aw.writer, .{}) catch |err| switch (err) {
         error.WriteFailed => return error.OutOfMemory,
         else => |e| return e,
     };
-    }
+    const end_offset: u32 = @intCast(emit.w.end);
-    const end_offset: u32 = @intCast(emit.code.items.len);
     for (reloc_info) |reloc| switch (reloc.target.type) {
         .inst => {
             const inst_length: u4 = @intCast(end_offset - start_offset);
@@ -769,7 +795,13 @@ fn encodeInst(emit: *Emit, lowered_inst: Instruction, reloc_info: []const RelocI
                     .symbolnum = @intCast(reloc.target.index),
                 },
             });
-        } else if (emit.bin_file.cast(.coff)) |coff_file| {
+        } else if (emit.bin_file.cast(.elf2)) |elf| try elf.addReloc(
+            @enumFromInt(emit.atom_index),
+            end_offset - 4,
+            @enumFromInt(reloc.target.index),
+            reloc.off,
+            .{ .x86_64 = .@"32" },
+        ) else if (emit.bin_file.cast(.coff)) |coff_file| {
             const atom_index = coff_file.getAtomIndexForSymbol(
                 .{ .sym_index = emit.atom_index, .file = null },
             ).?;
@@ -794,7 +826,13 @@ fn encodeInst(emit: *Emit, lowered_inst: Instruction, reloc_info: []const RelocI
                 .r_info = @as(u64, reloc.target.index) << 32 | @intFromEnum(r_type),
                 .r_addend = reloc.off - 4,
             }, zo);
-        } else if (emit.bin_file.cast(.macho)) |macho_file| {
+        } else if (emit.bin_file.cast(.elf2)) |elf| try elf.addReloc(
+            @enumFromInt(emit.atom_index),
+            end_offset - 4,
+            @enumFromInt(reloc.target.index),
+            reloc.off - 4,
+            .{ .x86_64 = .PC32 },
+        ) else if (emit.bin_file.cast(.macho)) |macho_file| {
             const zo = macho_file.getZigObject().?;
             const atom = zo.symbols.items[emit.atom_index].getAtom(macho_file).?;
             try atom.addReloc(macho_file, .{
@@ -849,7 +887,13 @@ fn encodeInst(emit: *Emit, lowered_inst: Instruction, reloc_info: []const RelocI
                 .r_info = @as(u64, reloc.target.index) << 32 | @intFromEnum(r_type),
                 .r_addend = reloc.off,
             }, zo);
-        } else if (emit.bin_file.cast(.macho)) |macho_file| {
+        } else if (emit.bin_file.cast(.elf2)) |elf| try elf.addReloc(
+            @enumFromInt(emit.atom_index),
+            end_offset - 4,
+            @enumFromInt(reloc.target.index),
+            reloc.off,
+            .{ .x86_64 = .TPOFF32 },
+        ) else if (emit.bin_file.cast(.macho)) |macho_file| {
             const zo = macho_file.getZigObject().?;
             const atom = zo.symbols.items[emit.atom_index].getAtom(macho_file).?;
             try atom.addReloc(macho_file, .{
@@ -908,7 +952,7 @@ const Loc = struct {
 
 fn dbgAdvancePCAndLine(emit: *Emit, loc: Loc) Error!void {
     const delta_line = @as(i33, loc.line) - @as(i33, emit.prev_di_loc.line);
-    const delta_pc: usize = emit.code.items.len - emit.prev_di_pc;
+    const delta_pc: usize = emit.w.end - emit.prev_di_pc;
     log.debug("  (advance pc={d} and line={d})", .{ delta_pc, delta_line });
     switch (emit.debug_output) {
         .dwarf => |dwarf| {
@@ -916,7 +960,7 @@ fn dbgAdvancePCAndLine(emit: *Emit, loc: Loc) Error!void {
             if (loc.column != emit.prev_di_loc.column) try dwarf.setColumn(loc.column);
             try dwarf.advancePCAndLine(delta_line, delta_pc);
             emit.prev_di_loc = loc;
-            emit.prev_di_pc = emit.code.items.len;
+            emit.prev_di_pc = emit.w.end;
         },
         .none => {},
     }

src/arch/x86_64/Mir.zig

@@ -1976,7 +1976,8 @@ pub fn emit(
     pt: Zcu.PerThread,
     src_loc: Zcu.LazySrcLoc,
     func_index: InternPool.Index,
-    code: *std.ArrayListUnmanaged(u8),
+    atom_index: u32,
+    w: *std.Io.Writer,
     debug_output: link.File.DebugInfoOutput,
 ) codegen.CodeGenError!void {
     const zcu = pt.zcu;
@@ -1997,17 +1998,9 @@ pub fn emit(
         .bin_file = lf,
         .pt = pt,
         .pic = mod.pic,
-        .atom_index = sym: {
+        .atom_index = atom_index,
-            if (lf.cast(.elf)) |ef| break :sym try ef.zigObjectPtr().?.getOrCreateMetadataForNav(zcu, nav);
-            if (lf.cast(.macho)) |mf| break :sym try mf.getZigObject().?.getOrCreateMetadataForNav(mf, nav);
-            if (lf.cast(.coff)) |cf| {
-                const atom = try cf.getOrCreateAtomForNav(nav);
-                break :sym cf.getAtom(atom).getSymbolIndex().?;
-            }
-            unreachable;
-        },
         .debug_output = debug_output,
-        .code = code,
+        .w = w,
 
         .prev_di_loc = .{
             .line = func.lbrace_line,
@@ -2037,7 +2030,8 @@ pub fn emitLazy(
     pt: Zcu.PerThread,
     src_loc: Zcu.LazySrcLoc,
     lazy_sym: link.File.LazySymbol,
-    code: *std.ArrayListUnmanaged(u8),
+    atom_index: u32,
+    w: *std.Io.Writer,
     debug_output: link.File.DebugInfoOutput,
 ) codegen.CodeGenError!void {
     const zcu = pt.zcu;
@@ -2055,20 +2049,9 @@ pub fn emitLazy(
         .bin_file = lf,
         .pt = pt,
         .pic = mod.pic,
-        .atom_index = sym: {
+        .atom_index = atom_index,
-            if (lf.cast(.elf)) |ef| break :sym ef.zigObjectPtr().?.getOrCreateMetadataForLazySymbol(ef, pt, lazy_sym) catch |err|
-                return zcu.codegenFailType(lazy_sym.ty, "{s} creating lazy symbol", .{@errorName(err)});
-            if (lf.cast(.macho)) |mf| break :sym mf.getZigObject().?.getOrCreateMetadataForLazySymbol(mf, pt, lazy_sym) catch |err|
-                return zcu.codegenFailType(lazy_sym.ty, "{s} creating lazy symbol", .{@errorName(err)});
-            if (lf.cast(.coff)) |cf| {
-                const atom = cf.getOrCreateAtomForLazySymbol(pt, lazy_sym) catch |err|
-                    return zcu.codegenFailType(lazy_sym.ty, "{s} creating lazy symbol", .{@errorName(err)});
-                break :sym cf.getAtom(atom).getSymbolIndex().?;
-            }
-            unreachable;
-        },
         .debug_output = debug_output,
-        .code = code,
+        .w = w,
 
         .prev_di_loc = undefined,
         .prev_di_pc = undefined,

src/arch/x86_64/bits.zig

@@ -727,6 +727,14 @@ pub const FrameIndex = enum(u32) {
     pub fn isNamed(fi: FrameIndex) bool {
         return @intFromEnum(fi) < named_count;
     }
+
+    pub fn format(fi: FrameIndex, writer: *std.Io.Writer) std.Io.Writer.Error!void {
+        if (fi.isNamed()) {
+            try writer.print("FrameIndex.{t}", .{fi});
+        } else {
+            try writer.print("FrameIndex({d})", .{@intFromEnum(fi)});
+        }
+    }
 };
 
 pub const FrameAddr = struct { index: FrameIndex, off: i32 = 0 };

src/arch/x86_64/encoder.zig

@@ -259,7 +259,7 @@ pub const Instruction = struct {
                             switch (sib.base) {
                                 .none => any = false,
                                 .reg => |reg| try w.print("{s}", .{@tagName(reg)}),
-                                .frame => |frame_index| try w.print("{}", .{frame_index}),
+                                .frame => |frame_index| try w.print("{f}", .{frame_index}),
                                 .table => try w.print("Table", .{}),
                                 .rip_inst => |inst_index| try w.print("RipInst({d})", .{inst_index}),
                                 .nav => |nav| try w.print("Nav({d})", .{@intFromEnum(nav)}),

src/codegen.zig

@@ -6,7 +6,6 @@ const link = @import("link.zig");
 const log = std.log.scoped(.codegen);
 const mem = std.mem;
 const math = std.math;
-const ArrayList = std.ArrayList;
 const target_util = @import("target.zig");
 const trace = @import("tracy.zig").trace;
 
@@ -179,10 +178,11 @@ pub fn emitFunction(
     pt: Zcu.PerThread,
     src_loc: Zcu.LazySrcLoc,
     func_index: InternPool.Index,
+    atom_index: u32,
     any_mir: *const AnyMir,
-    code: *ArrayList(u8),
+    w: *std.Io.Writer,
     debug_output: link.File.DebugInfoOutput,
-) CodeGenError!void {
+) (CodeGenError || std.Io.Writer.Error)!void {
     const zcu = pt.zcu;
     const func = zcu.funcInfo(func_index);
     const target = &zcu.navFileScope(func.owner_nav).mod.?.resolved_target.result;
@@ -195,7 +195,7 @@ pub fn emitFunction(
         => |backend| {
             dev.check(devFeatureForBackend(backend));
             const mir = &@field(any_mir, AnyMir.tag(backend));
-            return mir.emit(lf, pt, src_loc, func_index, code, debug_output);
+            return mir.emit(lf, pt, src_loc, func_index, atom_index, w, debug_output);
         },
     }
 }
@@ -205,9 +205,10 @@ pub fn generateLazyFunction(
     pt: Zcu.PerThread,
     src_loc: Zcu.LazySrcLoc,
     lazy_sym: link.File.LazySymbol,
-    code: *ArrayList(u8),
+    atom_index: u32,
+    w: *std.Io.Writer,
     debug_output: link.File.DebugInfoOutput,
-) CodeGenError!void {
+) (CodeGenError || std.Io.Writer.Error)!void {
     const zcu = pt.zcu;
     const target = if (Type.fromInterned(lazy_sym.ty).typeDeclInstAllowGeneratedTag(zcu)) |inst_index|
         &zcu.fileByIndex(inst_index.resolveFile(&zcu.intern_pool)).mod.?.resolved_target.result
@@ -217,19 +218,11 @@ pub fn generateLazyFunction(
         else => unreachable,
         inline .stage2_riscv64, .stage2_x86_64 => |backend| {
             dev.check(devFeatureForBackend(backend));
-            return importBackend(backend).generateLazy(lf, pt, src_loc, lazy_sym, code, debug_output);
+            return importBackend(backend).generateLazy(lf, pt, src_loc, lazy_sym, atom_index, w, debug_output);
         },
     }
 }
 
-fn writeFloat(comptime F: type, f: F, target: *const std.Target, endian: std.builtin.Endian, code: []u8) void {
-    _ = target;
-    const bits = @typeInfo(F).float.bits;
-    const Int = @Type(.{ .int = .{ .signedness = .unsigned, .bits = bits } });
-    const int: Int = @bitCast(f);
-    mem.writeInt(Int, code[0..@divExact(bits, 8)], int, endian);
-}
-
 pub fn generateLazySymbol(
     bin_file: *link.File,
     pt: Zcu.PerThread,
@@ -237,17 +230,14 @@ pub fn generateLazySymbol(
     lazy_sym: link.File.LazySymbol,
     // TODO don't use an "out" parameter like this; put it in the result instead
     alignment: *Alignment,
-    code: *ArrayList(u8),
+    w: *std.Io.Writer,
     debug_output: link.File.DebugInfoOutput,
     reloc_parent: link.File.RelocInfo.Parent,
-) CodeGenError!void {
+) (CodeGenError || std.Io.Writer.Error)!void {
-    _ = reloc_parent;
-
     const tracy = trace(@src());
     defer tracy.end();
 
     const comp = bin_file.comp;
-    const gpa = comp.gpa;
     const zcu = pt.zcu;
     const ip = &zcu.intern_pool;
     const target = &comp.root_mod.resolved_target.result;
@@ -260,37 +250,36 @@ pub fn generateLazySymbol(
 
     if (lazy_sym.kind == .code) {
         alignment.* = target_util.defaultFunctionAlignment(target);
-        return generateLazyFunction(bin_file, pt, src_loc, lazy_sym, code, debug_output);
+        return generateLazyFunction(bin_file, pt, src_loc, lazy_sym, reloc_parent.atom_index, w, debug_output);
     }
 
     if (lazy_sym.ty == .anyerror_type) {
         alignment.* = .@"4";
         const err_names = ip.global_error_set.getNamesFromMainThread();
-        var offset_index: u32 = @intCast(code.items.len);
+        const strings_start: u32 = @intCast(4 * (1 + err_names.len + @intFromBool(err_names.len > 0)));
-        var string_index: u32 = @intCast(4 * (1 + err_names.len + @intFromBool(err_names.len > 0)));
+        var string_index = strings_start;
-        try code.resize(gpa, offset_index + string_index);
+        try w.rebase(w.end, string_index);
-        mem.writeInt(u32, code.items[offset_index..][0..4], @intCast(err_names.len), endian);
+        w.writeInt(u32, @intCast(err_names.len), endian) catch unreachable;
         if (err_names.len == 0) return;
-        offset_index += 4;
         for (err_names) |err_name_nts| {
-            const err_name = err_name_nts.toSlice(ip);
+            w.writeInt(u32, string_index, endian) catch unreachable;
-            mem.writeInt(u32, code.items[offset_index..][0..4], string_index, endian);
+            string_index += @intCast(err_name_nts.toSlice(ip).len + 1);
-            offset_index += 4;
+        }
-            try code.ensureUnusedCapacity(gpa, err_name.len + 1);
+        w.writeInt(u32, string_index, endian) catch unreachable;
-            code.appendSliceAssumeCapacity(err_name);
+        try w.rebase(w.end, string_index - strings_start);
-            code.appendAssumeCapacity(0);
+        for (err_names) |err_name_nts| {
-            string_index += @intCast(err_name.len + 1);
+            w.writeAll(err_name_nts.toSlice(ip)) catch unreachable;
+            w.writeByte(0) catch unreachable;
         }
-        mem.writeInt(u32, code.items[offset_index..][0..4], string_index, endian);
     } else if (Type.fromInterned(lazy_sym.ty).zigTypeTag(zcu) == .@"enum") {
         alignment.* = .@"1";
         const enum_ty = Type.fromInterned(lazy_sym.ty);
         const tag_names = enum_ty.enumFields(zcu);
         for (0..tag_names.len) |tag_index| {
             const tag_name = tag_names.get(ip)[tag_index].toSlice(ip);
-            try code.ensureUnusedCapacity(gpa, tag_name.len + 1);
+            try w.rebase(w.end, tag_name.len + 1);
-            code.appendSliceAssumeCapacity(tag_name);
+            w.writeAll(tag_name) catch unreachable;
-            code.appendAssumeCapacity(0);
+            w.writeByte(0) catch unreachable;
         }
     } else {
         return zcu.codegenFailType(lazy_sym.ty, "TODO implement generateLazySymbol for {s} {f}", .{
@@ -312,14 +301,13 @@ pub fn generateSymbol(
     pt: Zcu.PerThread,
     src_loc: Zcu.LazySrcLoc,
     val: Value,
-    code: *ArrayList(u8),
+    w: *std.Io.Writer,
     reloc_parent: link.File.RelocInfo.Parent,
-) GenerateSymbolError!void {
+) (GenerateSymbolError || std.Io.Writer.Error)!void {
     const tracy = trace(@src());
     defer tracy.end();
 
     const zcu = pt.zcu;
-    const gpa = zcu.gpa;
     const ip = &zcu.intern_pool;
     const ty = val.typeOf(zcu);
 
@@ -330,7 +318,7 @@ pub fn generateSymbol(
 
     if (val.isUndef(zcu)) {
         const abi_size = math.cast(usize, ty.abiSize(zcu)) orelse return error.Overflow;
-        try code.appendNTimes(gpa, 0xaa, abi_size);
+        try w.splatByteAll(0xaa, abi_size);
         return;
     }
 
@@ -360,7 +348,7 @@ pub fn generateSymbol(
             .null => unreachable, // non-runtime value
             .@"unreachable" => unreachable, // non-runtime value
             .empty_tuple => return,
-            .false, .true => try code.append(gpa, switch (simple_value) {
+            .false, .true => try w.writeByte(switch (simple_value) {
                 .false => 0,
                 .true => 1,
                 else => unreachable,
@@ -376,11 +364,11 @@ pub fn generateSymbol(
             const abi_size = math.cast(usize, ty.abiSize(zcu)) orelse return error.Overflow;
             var space: Value.BigIntSpace = undefined;
             const int_val = val.toBigInt(&space, zcu);
-            int_val.writeTwosComplement(try code.addManyAsSlice(gpa, abi_size), endian);
+            int_val.writeTwosComplement(try w.writableSlice(abi_size), endian);
         },
         .err => |err| {
             const int = try pt.getErrorValue(err.name);
-            mem.writeInt(u16, try code.addManyAsArray(gpa, 2), @intCast(int), endian);
+            try w.writeInt(u16, @intCast(int), endian);
         },
         .error_union => |error_union| {
             const payload_ty = ty.errorUnionPayload(zcu);
@@ -390,7 +378,7 @@ pub fn generateSymbol(
             };
 
             if (!payload_ty.hasRuntimeBitsIgnoreComptime(zcu)) {
-                mem.writeInt(u16, try code.addManyAsArray(gpa, 2), err_val, endian);
+                try w.writeInt(u16, err_val, endian);
                 return;
             }
 
@@ -400,63 +388,63 @@ pub fn generateSymbol(
 
             // error value first when its type is larger than the error union's payload
             if (error_align.order(payload_align) == .gt) {
-                mem.writeInt(u16, try code.addManyAsArray(gpa, 2), err_val, endian);
+                try w.writeInt(u16, err_val, endian);
             }
 
             // emit payload part of the error union
             {
-                const begin = code.items.len;
+                const begin = w.end;
                 try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(switch (error_union.val) {
                     .err_name => try pt.intern(.{ .undef = payload_ty.toIntern() }),
                     .payload => |payload| payload,
-                }), code, reloc_parent);
+                }), w, reloc_parent);
-                const unpadded_end = code.items.len - begin;
+                const unpadded_end = w.end - begin;
                 const padded_end = abi_align.forward(unpadded_end);
                 const padding = math.cast(usize, padded_end - unpadded_end) orelse return error.Overflow;
 
                 if (padding > 0) {
-                    try code.appendNTimes(gpa, 0, padding);
+                    try w.splatByteAll(0, padding);
                 }
             }
 
             // Payload size is larger than error set, so emit our error set last
             if (error_align.compare(.lte, payload_align)) {
-                const begin = code.items.len;
+                const begin = w.end;
-                mem.writeInt(u16, try code.addManyAsArray(gpa, 2), err_val, endian);
+                try w.writeInt(u16, err_val, endian);
-                const unpadded_end = code.items.len - begin;
+                const unpadded_end = w.end - begin;
                 const padded_end = abi_align.forward(unpadded_end);
                 const padding = math.cast(usize, padded_end - unpadded_end) orelse return error.Overflow;
 
                 if (padding > 0) {
-                    try code.appendNTimes(gpa, 0, padding);
+                    try w.splatByteAll(0, padding);
                 }
             }
         },
         .enum_tag => |enum_tag| {
             const int_tag_ty = ty.intTagType(zcu);
-            try generateSymbol(bin_file, pt, src_loc, try pt.getCoerced(Value.fromInterned(enum_tag.int), int_tag_ty), code, reloc_parent);
+            try generateSymbol(bin_file, pt, src_loc, try pt.getCoerced(Value.fromInterned(enum_tag.int), int_tag_ty), w, reloc_parent);
         },
         .float => |float| storage: switch (float.storage) {
-            .f16 => |f16_val| writeFloat(f16, f16_val, target, endian, try code.addManyAsArray(gpa, 2)),
+            .f16 => |f16_val| try w.writeInt(u16, @bitCast(f16_val), endian),
-            .f32 => |f32_val| writeFloat(f32, f32_val, target, endian, try code.addManyAsArray(gpa, 4)),
+            .f32 => |f32_val| try w.writeInt(u32, @bitCast(f32_val), endian),
-            .f64 => |f64_val| writeFloat(f64, f64_val, target, endian, try code.addManyAsArray(gpa, 8)),
+            .f64 => |f64_val| try w.writeInt(u64, @bitCast(f64_val), endian),
             .f80 => |f80_val| {
-                writeFloat(f80, f80_val, target, endian, try code.addManyAsArray(gpa, 10));
+                try w.writeInt(u80, @bitCast(f80_val), endian);
                 const abi_size = math.cast(usize, ty.abiSize(zcu)) orelse return error.Overflow;
-                try code.appendNTimes(gpa, 0, abi_size - 10);
+                try w.splatByteAll(0, abi_size - 10);
             },
             .f128 => |f128_val| switch (Type.fromInterned(float.ty).floatBits(target)) {
                 else => unreachable,
                 16 => continue :storage .{ .f16 = @floatCast(f128_val) },
                 32 => continue :storage .{ .f32 = @floatCast(f128_val) },
                 64 => continue :storage .{ .f64 = @floatCast(f128_val) },
-                128 => writeFloat(f128, f128_val, target, endian, try code.addManyAsArray(gpa, 16)),
+                128 => try w.writeInt(u128, @bitCast(f128_val), endian),
             },
         },
-        .ptr => try lowerPtr(bin_file, pt, src_loc, val.toIntern(), code, reloc_parent, 0),
+        .ptr => try lowerPtr(bin_file, pt, src_loc, val.toIntern(), w, reloc_parent, 0),
         .slice => |slice| {
-            try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(slice.ptr), code, reloc_parent);
+            try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(slice.ptr), w, reloc_parent);
-            try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(slice.len), code, reloc_parent);
+            try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(slice.len), w, reloc_parent);
         },
         .opt => {
             const payload_type = ty.optionalChild(zcu);
@@ -465,9 +453,9 @@ pub fn generateSymbol(
 
             if (ty.optionalReprIsPayload(zcu)) {
                 if (payload_val) |value| {
-                    try generateSymbol(bin_file, pt, src_loc, value, code, reloc_parent);
+                    try generateSymbol(bin_file, pt, src_loc, value, w, reloc_parent);
                 } else {
-                    try code.appendNTimes(gpa, 0, abi_size);
+                    try w.splatByteAll(0, abi_size);
                 }
             } else {
                 const padding = abi_size - (math.cast(usize, payload_type.abiSize(zcu)) orelse return error.Overflow) - 1;
@@ -475,15 +463,15 @@ pub fn generateSymbol(
                     const value = payload_val orelse Value.fromInterned(try pt.intern(.{
                         .undef = payload_type.toIntern(),
                     }));
-                    try generateSymbol(bin_file, pt, src_loc, value, code, reloc_parent);
+                    try generateSymbol(bin_file, pt, src_loc, value, w, reloc_parent);
                 }
-                try code.append(gpa, @intFromBool(payload_val != null));
+                try w.writeByte(@intFromBool(payload_val != null));
-                try code.appendNTimes(gpa, 0, padding);
+                try w.splatByteAll(0, padding);
             }
         },
         .aggregate => |aggregate| switch (ip.indexToKey(ty.toIntern())) {
             .array_type => |array_type| switch (aggregate.storage) {
-                .bytes => |bytes| try code.appendSlice(gpa, bytes.toSlice(array_type.lenIncludingSentinel(), ip)),
+                .bytes => |bytes| try w.writeAll(bytes.toSlice(array_type.lenIncludingSentinel(), ip)),
                 .elems, .repeated_elem => {
                     var index: u64 = 0;
                     while (index < array_type.lenIncludingSentinel()) : (index += 1) {
@@ -494,14 +482,14 @@ pub fn generateSymbol(
                                 elem
                             else
                                 array_type.sentinel,
-                        }), code, reloc_parent);
+                        }), w, reloc_parent);
                     }
                 },
             },
             .vector_type => |vector_type| {
                 const abi_size = math.cast(usize, ty.abiSize(zcu)) orelse return error.Overflow;
                 if (vector_type.child == .bool_type) {
-                    const bytes = try code.addManyAsSlice(gpa, abi_size);
+                    const bytes = try w.writableSlice(abi_size);
                     @memset(bytes, 0xaa);
                     var index: usize = 0;
                     const len = math.cast(usize, vector_type.len) orelse return error.Overflow;
@@ -540,7 +528,7 @@ pub fn generateSymbol(
                     }
                 } else {
                     switch (aggregate.storage) {
-                        .bytes => |bytes| try code.appendSlice(gpa, bytes.toSlice(vector_type.len, ip)),
+                        .bytes => |bytes| try w.writeAll(bytes.toSlice(vector_type.len, ip)),
                         .elems, .repeated_elem => {
                             var index: u64 = 0;
                             while (index < vector_type.len) : (index += 1) {
@@ -550,7 +538,7 @@ pub fn generateSymbol(
                                         math.cast(usize, index) orelse return error.Overflow
                                     ],
                                     .repeated_elem => |elem| elem,
-                                }), code, reloc_parent);
+                                }), w, reloc_parent);
                             }
                         },
                     }
@@ -558,11 +546,11 @@ pub fn generateSymbol(
                     const padding = abi_size -
                         (math.cast(usize, Type.fromInterned(vector_type.child).abiSize(zcu) * vector_type.len) orelse
                             return error.Overflow);
-                    if (padding > 0) try code.appendNTimes(gpa, 0, padding);
+                    if (padding > 0) try w.splatByteAll(0, padding);
                 }
             },
             .tuple_type => |tuple| {
-                const struct_begin = code.items.len;
+                const struct_begin = w.end;
                 for (
                     tuple.types.get(ip),
                     tuple.values.get(ip),
@@ -580,8 +568,8 @@ pub fn generateSymbol(
                         .repeated_elem => |elem| elem,
                     };
 
-                    try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(field_val), code, reloc_parent);
+                    try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(field_val), w, reloc_parent);
-                    const unpadded_field_end = code.items.len - struct_begin;
+                    const unpadded_field_end = w.end - struct_begin;
 
                     // Pad struct members if required
                     const padded_field_end = ty.structFieldOffset(index + 1, zcu);
@@ -589,7 +577,7 @@ pub fn generateSymbol(
                         return error.Overflow;
 
                     if (padding > 0) {
-                        try code.appendNTimes(gpa, 0, padding);
+                        try w.splatByteAll(0, padding);
                     }
                 }
             },
@@ -598,8 +586,9 @@ pub fn generateSymbol(
                 switch (struct_type.layout) {
                     .@"packed" => {
                         const abi_size = math.cast(usize, ty.abiSize(zcu)) orelse return error.Overflow;
-                        const current_pos = code.items.len;
+                        const start = w.end;
-                        try code.appendNTimes(gpa, 0, abi_size);
+                        const buffer = try w.writableSlice(abi_size);
+                        @memset(buffer, 0);
                         var bits: u16 = 0;
 
                         for (struct_type.field_types.get(ip), 0..) |field_ty, index| {
@@ -619,22 +608,20 @@ pub fn generateSymbol(
                                     error.DivisionByZero => unreachable,
                                     error.UnexpectedRemainder => return error.RelocationNotByteAligned,
                                 };
-                                code.items.len = current_pos + field_offset;
+                                w.end = start + field_offset;
-                                // TODO: code.lockPointers();
                                 defer {
-                                    assert(code.items.len == current_pos + field_offset + @divExact(target.ptrBitWidth(), 8));
+                                    assert(w.end == start + field_offset + @divExact(target.ptrBitWidth(), 8));
-                                    // TODO: code.unlockPointers();
+                                    w.end = start + abi_size;
-                                    code.items.len = current_pos + abi_size;
                                 }
-                                try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(field_val), code, reloc_parent);
+                                try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(field_val), w, reloc_parent);
                             } else {
-                                Value.fromInterned(field_val).writeToPackedMemory(Type.fromInterned(field_ty), pt, code.items[current_pos..], bits) catch unreachable;
+                                Value.fromInterned(field_val).writeToPackedMemory(.fromInterned(field_ty), pt, buffer, bits) catch unreachable;
                             }
                             bits += @intCast(Type.fromInterned(field_ty).bitSize(zcu));
                         }
                     },
                     .auto, .@"extern" => {
-                        const struct_begin = code.items.len;
+                        const struct_begin = w.end;
                         const field_types = struct_type.field_types.get(ip);
                         const offsets = struct_type.offsets.get(ip);
 
@@ -654,11 +641,11 @@ pub fn generateSymbol(
 
                             const padding = math.cast(
                                 usize,
-                                offsets[field_index] - (code.items.len - struct_begin),
+                                offsets[field_index] - (w.end - struct_begin),
                             ) orelse return error.Overflow;
-                            if (padding > 0) try code.appendNTimes(gpa, 0, padding);
+                            if (padding > 0) try w.splatByteAll(0, padding);
 
-                            try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(field_val), code, reloc_parent);
+                            try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(field_val), w, reloc_parent);
                         }
 
                         const size = struct_type.sizeUnordered(ip);
@@ -666,10 +653,9 @@ pub fn generateSymbol(
 
                         const padding = math.cast(
                             usize,
-                            std.mem.alignForward(u64, size, @max(alignment, 1)) -
+                            std.mem.alignForward(u64, size, @max(alignment, 1)) - (w.end - struct_begin),
-                                (code.items.len - struct_begin),
                         ) orelse return error.Overflow;
-                        if (padding > 0) try code.appendNTimes(gpa, 0, padding);
+                        if (padding > 0) try w.splatByteAll(0, padding);
                     },
                 }
             },
@@ -679,12 +665,12 @@ pub fn generateSymbol(
             const layout = ty.unionGetLayout(zcu);
 
             if (layout.payload_size == 0) {
-                return generateSymbol(bin_file, pt, src_loc, Value.fromInterned(un.tag), code, reloc_parent);
+                return generateSymbol(bin_file, pt, src_loc, Value.fromInterned(un.tag), w, reloc_parent);
             }
 
             // Check if we should store the tag first.
             if (layout.tag_size > 0 and layout.tag_align.compare(.gte, layout.payload_align)) {
-                try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(un.tag), code, reloc_parent);
+                try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(un.tag), w, reloc_parent);
             }
 
             const union_obj = zcu.typeToUnion(ty).?;
@@ -692,24 +678,24 @@ pub fn generateSymbol(
                 const field_index = ty.unionTagFieldIndex(Value.fromInterned(un.tag), zcu).?;
                 const field_ty = Type.fromInterned(union_obj.field_types.get(ip)[field_index]);
                 if (!field_ty.hasRuntimeBits(zcu)) {
-                    try code.appendNTimes(gpa, 0xaa, math.cast(usize, layout.payload_size) orelse return error.Overflow);
+                    try w.splatByteAll(0xaa, math.cast(usize, layout.payload_size) orelse return error.Overflow);
                 } else {
-                    try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(un.val), code, reloc_parent);
+                    try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(un.val), w, reloc_parent);
 
                     const padding = math.cast(usize, layout.payload_size - field_ty.abiSize(zcu)) orelse return error.Overflow;
                     if (padding > 0) {
-                        try code.appendNTimes(gpa, 0, padding);
+                        try w.splatByteAll(0, padding);
                     }
                 }
             } else {
-                try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(un.val), code, reloc_parent);
+                try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(un.val), w, reloc_parent);
             }
 
             if (layout.tag_size > 0 and layout.tag_align.compare(.lt, layout.payload_align)) {
-                try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(un.tag), code, reloc_parent);
+                try generateSymbol(bin_file, pt, src_loc, Value.fromInterned(un.tag), w, reloc_parent);
 
                 if (layout.padding > 0) {
-                    try code.appendNTimes(gpa, 0, layout.padding);
+                    try w.splatByteAll(0, layout.padding);
                 }
             }
         },
@@ -722,30 +708,30 @@ fn lowerPtr(
     pt: Zcu.PerThread,
     src_loc: Zcu.LazySrcLoc,
     ptr_val: InternPool.Index,
-    code: *ArrayList(u8),
+    w: *std.Io.Writer,
     reloc_parent: link.File.RelocInfo.Parent,
     prev_offset: u64,
-) GenerateSymbolError!void {
+) (GenerateSymbolError || std.Io.Writer.Error)!void {
     const zcu = pt.zcu;
     const ptr = zcu.intern_pool.indexToKey(ptr_val).ptr;
     const offset: u64 = prev_offset + ptr.byte_offset;
     return switch (ptr.base_addr) {
-        .nav => |nav| try lowerNavRef(bin_file, pt, nav, code, reloc_parent, offset),
+        .nav => |nav| try lowerNavRef(bin_file, pt, nav, w, reloc_parent, offset),
-        .uav => |uav| try lowerUavRef(bin_file, pt, src_loc, uav, code, reloc_parent, offset),
+        .uav => |uav| try lowerUavRef(bin_file, pt, src_loc, uav, w, reloc_parent, offset),
-        .int => try generateSymbol(bin_file, pt, src_loc, try pt.intValue(Type.usize, offset), code, reloc_parent),
+        .int => try generateSymbol(bin_file, pt, src_loc, try pt.intValue(Type.usize, offset), w, reloc_parent),
         .eu_payload => |eu_ptr| try lowerPtr(
             bin_file,
             pt,
             src_loc,
             eu_ptr,
-            code,
+            w,
             reloc_parent,
             offset + errUnionPayloadOffset(
                 Value.fromInterned(eu_ptr).typeOf(zcu).childType(zcu).errorUnionPayload(zcu),
                 zcu,
             ),
         ),
-        .opt_payload => |opt_ptr| try lowerPtr(bin_file, pt, src_loc, opt_ptr, code, reloc_parent, offset),
+        .opt_payload => |opt_ptr| try lowerPtr(bin_file, pt, src_loc, opt_ptr, w, reloc_parent, offset),
         .field => |field| {
             const base_ptr = Value.fromInterned(field.base);
             const base_ty = base_ptr.typeOf(zcu).childType(zcu);
@@ -764,7 +750,7 @@ fn lowerPtr(
                 },
                 else => unreachable,
             };
-            return lowerPtr(bin_file, pt, src_loc, field.base, code, reloc_parent, offset + field_off);
+            return lowerPtr(bin_file, pt, src_loc, field.base, w, reloc_parent, offset + field_off);
         },
         .arr_elem, .comptime_field, .comptime_alloc => unreachable,
     };
@@ -775,12 +761,11 @@ fn lowerUavRef(
     pt: Zcu.PerThread,
     src_loc: Zcu.LazySrcLoc,
     uav: InternPool.Key.Ptr.BaseAddr.Uav,
-    code: *ArrayList(u8),
+    w: *std.Io.Writer,
     reloc_parent: link.File.RelocInfo.Parent,
     offset: u64,
-) GenerateSymbolError!void {
+) (GenerateSymbolError || std.Io.Writer.Error)!void {
     const zcu = pt.zcu;
-    const gpa = zcu.gpa;
     const ip = &zcu.intern_pool;
     const comp = lf.comp;
     const target = &comp.root_mod.resolved_target.result;
@@ -790,10 +775,9 @@ fn lowerUavRef(
     const is_fn_body = uav_ty.zigTypeTag(zcu) == .@"fn";
 
     log.debug("lowerUavRef: ty = {f}", .{uav_ty.fmt(pt)});
-    try code.ensureUnusedCapacity(gpa, ptr_width_bytes);
 
     if (!is_fn_body and !uav_ty.hasRuntimeBits(zcu)) {
-        code.appendNTimesAssumeCapacity(0xaa, ptr_width_bytes);
+        try w.splatByteAll(0xaa, ptr_width_bytes);
         return;
     }
 
@@ -804,29 +788,32 @@ fn lowerUavRef(
             dev.check(link.File.Tag.wasm.devFeature());
             const wasm = lf.cast(.wasm).?;
             assert(reloc_parent == .none);
-            try wasm.addUavReloc(code.items.len, uav.val, uav.orig_ty, @intCast(offset));
+            try wasm.addUavReloc(w.end, uav.val, uav.orig_ty, @intCast(offset));
-            code.appendNTimesAssumeCapacity(0, ptr_width_bytes);
+            try w.splatByteAll(0, ptr_width_bytes);
             return;
         },
         else => {},
     }
 
-    const uav_align = ip.indexToKey(uav.orig_ty).ptr_type.flags.alignment;
+    const uav_align = Type.fromInterned(uav.orig_ty).ptrAlignment(zcu);
     switch (try lf.lowerUav(pt, uav_val, uav_align, src_loc)) {
         .sym_index => {},
         .fail => |em| std.debug.panic("TODO rework lowerUav. internal error: {s}", .{em.msg}),
     }
 
-    const vaddr = try lf.getUavVAddr(uav_val, .{
+    const vaddr = lf.getUavVAddr(uav_val, .{
         .parent = reloc_parent,
-        .offset = code.items.len,
+        .offset = w.end,
         .addend = @intCast(offset),
-    });
+    }) catch |err| switch (err) {
+        error.OutOfMemory => return error.OutOfMemory,
+        else => |e| std.debug.panic("TODO rework lowerUav. internal error: {t}", .{e}),
+    };
     const endian = target.cpu.arch.endian();
     switch (ptr_width_bytes) {
-        2 => mem.writeInt(u16, code.addManyAsArrayAssumeCapacity(2), @intCast(vaddr), endian),
+        2 => try w.writeInt(u16, @intCast(vaddr), endian),
-        4 => mem.writeInt(u32, code.addManyAsArrayAssumeCapacity(4), @intCast(vaddr), endian),
+        4 => try w.writeInt(u32, @intCast(vaddr), endian),
-        8 => mem.writeInt(u64, code.addManyAsArrayAssumeCapacity(8), vaddr, endian),
+        8 => try w.writeInt(u64, vaddr, endian),
         else => unreachable,
     }
 }
@@ -835,10 +822,10 @@ fn lowerNavRef(
     lf: *link.File,
     pt: Zcu.PerThread,
     nav_index: InternPool.Nav.Index,
-    code: *ArrayList(u8),
+    w: *std.Io.Writer,
     reloc_parent: link.File.RelocInfo.Parent,
     offset: u64,
-) GenerateSymbolError!void {
+) (GenerateSymbolError || std.Io.Writer.Error)!void {
     const zcu = pt.zcu;
     const gpa = zcu.gpa;
     const ip = &zcu.intern_pool;
@@ -848,10 +835,8 @@ fn lowerNavRef(
     const nav_ty = Type.fromInterned(ip.getNav(nav_index).typeOf(ip));
     const is_fn_body = nav_ty.zigTypeTag(zcu) == .@"fn";
 
-    try code.ensureUnusedCapacity(gpa, ptr_width_bytes);
-
     if (!is_fn_body and !nav_ty.hasRuntimeBits(zcu)) {
-        code.appendNTimesAssumeCapacity(0xaa, ptr_width_bytes);
+        try w.splatByteAll(0xaa, ptr_width_bytes);
         return;
     }
 
@@ -870,13 +855,13 @@ fn lowerNavRef(
                 } else {
                     try wasm.func_table_fixups.append(gpa, .{
                         .table_index = @enumFromInt(gop.index),
-                        .offset = @intCast(code.items.len),
+                        .offset = @intCast(w.end),
                     });
                 }
             } else {
                 if (is_obj) {
                     try wasm.out_relocs.append(gpa, .{
-                        .offset = @intCast(code.items.len),
+                        .offset = @intCast(w.end),
                         .pointee = .{ .symbol_index = try wasm.navSymbolIndex(nav_index) },
                         .tag = if (ptr_width_bytes == 4) .memory_addr_i32 else .memory_addr_i64,
                         .addend = @intCast(offset),
@@ -885,12 +870,12 @@ fn lowerNavRef(
                     try wasm.nav_fixups.ensureUnusedCapacity(gpa, 1);
                     wasm.nav_fixups.appendAssumeCapacity(.{
                         .navs_exe_index = try wasm.refNavExe(nav_index),
-                        .offset = @intCast(code.items.len),
+                        .offset = @intCast(w.end),
                         .addend = @intCast(offset),
                     });
                 }
             }
-            code.appendNTimesAssumeCapacity(0, ptr_width_bytes);
+            try w.splatByteAll(0, ptr_width_bytes);
             return;
         },
         else => {},
@@ -898,14 +883,14 @@ fn lowerNavRef(
 
     const vaddr = lf.getNavVAddr(pt, nav_index, .{
         .parent = reloc_parent,
-        .offset = code.items.len,
+        .offset = w.end,
         .addend = @intCast(offset),
     }) catch @panic("TODO rework getNavVAddr");
     const endian = target.cpu.arch.endian();
     switch (ptr_width_bytes) {
-        2 => mem.writeInt(u16, code.addManyAsArrayAssumeCapacity(2), @intCast(vaddr), endian),
+        2 => try w.writeInt(u16, @intCast(vaddr), endian),
-        4 => mem.writeInt(u32, code.addManyAsArrayAssumeCapacity(4), @intCast(vaddr), endian),
+        4 => try w.writeInt(u32, @intCast(vaddr), endian),
-        8 => mem.writeInt(u64, code.addManyAsArrayAssumeCapacity(8), vaddr, endian),
+        8 => try w.writeInt(u64, vaddr, endian),
         else => unreachable,
     }
 }
@@ -962,6 +947,16 @@ pub fn genNavRef(
             },
             .link_once => unreachable,
         }
+    } else if (lf.cast(.elf2)) |elf| {
+        return .{ .sym_index = @intFromEnum(elf.navSymbol(zcu, nav_index) catch |err| switch (err) {
+            error.OutOfMemory => return error.OutOfMemory,
+            else => |e| return .{ .fail = try ErrorMsg.create(
+                zcu.gpa,
+                src_loc,
+                "linker failed to create a nav: {t}",
+                .{e},
+            ) },
+        }) };
     } else if (lf.cast(.macho)) |macho_file| {
         const zo = macho_file.getZigObject().?;
         switch (linkage) {

src/codegen/aarch64/Mir.zig

@@ -56,13 +56,13 @@ pub fn emit(
     pt: Zcu.PerThread,
     src_loc: Zcu.LazySrcLoc,
     func_index: InternPool.Index,
-    code: *std.ArrayListUnmanaged(u8),
+    atom_index: u32,
+    w: *std.Io.Writer,
     debug_output: link.File.DebugInfoOutput,
 ) !void {
     _ = debug_output;
     const zcu = pt.zcu;
     const ip = &zcu.intern_pool;
-    const gpa = zcu.gpa;
     const func = zcu.funcInfo(func_index);
     const nav = ip.getNav(func.owner_nav);
     const mod = zcu.navFileScope(func.owner_nav).mod.?;
@@ -81,20 +81,19 @@ pub fn emit(
         @as(u5, @intCast(func_align.minStrict(.@"16").toByteUnits().?)),
         Instruction.size,
     ) - 1);
-    try code.ensureUnusedCapacity(gpa, Instruction.size *
+    try w.rebase(w.end, Instruction.size * (code_len + literals_align_gap + mir.literals.len));
-        (code_len + literals_align_gap + mir.literals.len));
+    emitInstructionsForward(w, mir.prologue) catch unreachable;
-    emitInstructionsForward(code, mir.prologue);
+    emitInstructionsBackward(w, mir.body) catch unreachable;
-    emitInstructionsBackward(code, mir.body);
+    const body_end: u32 = @intCast(w.end);
-    const body_end: u32 = @intCast(code.items.len);
+    emitInstructionsBackward(w, mir.epilogue) catch unreachable;
-    emitInstructionsBackward(code, mir.epilogue);
+    w.splatByteAll(0, Instruction.size * literals_align_gap) catch unreachable;
-    code.appendNTimesAssumeCapacity(0, Instruction.size * literals_align_gap);
+    w.writeAll(@ptrCast(mir.literals)) catch unreachable;
-    code.appendSliceAssumeCapacity(@ptrCast(mir.literals));
     mir_log.debug("", .{});
 
     for (mir.nav_relocs) |nav_reloc| try emitReloc(
         lf,
         zcu,
-        func.owner_nav,
+        atom_index,
         switch (try @import("../../codegen.zig").genNavRef(
             lf,
             pt,
@@ -112,7 +111,7 @@ pub fn emit(
     for (mir.uav_relocs) |uav_reloc| try emitReloc(
         lf,
         zcu,
-        func.owner_nav,
+        atom_index,
         switch (try lf.lowerUav(
             pt,
             uav_reloc.uav.val,
@@ -129,7 +128,7 @@ pub fn emit(
     for (mir.lazy_relocs) |lazy_reloc| try emitReloc(
         lf,
         zcu,
-        func.owner_nav,
+        atom_index,
         if (lf.cast(.elf)) |ef|
             ef.zigObjectPtr().?.getOrCreateMetadataForLazySymbol(ef, pt, lazy_reloc.symbol) catch |err|
                 return zcu.codegenFail(func.owner_nav, "{s} creating lazy symbol", .{@errorName(err)})
@@ -150,7 +149,7 @@ pub fn emit(
     for (mir.global_relocs) |global_reloc| try emitReloc(
         lf,
         zcu,
-        func.owner_nav,
+        atom_index,
         if (lf.cast(.elf)) |ef|
             try ef.getGlobalSymbol(std.mem.span(global_reloc.name), null)
         else if (lf.cast(.macho)) |mf|
@@ -168,30 +167,30 @@ pub fn emit(
         var instruction = mir.body[literal_reloc.label];
         instruction.load_store.register_literal.group.imm19 += literal_reloc_offset;
         instruction.write(
-            code.items[body_end - Instruction.size * (1 + literal_reloc.label) ..][0..Instruction.size],
+            w.buffered()[body_end - Instruction.size * (1 + literal_reloc.label) ..][0..Instruction.size],
         );
     }
 }
 
-fn emitInstructionsForward(code: *std.ArrayListUnmanaged(u8), instructions: []const Instruction) void {
+fn emitInstructionsForward(w: *std.Io.Writer, instructions: []const Instruction) !void {
-    for (instructions) |instruction| emitInstruction(code, instruction);
+    for (instructions) |instruction| try emitInstruction(w, instruction);
 }
-fn emitInstructionsBackward(code: *std.ArrayListUnmanaged(u8), instructions: []const Instruction) void {
+fn emitInstructionsBackward(w: *std.Io.Writer, instructions: []const Instruction) !void {
     var instruction_index = instructions.len;
     while (instruction_index > 0) {
         instruction_index -= 1;
-        emitInstruction(code, instructions[instruction_index]);
+        try emitInstruction(w, instructions[instruction_index]);
     }
 }
-fn emitInstruction(code: *std.ArrayListUnmanaged(u8), instruction: Instruction) void {
+fn emitInstruction(w: *std.Io.Writer, instruction: Instruction) !void {
     mir_log.debug("    {f}", .{instruction});
-    instruction.write(code.addManyAsArrayAssumeCapacity(Instruction.size));
+    instruction.write(try w.writableArray(Instruction.size));
 }
 
 fn emitReloc(
     lf: *link.File,
     zcu: *Zcu,
-    owner_nav: InternPool.Nav.Index,
+    atom_index: u32,
     sym_index: u32,
     instruction: Instruction,
     offset: u32,
@@ -202,7 +201,7 @@ fn emitReloc(
         else => unreachable,
         .data_processing_immediate => |decoded| if (lf.cast(.elf)) |ef| {
             const zo = ef.zigObjectPtr().?;
-            const atom = zo.symbol(try zo.getOrCreateMetadataForNav(zcu, owner_nav)).atom(ef).?;
+            const atom = zo.symbol(atom_index).atom(ef).?;
             const r_type: std.elf.R_AARCH64 = switch (decoded.decode()) {
                 else => unreachable,
                 .pc_relative_addressing => |pc_relative_addressing| switch (pc_relative_addressing.group.op) {
@@ -221,7 +220,7 @@ fn emitReloc(
             }, zo);
         } else if (lf.cast(.macho)) |mf| {
             const zo = mf.getZigObject().?;
-            const atom = zo.symbols.items[try zo.getOrCreateMetadataForNav(mf, owner_nav)].getAtom(mf).?;
+            const atom = zo.symbols.items[atom_index].getAtom(mf).?;
             switch (decoded.decode()) {
                 else => unreachable,
                 .pc_relative_addressing => |pc_relative_addressing| switch (pc_relative_addressing.group.op) {
@@ -260,7 +259,7 @@ fn emitReloc(
         },
         .branch_exception_generating_system => |decoded| if (lf.cast(.elf)) |ef| {
             const zo = ef.zigObjectPtr().?;
-            const atom = zo.symbol(try zo.getOrCreateMetadataForNav(zcu, owner_nav)).atom(ef).?;
+            const atom = zo.symbol(atom_index).atom(ef).?;
             const r_type: std.elf.R_AARCH64 = switch (decoded.decode().unconditional_branch_immediate.group.op) {
                 .b => .JUMP26,
                 .bl => .CALL26,
@@ -272,7 +271,7 @@ fn emitReloc(
             }, zo);
         } else if (lf.cast(.macho)) |mf| {
             const zo = mf.getZigObject().?;
-            const atom = zo.symbols.items[try zo.getOrCreateMetadataForNav(mf, owner_nav)].getAtom(mf).?;
+            const atom = zo.symbols.items[atom_index].getAtom(mf).?;
             try atom.addReloc(mf, .{
                 .tag = .@"extern",
                 .offset = offset,
@@ -289,7 +288,7 @@ fn emitReloc(
         },
         .load_store => |decoded| if (lf.cast(.elf)) |ef| {
             const zo = ef.zigObjectPtr().?;
-            const atom = zo.symbol(try zo.getOrCreateMetadataForNav(zcu, owner_nav)).atom(ef).?;
+            const atom = zo.symbol(atom_index).atom(ef).?;
             const r_type: std.elf.R_AARCH64 = switch (decoded.decode().register_unsigned_immediate.decode()) {
                 .integer => |integer| switch (integer.decode()) {
                     .unallocated, .prfm => unreachable,
@@ -316,7 +315,7 @@ fn emitReloc(
             }, zo);
         } else if (lf.cast(.macho)) |mf| {
             const zo = mf.getZigObject().?;
-            const atom = zo.symbols.items[try zo.getOrCreateMetadataForNav(mf, owner_nav)].getAtom(mf).?;
+            const atom = zo.symbols.items[atom_index].getAtom(mf).?;
             try atom.addReloc(mf, .{
                 .tag = .@"extern",
                 .offset = offset,

src/dev.zig

@@ -97,6 +97,7 @@ pub const Env = enum {
                 .lld_linker,
                 .coff_linker,
                 .elf_linker,
+                .elf2_linker,
                 .macho_linker,
                 .c_linker,
                 .wasm_linker,
@@ -163,6 +164,7 @@ pub const Env = enum {
                 .incremental,
                 .aarch64_backend,
                 .elf_linker,
+                .elf2_linker,
                 => true,
                 else => Env.sema.supports(feature),
             },
@@ -210,6 +212,7 @@ pub const Env = enum {
                 .legalize,
                 .x86_64_backend,
                 .elf_linker,
+                .elf2_linker,
                 => true,
                 else => Env.sema.supports(feature),
             },
@@ -282,6 +285,7 @@ pub const Feature = enum {
     lld_linker,
     coff_linker,
     elf_linker,
+    elf2_linker,
     macho_linker,
     c_linker,
     wasm_linker,

src/link.zig

@@ -219,6 +219,7 @@ pub const Diags = struct {
     }
 
     pub fn addError(diags: *Diags, comptime format: []const u8, args: anytype) void {
+        @branchHint(.cold);
         return addErrorSourceLocation(diags, .none, format, args);
     }
 
@@ -529,7 +530,7 @@ pub const File = struct {
             const lld: *Lld = try .createEmpty(arena, comp, emit, options);
             return &lld.base;
         }
-        switch (Tag.fromObjectFormat(comp.root_mod.resolved_target.result.ofmt)) {
+        switch (Tag.fromObjectFormat(comp.root_mod.resolved_target.result.ofmt, comp.config.use_new_linker)) {
             .plan9 => return error.UnsupportedObjectFormat,
             inline else => |tag| {
                 dev.check(tag.devFeature());
@@ -552,7 +553,7 @@ pub const File = struct {
             const lld: *Lld = try .createEmpty(arena, comp, emit, options);
             return &lld.base;
         }
-        switch (Tag.fromObjectFormat(comp.root_mod.resolved_target.result.ofmt)) {
+        switch (Tag.fromObjectFormat(comp.root_mod.resolved_target.result.ofmt, comp.config.use_new_linker)) {
             .plan9 => return error.UnsupportedObjectFormat,
             inline else => |tag| {
                 dev.check(tag.devFeature());
@@ -579,7 +580,8 @@ pub const File = struct {
                 const emit = base.emit;
                 if (base.child_pid) |pid| {
                     if (builtin.os.tag == .windows) {
-                        base.cast(.coff).?.ptraceAttach(pid) catch |err| {
+                        const coff_file = base.cast(.coff).?;
+                        coff_file.ptraceAttach(pid) catch |err| {
                             log.warn("attaching failed with error: {s}", .{@errorName(err)});
                         };
                     } else {
@@ -597,8 +599,11 @@ pub const File = struct {
                             .linux => std.posix.ptrace(std.os.linux.PTRACE.ATTACH, pid, 0, 0) catch |err| {
                                 log.warn("ptrace failure: {s}", .{@errorName(err)});
                             },
-                            .macos => base.cast(.macho).?.ptraceAttach(pid) catch |err| {
+                            .macos => {
+                                const macho_file = base.cast(.macho).?;
+                                macho_file.ptraceAttach(pid) catch |err| {
                                     log.warn("attaching failed with error: {s}", .{@errorName(err)});
+                                };
                             },
                             .windows => unreachable,
                             else => return error.HotSwapUnavailableOnHostOperatingSystem,
@@ -613,6 +618,20 @@ pub const File = struct {
                     .mode = determineMode(output_mode, link_mode),
                 });
             },
+            .elf2 => {
+                const elf = base.cast(.elf2).?;
+                if (base.file == null) {
+                    elf.mf.file = try base.emit.root_dir.handle.createFile(base.emit.sub_path, .{
+                        .truncate = false,
+                        .read = true,
+                        .mode = determineMode(comp.config.output_mode, comp.config.link_mode),
+                    });
+                    base.file = elf.mf.file;
+                    try elf.mf.ensureTotalCapacity(
+                        @intCast(elf.mf.nodes.items[0].location().resolve(&elf.mf)[1]),
+                    );
+                }
+            },
             .c, .spirv => dev.checkAny(&.{ .c_linker, .spirv_linker }),
             .plan9 => unreachable,
         }
@@ -669,14 +688,30 @@ pub const File = struct {
 
                 if (base.child_pid) |pid| {
                     switch (builtin.os.tag) {
-                        .macos => base.cast(.macho).?.ptraceDetach(pid) catch |err| {
+                        .macos => {
+                            const macho_file = base.cast(.macho).?;
+                            macho_file.ptraceDetach(pid) catch |err| {
                                 log.warn("detaching failed with error: {s}", .{@errorName(err)});
+                            };
+                        },
+                        .windows => {
+                            const coff_file = base.cast(.coff).?;
+                            coff_file.ptraceDetach(pid);
                         },
-                        .windows => base.cast(.coff).?.ptraceDetach(pid),
                         else => return error.HotSwapUnavailableOnHostOperatingSystem,
                     }
                 }
             },
+            .elf2 => {
+                const elf = base.cast(.elf2).?;
+                if (base.file) |f| {
+                    elf.mf.unmap();
+                    assert(elf.mf.file.handle == f.handle);
+                    elf.mf.file = undefined;
+                    f.close();
+                    base.file = null;
+                }
+            },
             .c, .spirv => dev.checkAny(&.{ .c_linker, .spirv_linker }),
             .plan9 => unreachable,
         }
@@ -793,6 +828,7 @@ pub const File = struct {
             .spirv => {},
             .goff, .xcoff => {},
             .plan9 => unreachable,
+            .elf2 => {},
             inline else => |tag| {
                 dev.check(tag.devFeature());
                 return @as(*tag.Type(), @fieldParentPtr("base", base)).updateLineNumber(pt, ti_id);
@@ -825,6 +861,26 @@ pub const File = struct {
         }
     }
 
+    pub fn idle(base: *File, tid: Zcu.PerThread.Id) !bool {
+        switch (base.tag) {
+            else => return false,
+            inline .elf2 => |tag| {
+                dev.check(tag.devFeature());
+                return @as(*tag.Type(), @fieldParentPtr("base", base)).idle(tid);
+            },
+        }
+    }
+
+    pub fn updateErrorData(base: *File, pt: Zcu.PerThread) !void {
+        switch (base.tag) {
+            else => {},
+            inline .elf2 => |tag| {
+                dev.check(tag.devFeature());
+                return @as(*tag.Type(), @fieldParentPtr("base", base)).updateErrorData(pt);
+            },
+        }
+    }
+
     pub const FlushError = error{
         /// Indicates an error will be present in `Compilation.link_diags`.
         LinkFailure,
@@ -1099,7 +1155,7 @@ pub const File = struct {
         if (base.zcu_object_basename != null) return;
 
         switch (base.tag) {
-            inline .wasm => |tag| {
+            inline .elf2, .wasm => |tag| {
                 dev.check(tag.devFeature());
                 return @as(*tag.Type(), @fieldParentPtr("base", base)).prelink(base.comp.link_prog_node);
             },
@@ -1110,6 +1166,7 @@ pub const File = struct {
     pub const Tag = enum {
         coff,
         elf,
+        elf2,
         macho,
         c,
         wasm,
@@ -1123,6 +1180,7 @@ pub const File = struct {
             return switch (tag) {
                 .coff => Coff,
                 .elf => Elf,
+                .elf2 => Elf2,
                 .macho => MachO,
                 .c => C,
                 .wasm => Wasm,
@@ -1134,10 +1192,10 @@ pub const File = struct {
             };
         }
 
-        fn fromObjectFormat(ofmt: std.Target.ObjectFormat) Tag {
+        fn fromObjectFormat(ofmt: std.Target.ObjectFormat, use_new_linker: bool) Tag {
             return switch (ofmt) {
                 .coff => .coff,
-                .elf => .elf,
+                .elf => if (use_new_linker) .elf2 else .elf,
                 .macho => .macho,
                 .wasm => .wasm,
                 .plan9 => .plan9,
@@ -1223,6 +1281,7 @@ pub const File = struct {
     pub const C = @import("link/C.zig");
     pub const Coff = @import("link/Coff.zig");
     pub const Elf = @import("link/Elf.zig");
+    pub const Elf2 = @import("link/Elf2.zig");
     pub const MachO = @import("link/MachO.zig");
     pub const SpirV = @import("link/SpirV.zig");
     pub const Wasm = @import("link/Wasm.zig");
@@ -1548,6 +1607,9 @@ pub fn doZcuTask(comp: *Compilation, tid: usize, task: ZcuTask) void {
         }
     }
 }
+pub fn doIdleTask(comp: *Compilation, tid: usize) error{ OutOfMemory, LinkFailure }!bool {
+    return if (comp.bin_file) |lf| lf.idle(@enumFromInt(tid)) else false;
+}
 /// After the main pipeline is done, but before flush, the compilation may need to link one final
 /// `Nav` into the binary: the `builtin.test_functions` value. Since the link thread isn't running
 /// by then, we expose this function which can be called directly.
@@ -1573,6 +1635,13 @@ pub fn linkTestFunctionsNav(pt: Zcu.PerThread, nav_index: InternPool.Nav.Index)
         };
     }
 }
+pub fn updateErrorData(pt: Zcu.PerThread) void {
+    const comp = pt.zcu.comp;
+    if (comp.bin_file) |lf| lf.updateErrorData(pt) catch |err| switch (err) {
+        error.OutOfMemory => comp.link_diags.setAllocFailure(),
+        error.LinkFailure => {},
+    };
+}
 
 /// Provided by the CLI, processed into `LinkInput` instances at the start of
 /// the compilation pipeline.

src/link/Coff.zig

@@ -953,7 +953,7 @@ fn writeOffsetTableEntry(coff: *Coff, index: usize) !void {
 }
 
 fn markRelocsDirtyByTarget(coff: *Coff, target: SymbolWithLoc) void {
-    if (!coff.base.comp.incremental) return;
+    if (!coff.base.comp.config.incremental) return;
     // TODO: reverse-lookup might come in handy here
     for (coff.relocs.values()) |*relocs| {
         for (relocs.items) |*reloc| {
@@ -964,7 +964,7 @@ fn markRelocsDirtyByTarget(coff: *Coff, target: SymbolWithLoc) void {
 }
 
 fn markRelocsDirtyByAddress(coff: *Coff, addr: u32) void {
-    if (!coff.base.comp.incremental) return;
+    if (!coff.base.comp.config.incremental) return;
     const got_moved = blk: {
         const sect_id = coff.got_section_index orelse break :blk false;
         break :blk coff.sections.items(.header)[sect_id].virtual_address >= addr;
@@ -1111,20 +1111,24 @@ pub fn updateFunc(
 
     coff.navs.getPtr(func.owner_nav).?.section = coff.text_section_index.?;
 
-    var code_buffer: std.ArrayListUnmanaged(u8) = .empty;
+    var aw: std.Io.Writer.Allocating = .init(gpa);
-    defer code_buffer.deinit(gpa);
+    defer aw.deinit();
 
-    try codegen.emitFunction(
+    codegen.emitFunction(
         &coff.base,
         pt,
         zcu.navSrcLoc(nav_index),
         func_index,
+        coff.getAtom(atom_index).getSymbolIndex().?,
         mir,
-        &code_buffer,
+        &aw.writer,
         .none,
-    );
+    ) catch |err| switch (err) {
+        error.WriteFailed => return error.OutOfMemory,
+        else => |e| return e,
+    };
 
-    try coff.updateNavCode(pt, nav_index, code_buffer.items, .FUNCTION);
+    try coff.updateNavCode(pt, nav_index, aw.written(), .FUNCTION);
 
     // Exports will be updated by `Zcu.processExports` after the update.
 }
@@ -1145,18 +1149,18 @@ fn lowerConst(
 ) !LowerConstResult {
     const gpa = coff.base.comp.gpa;
 
-    var code_buffer: std.ArrayListUnmanaged(u8) = .empty;
+    var aw: std.Io.Writer.Allocating = .init(gpa);
-    defer code_buffer.deinit(gpa);
+    defer aw.deinit();
 
     const atom_index = try coff.createAtom();
     const sym = coff.getAtom(atom_index).getSymbolPtr(coff);
     try coff.setSymbolName(sym, name);
     sym.section_number = @as(coff_util.SectionNumber, @enumFromInt(sect_id + 1));
 
-    try codegen.generateSymbol(&coff.base, pt, src_loc, val, &code_buffer, .{
+    try codegen.generateSymbol(&coff.base, pt, src_loc, val, &aw.writer, .{
         .atom_index = coff.getAtom(atom_index).getSymbolIndex().?,
     });
-    const code = code_buffer.items;
+    const code = aw.written();
 
     const atom = coff.getAtomPtr(atom_index);
     atom.size = @intCast(code.len);
@@ -1170,7 +1174,7 @@ fn lowerConst(
     log.debug("allocated atom for {s} at 0x{x}", .{ name, atom.getSymbol(coff).value });
     log.debug("  (required alignment 0x{x})", .{required_alignment});
 
-    try coff.writeAtom(atom_index, code, coff.base.comp.incremental);
+    try coff.writeAtom(atom_index, code, coff.base.comp.config.incremental);
 
     return .{ .ok = atom_index };
 }
@@ -1214,19 +1218,22 @@ pub fn updateNav(
 
         coff.navs.getPtr(nav_index).?.section = coff.getNavOutputSection(nav_index);
 
-        var code_buffer: std.ArrayListUnmanaged(u8) = .empty;
+        var aw: std.Io.Writer.Allocating = .init(gpa);
-        defer code_buffer.deinit(gpa);
+        defer aw.deinit();
 
-        try codegen.generateSymbol(
+        codegen.generateSymbol(
             &coff.base,
             pt,
             zcu.navSrcLoc(nav_index),
             nav_init,
-            &code_buffer,
+            &aw.writer,
             .{ .atom_index = atom.getSymbolIndex().? },
-        );
+        ) catch |err| switch (err) {
+            error.WriteFailed => return error.OutOfMemory,
+            else => |e| return e,
+        };
 
-        try coff.updateNavCode(pt, nav_index, code_buffer.items, .NULL);
+        try coff.updateNavCode(pt, nav_index, aw.written(), .NULL);
     }
 
     // Exports will be updated by `Zcu.processExports` after the update.
@@ -1244,8 +1251,8 @@ fn updateLazySymbolAtom(
     const gpa = comp.gpa;
 
     var required_alignment: InternPool.Alignment = .none;
-    var code_buffer: std.ArrayListUnmanaged(u8) = .empty;
+    var aw: std.Io.Writer.Allocating = .init(gpa);
-    defer code_buffer.deinit(gpa);
+    defer aw.deinit();
 
     const name = try allocPrint(gpa, "__lazy_{s}_{f}", .{
         @tagName(sym.kind),
@@ -1262,11 +1269,11 @@ fn updateLazySymbolAtom(
         src,
         sym,
         &required_alignment,
-        &code_buffer,
+        &aw.writer,
         .none,
         .{ .atom_index = local_sym_index },
     );
-    const code = code_buffer.items;
+    const code = aw.written();
 
     const atom = coff.getAtomPtr(atom_index);
     const symbol = atom.getSymbolPtr(coff);
@@ -1285,7 +1292,7 @@ fn updateLazySymbolAtom(
     symbol.value = vaddr;
 
     try coff.addGotEntry(.{ .sym_index = local_sym_index });
-    try coff.writeAtom(atom_index, code, coff.base.comp.incremental);
+    try coff.writeAtom(atom_index, code, coff.base.comp.config.incremental);
 }
 
 pub fn getOrCreateAtomForLazySymbol(
@@ -1437,7 +1444,7 @@ fn updateNavCode(
         };
     }
 
-    coff.writeAtom(atom_index, code, coff.base.comp.incremental) catch |err| switch (err) {
+    coff.writeAtom(atom_index, code, coff.base.comp.config.incremental) catch |err| switch (err) {
         error.OutOfMemory => return error.OutOfMemory,
         else => |e| return coff.base.cgFail(nav_index, "failed to write atom: {s}", .{@errorName(e)}),
     };
@@ -1539,14 +1546,12 @@ pub fn updateExports(
         sym.section_number = @as(coff_util.SectionNumber, @enumFromInt(metadata.section + 1));
         sym.type = atom.getSymbol(coff).type;
 
-        switch (exp.opts.linkage) {
+        sym.storage_class = switch (exp.opts.linkage) {
-            .strong => {
+            .internal => .EXTERNAL,
-                sym.storage_class = .EXTERNAL;
+            .strong => .EXTERNAL,
-            },
-            .internal => @panic("TODO Internal"),
             .weak => @panic("TODO WeakExternal"),
             else => unreachable,
-        }
+        };
 
         try coff.resolveGlobalSymbol(sym_loc);
     }

src/link/Dwarf.zig

@@ -2126,19 +2126,22 @@ pub const WipNav = struct {
         const size = if (ty.hasRuntimeBits(wip_nav.pt.zcu)) ty.abiSize(wip_nav.pt.zcu) else 0;
         try diw.writeUleb128(size);
         if (size == 0) return;
-        var bytes = wip_nav.debug_info.toArrayList();
+        const old_end = wip_nav.debug_info.writer.end;
-        defer wip_nav.debug_info = .fromArrayList(wip_nav.dwarf.gpa, &bytes);
-        const old_len = bytes.items.len;
         try codegen.generateSymbol(
             wip_nav.dwarf.bin_file,
             wip_nav.pt,
             src_loc,
             val,
-            &bytes,
+            &wip_nav.debug_info.writer,
             .{ .debug_output = .{ .dwarf = wip_nav } },
         );
-        if (old_len + size != bytes.items.len) {
+        if (old_end + size != wip_nav.debug_info.writer.end) {
-            std.debug.print("{f} [{}]: {} != {}\n", .{ ty.fmt(wip_nav.pt), ty.toIntern(), size, bytes.items.len - old_len });
+            std.debug.print("{f} [{}]: {} != {}\n", .{
+                ty.fmt(wip_nav.pt),
+                ty.toIntern(),
+                size,
+                wip_nav.debug_info.writer.end - old_end,
+            });
             unreachable;
         }
     }
@@ -6429,7 +6432,7 @@ fn sleb128Bytes(value: anytype) u32 {
 /// overrides `-fno-incremental` for testing incremental debug info until `-fincremental` is functional
 const force_incremental = false;
 inline fn incremental(dwarf: Dwarf) bool {
-    return force_incremental or dwarf.bin_file.comp.incremental;
+    return force_incremental or dwarf.bin_file.comp.config.incremental;
 }
 
 const Allocator = std.mem.Allocator;

src/link/Elf/LinkerDefined.zig

@@ -47,7 +47,7 @@ fn newSymbolAssumeCapacity(self: *LinkerDefined, name_off: u32, elf_file: *Elf)
     const esym = self.symtab.addOneAssumeCapacity();
     esym.* = .{
         .st_name = name_off,
-        .st_info = elf.STB_WEAK << 4,
+        .st_info = @as(u8, elf.STB_WEAK) << 4,
         .st_other = @intFromEnum(elf.STV.HIDDEN),
         .st_shndx = elf.SHN_ABS,
         .st_value = 0,

src/link/Elf/SharedObject.zig

@@ -105,7 +105,7 @@ pub fn parseHeader(
         if (amt != buf.len) return error.UnexpectedEndOfFile;
     }
     if (!mem.eql(u8, ehdr.e_ident[0..4], "\x7fELF")) return error.BadMagic;
-    if (ehdr.e_ident[elf.EI_VERSION] != 1) return error.BadElfVersion;
+    if (ehdr.e_ident[elf.EI.VERSION] != 1) return error.BadElfVersion;
     if (ehdr.e_type != elf.ET.DYN) return error.NotSharedObject;
 
     if (target.toElfMachine() != ehdr.e_machine)

src/link/Elf/ZigObject.zig

@@ -277,8 +277,8 @@ pub fn flush(self: *ZigObject, elf_file: *Elf, tid: Zcu.PerThread.Id) !void {
             pt,
             .{ .kind = .code, .ty = .anyerror_type },
             metadata.text_symbol_index,
-        ) catch |err| return switch (err) {
+        ) catch |err| switch (err) {
-            error.CodegenFail => error.LinkFailure,
+            error.CodegenFail => return error.LinkFailure,
             else => |e| return e,
         };
         if (metadata.rodata_state != .unused) self.updateLazySymbol(
@@ -286,8 +286,8 @@ pub fn flush(self: *ZigObject, elf_file: *Elf, tid: Zcu.PerThread.Id) !void {
             pt,
             .{ .kind = .const_data, .ty = .anyerror_type },
             metadata.rodata_symbol_index,
-        ) catch |err| return switch (err) {
+        ) catch |err| switch (err) {
-            error.CodegenFail => error.LinkFailure,
+            error.CodegenFail => return error.LinkFailure,
             else => |e| return e,
         };
     }
@@ -1533,22 +1533,26 @@ pub fn updateFunc(
     const sym_index = try self.getOrCreateMetadataForNav(zcu, func.owner_nav);
     self.atom(self.symbol(sym_index).ref.index).?.freeRelocs(self);
 
-    var code_buffer: std.ArrayListUnmanaged(u8) = .empty;
+    var aw: std.Io.Writer.Allocating = .init(gpa);
-    defer code_buffer.deinit(gpa);
+    defer aw.deinit();
 
     var debug_wip_nav = if (self.dwarf) |*dwarf| try dwarf.initWipNav(pt, func.owner_nav, sym_index) else null;
     defer if (debug_wip_nav) |*wip_nav| wip_nav.deinit();
 
-    try codegen.emitFunction(
+    codegen.emitFunction(
         &elf_file.base,
         pt,
         zcu.navSrcLoc(func.owner_nav),
         func_index,
+        sym_index,
         mir,
-        &code_buffer,
+        &aw.writer,
         if (debug_wip_nav) |*dn| .{ .dwarf = dn } else .none,
-    );
+    ) catch |err| switch (err) {
-    const code = code_buffer.items;
+        error.WriteFailed => return error.OutOfMemory,
+        else => |e| return e,
+    };
+    const code = aw.written();
 
     const shndx = try self.getNavShdrIndex(elf_file, zcu, func.owner_nav, sym_index, code);
     log.debug("setting shdr({x},{s}) for {f}", .{
@@ -1663,21 +1667,24 @@ pub fn updateNav(
         const sym_index = try self.getOrCreateMetadataForNav(zcu, nav_index);
         self.symbol(sym_index).atom(elf_file).?.freeRelocs(self);
 
-        var code_buffer: std.ArrayListUnmanaged(u8) = .empty;
+        var aw: std.Io.Writer.Allocating = .init(zcu.gpa);
-        defer code_buffer.deinit(zcu.gpa);
+        defer aw.deinit();
 
         var debug_wip_nav = if (self.dwarf) |*dwarf| try dwarf.initWipNav(pt, nav_index, sym_index) else null;
         defer if (debug_wip_nav) |*wip_nav| wip_nav.deinit();
 
-        try codegen.generateSymbol(
+        codegen.generateSymbol(
             &elf_file.base,
             pt,
             zcu.navSrcLoc(nav_index),
             Value.fromInterned(nav_init),
-            &code_buffer,
+            &aw.writer,
             .{ .atom_index = sym_index },
-        );
+        ) catch |err| switch (err) {
-        const code = code_buffer.items;
+            error.WriteFailed => return error.OutOfMemory,
+            else => |e| return e,
+        };
+        const code = aw.written();
 
         const shndx = try self.getNavShdrIndex(elf_file, zcu, nav_index, sym_index, code);
         log.debug("setting shdr({x},{s}) for {f}", .{
@@ -1722,8 +1729,8 @@ fn updateLazySymbol(
     const gpa = zcu.gpa;
 
     var required_alignment: InternPool.Alignment = .none;
-    var code_buffer: std.ArrayListUnmanaged(u8) = .empty;
+    var aw: std.Io.Writer.Allocating = .init(gpa);
-    defer code_buffer.deinit(gpa);
+    defer aw.deinit();
 
     const name_str_index = blk: {
         const name = try std.fmt.allocPrint(gpa, "__lazy_{s}_{f}", .{
@@ -1734,18 +1741,20 @@ fn updateLazySymbol(
         break :blk try self.strtab.insert(gpa, name);
     };
 
-    const src = Type.fromInterned(sym.ty).srcLocOrNull(zcu) orelse Zcu.LazySrcLoc.unneeded;
+    codegen.generateLazySymbol(
-    try codegen.generateLazySymbol(
         &elf_file.base,
         pt,
-        src,
+        Type.fromInterned(sym.ty).srcLocOrNull(zcu) orelse .unneeded,
         sym,
         &required_alignment,
-        &code_buffer,
+        &aw.writer,
         .none,
         .{ .atom_index = symbol_index },
-    );
+    ) catch |err| switch (err) {
-    const code = code_buffer.items;
+        error.WriteFailed => return error.OutOfMemory,
+        else => |e| return e,
+    };
+    const code = aw.written();
 
     const output_section_index = switch (sym.kind) {
         .code => if (self.text_index) |sym_index|
@@ -1807,21 +1816,24 @@ fn lowerConst(
 ) !codegen.SymbolResult {
     const gpa = pt.zcu.gpa;
 
-    var code_buffer: std.ArrayListUnmanaged(u8) = .empty;
+    var aw: std.Io.Writer.Allocating = .init(gpa);
-    defer code_buffer.deinit(gpa);
+    defer aw.deinit();
 
     const name_off = try self.addString(gpa, name);
     const sym_index = try self.newSymbolWithAtom(gpa, name_off);
 
-    try codegen.generateSymbol(
+    codegen.generateSymbol(
         &elf_file.base,
         pt,
         src_loc,
         val,
-        &code_buffer,
+        &aw.writer,
         .{ .atom_index = sym_index },
-    );
+    ) catch |err| switch (err) {
-    const code = code_buffer.items;
+        error.WriteFailed => return error.OutOfMemory,
+        else => |e| return e,
+    };
+    const code = aw.written();
 
     const local_sym = self.symbol(sym_index);
     const local_esym = &self.symtab.items(.elf_sym)[local_sym.esym_index];

src/link/MachO/ZigObject.zig

@@ -784,22 +784,26 @@ pub fn updateFunc(
     const sym_index = try self.getOrCreateMetadataForNav(macho_file, func.owner_nav);
     self.symbols.items[sym_index].getAtom(macho_file).?.freeRelocs(macho_file);
 
-    var code_buffer: std.ArrayListUnmanaged(u8) = .empty;
+    var aw: std.Io.Writer.Allocating = .init(gpa);
-    defer code_buffer.deinit(gpa);
+    defer aw.deinit();
 
     var debug_wip_nav = if (self.dwarf) |*dwarf| try dwarf.initWipNav(pt, func.owner_nav, sym_index) else null;
     defer if (debug_wip_nav) |*wip_nav| wip_nav.deinit();
 
-    try codegen.emitFunction(
+    codegen.emitFunction(
         &macho_file.base,
         pt,
         zcu.navSrcLoc(func.owner_nav),
         func_index,
+        sym_index,
         mir,
-        &code_buffer,
+        &aw.writer,
         if (debug_wip_nav) |*wip_nav| .{ .dwarf = wip_nav } else .none,
-    );
+    ) catch |err| switch (err) {
-    const code = code_buffer.items;
+        error.WriteFailed => return error.OutOfMemory,
+        else => |e| return e,
+    };
+    const code = aw.written();
 
     const sect_index = try self.getNavOutputSection(macho_file, zcu, func.owner_nav, code);
     const old_rva, const old_alignment = blk: {
@@ -895,21 +899,24 @@ pub fn updateNav(
         const sym_index = try self.getOrCreateMetadataForNav(macho_file, nav_index);
         self.symbols.items[sym_index].getAtom(macho_file).?.freeRelocs(macho_file);
 
-        var code_buffer: std.ArrayListUnmanaged(u8) = .empty;
+        var aw: std.Io.Writer.Allocating = .init(zcu.gpa);
-        defer code_buffer.deinit(zcu.gpa);
+        defer aw.deinit();
 
         var debug_wip_nav = if (self.dwarf) |*dwarf| try dwarf.initWipNav(pt, nav_index, sym_index) else null;
         defer if (debug_wip_nav) |*wip_nav| wip_nav.deinit();
 
-        try codegen.generateSymbol(
+        codegen.generateSymbol(
             &macho_file.base,
             pt,
             zcu.navSrcLoc(nav_index),
             Value.fromInterned(nav_init),
-            &code_buffer,
+            &aw.writer,
             .{ .atom_index = sym_index },
-        );
+        ) catch |err| switch (err) {
-        const code = code_buffer.items;
+            error.WriteFailed => return error.OutOfMemory,
+            else => |e| return e,
+        };
+        const code = aw.written();
 
         const sect_index = try self.getNavOutputSection(macho_file, zcu, nav_index, code);
         if (isThreadlocal(macho_file, nav_index))
@@ -1198,21 +1205,24 @@ fn lowerConst(
 ) !codegen.SymbolResult {
     const gpa = macho_file.base.comp.gpa;
 
-    var code_buffer: std.ArrayListUnmanaged(u8) = .empty;
+    var aw: std.Io.Writer.Allocating = .init(gpa);
-    defer code_buffer.deinit(gpa);
+    defer aw.deinit();
 
     const name_str = try self.addString(gpa, name);
     const sym_index = try self.newSymbolWithAtom(gpa, name_str, macho_file);
 
-    try codegen.generateSymbol(
+    codegen.generateSymbol(
         &macho_file.base,
         pt,
         src_loc,
         val,
-        &code_buffer,
+        &aw.writer,
         .{ .atom_index = sym_index },
-    );
+    ) catch |err| switch (err) {
-    const code = code_buffer.items;
+        error.WriteFailed => return error.OutOfMemory,
+        else => |e| return e,
+    };
+    const code = aw.written();
 
     const sym = &self.symbols.items[sym_index];
     sym.out_n_sect = output_section_index;
@@ -1349,8 +1359,8 @@ fn updateLazySymbol(
     const gpa = zcu.gpa;
 
     var required_alignment: Atom.Alignment = .none;
-    var code_buffer: std.ArrayListUnmanaged(u8) = .empty;
+    var aw: std.Io.Writer.Allocating = .init(gpa);
-    defer code_buffer.deinit(gpa);
+    defer aw.deinit();
 
     const name_str = blk: {
         const name = try std.fmt.allocPrint(gpa, "__lazy_{s}_{f}", .{
@@ -1368,11 +1378,11 @@ fn updateLazySymbol(
         src,
         lazy_sym,
         &required_alignment,
-        &code_buffer,
+        &aw.writer,
         .none,
         .{ .atom_index = symbol_index },
     );
-    const code = code_buffer.items;
+    const code = aw.written();
 
     const output_section_index = switch (lazy_sym.kind) {
         .code => macho_file.zig_text_sect_index.?,

src/link/MappedFile.zig

@@ -0,0 +1,929 @@
+file: std.fs.File,
+flags: packed struct {
+    block_size: std.mem.Alignment,
+    copy_file_range_unsupported: bool,
+    fallocate_punch_hole_unsupported: bool,
+    fallocate_insert_range_unsupported: bool,
+},
+section: if (is_windows) windows.HANDLE else void,
+contents: []align(std.heap.page_size_min) u8,
+nodes: std.ArrayList(Node),
+free_ni: Node.Index,
+large: std.ArrayList(u64),
+updates: std.ArrayList(Node.Index),
+update_prog_node: std.Progress.Node,
+writers: std.SinglyLinkedList,
+
+pub const Error = std.posix.MMapError ||
+    std.posix.MRemapError ||
+    std.fs.File.SetEndPosError ||
+    std.fs.File.CopyRangeError ||
+    error{NotFile};
+
+pub fn init(file: std.fs.File, gpa: std.mem.Allocator) !MappedFile {
+    var mf: MappedFile = .{
+        .file = file,
+        .flags = undefined,
+        .section = if (is_windows) windows.INVALID_HANDLE_VALUE else {},
+        .contents = &.{},
+        .nodes = .empty,
+        .free_ni = .none,
+        .large = .empty,
+        .updates = .empty,
+        .update_prog_node = .none,
+        .writers = .{},
+    };
+    errdefer mf.deinit(gpa);
+    const size: u64, const blksize = if (is_windows)
+        .{ try windows.GetFileSizeEx(file.handle), 1 }
+    else stat: {
+        const stat = try std.posix.fstat(mf.file.handle);
+        if (!std.posix.S.ISREG(stat.mode)) return error.PathAlreadyExists;
+        break :stat .{ @bitCast(stat.size), stat.blksize };
+    };
+    mf.flags = .{
+        .block_size = .fromByteUnits(
+            std.math.ceilPowerOfTwoAssert(usize, @max(std.heap.pageSize(), blksize)),
+        ),
+        .copy_file_range_unsupported = false,
+        .fallocate_insert_range_unsupported = false,
+        .fallocate_punch_hole_unsupported = false,
+    };
+    try mf.nodes.ensureUnusedCapacity(gpa, 1);
+    assert(try mf.addNode(gpa, .{
+        .add_node = .{
+            .size = size,
+            .fixed = true,
+        },
+    }) == Node.Index.root);
+    try mf.ensureTotalCapacity(@intCast(size));
+    return mf;
+}
+
+pub fn deinit(mf: *MappedFile, gpa: std.mem.Allocator) void {
+    mf.unmap();
+    mf.nodes.deinit(gpa);
+    mf.large.deinit(gpa);
+    mf.updates.deinit(gpa);
+    mf.update_prog_node.end();
+    assert(mf.writers.first == null);
+    mf.* = undefined;
+}
+
+pub const Node = extern struct {
+    parent: Node.Index,
+    prev: Node.Index,
+    next: Node.Index,
+    first: Node.Index,
+    last: Node.Index,
+    flags: Flags,
+    location_payload: Location.Payload,
+
+    pub const Flags = packed struct(u32) {
+        location_tag: Location.Tag,
+        alignment: std.mem.Alignment,
+        /// Whether this node can be moved.
+        fixed: bool,
+        /// Whether this node has been moved.
+        moved: bool,
+        /// Whether this node has been resized.
+        resized: bool,
+        /// Whether this node might contain non-zero bytes.
+        has_content: bool,
+        unused: @Type(.{ .int = .{
+            .signedness = .unsigned,
+            .bits = 32 - @bitSizeOf(std.mem.Alignment) - 5,
+        } }) = 0,
+    };
+
+    pub const Location = union(enum(u1)) {
+        small: extern struct {
+            /// Relative to `parent`.
+            offset: u32,
+            size: u32,
+        },
+        large: extern struct {
+            index: usize,
+            unused: @Type(.{ .int = .{
+                .signedness = .unsigned,
+                .bits = 64 - @bitSizeOf(usize),
+            } }) = 0,
+        },
+
+        pub const Tag = @typeInfo(Location).@"union".tag_type.?;
+        pub const Payload = @Type(.{ .@"union" = .{
+            .layout = .@"extern",
+            .tag_type = null,
+            .fields = @typeInfo(Location).@"union".fields,
+            .decls = &.{},
+        } });
+
+        pub fn resolve(loc: Location, mf: *const MappedFile) [2]u64 {
+            return switch (loc) {
+                .small => |small| .{ small.offset, small.size },
+                .large => |large| mf.large.items[large.index..][0..2].*,
+            };
+        }
+    };
+
+    pub const Index = enum(u32) {
+        none,
+        _,
+
+        pub const root: Node.Index = .none;
+
+        fn get(ni: Node.Index, mf: *const MappedFile) *Node {
+            return &mf.nodes.items[@intFromEnum(ni)];
+        }
+
+        pub fn childrenMoved(ni: Node.Index, gpa: std.mem.Allocator, mf: *MappedFile) !void {
+            var child_ni = ni.get(mf).last;
+            while (child_ni != .none) {
+                try child_ni.moved(gpa, mf);
+                child_ni = child_ni.get(mf).prev;
+            }
+        }
+
+        pub fn hasMoved(ni: Node.Index, mf: *const MappedFile) bool {
+            var parent_ni = ni;
+            while (parent_ni != .none) {
+                const parent = parent_ni.get(mf);
+                if (parent.flags.moved) return true;
+                parent_ni = parent.parent;
+            }
+            return false;
+        }
+        pub fn moved(ni: Node.Index, gpa: std.mem.Allocator, mf: *MappedFile) !void {
+            try mf.updates.ensureUnusedCapacity(gpa, 1);
+            ni.movedAssumeCapacity(mf);
+        }
+        pub fn cleanMoved(ni: Node.Index, mf: *const MappedFile) bool {
+            const node_moved = &ni.get(mf).flags.moved;
+            defer node_moved.* = false;
+            return node_moved.*;
+        }
+        fn movedAssumeCapacity(ni: Node.Index, mf: *MappedFile) void {
+            var parent_ni = ni;
+            while (parent_ni != .none) {
+                const parent_node = parent_ni.get(mf);
+                if (parent_node.flags.moved) return;
+                parent_ni = parent_node.parent;
+            }
+            const node = ni.get(mf);
+            node.flags.moved = true;
+            if (node.flags.resized) return;
+            mf.updates.appendAssumeCapacity(ni);
+            mf.update_prog_node.increaseEstimatedTotalItems(1);
+        }
+
+        pub fn hasResized(ni: Node.Index, mf: *const MappedFile) bool {
+            return ni.get(mf).flags.resized;
+        }
+        pub fn resized(ni: Node.Index, gpa: std.mem.Allocator, mf: *MappedFile) !void {
+            try mf.updates.ensureUnusedCapacity(gpa, 1);
+            ni.resizedAssumeCapacity(mf);
+        }
+        pub fn cleanResized(ni: Node.Index, mf: *const MappedFile) bool {
+            const node_resized = &ni.get(mf).flags.resized;
+            defer node_resized.* = false;
+            return node_resized.*;
+        }
+        fn resizedAssumeCapacity(ni: Node.Index, mf: *MappedFile) void {
+            const node = ni.get(mf);
+            if (node.flags.resized) return;
+            node.flags.resized = true;
+            if (node.flags.moved) return;
+            mf.updates.appendAssumeCapacity(ni);
+            mf.update_prog_node.increaseEstimatedTotalItems(1);
+        }
+
+        pub fn alignment(ni: Node.Index, mf: *const MappedFile) std.mem.Alignment {
+            return ni.get(mf).flags.alignment;
+        }
+
+        fn setLocationAssumeCapacity(ni: Node.Index, mf: *MappedFile, offset: u64, size: u64) void {
+            const node = ni.get(mf);
+            if (size == 0) node.flags.has_content = false;
+            switch (node.location()) {
+                .small => |small| {
+                    if (small.offset != offset) ni.movedAssumeCapacity(mf);
+                    if (small.size != size) ni.resizedAssumeCapacity(mf);
+                    if (std.math.cast(u32, offset)) |small_offset| {
+                        if (std.math.cast(u32, size)) |small_size| {
+                            node.location_payload.small = .{
+                                .offset = small_offset,
+                                .size = small_size,
+                            };
+                            return;
+                        }
+                    }
+                    defer mf.large.appendSliceAssumeCapacity(&.{ offset, size });
+                    node.flags.location_tag = .large;
+                    node.location_payload = .{ .large = .{ .index = mf.large.items.len } };
+                },
+                .large => |large| {
+                    const large_items = mf.large.items[large.index..][0..2];
+                    if (large_items[0] != offset) ni.movedAssumeCapacity(mf);
+                    if (large_items[1] != size) ni.resizedAssumeCapacity(mf);
+                    large_items.* = .{ offset, size };
+                },
+            }
+        }
+
+        pub fn location(ni: Node.Index, mf: *const MappedFile) Location {
+            return ni.get(mf).location();
+        }
+
+        pub fn fileLocation(
+            ni: Node.Index,
+            mf: *const MappedFile,
+            set_has_content: bool,
+        ) struct { offset: u64, size: u64 } {
+            var offset, const size = ni.location(mf).resolve(mf);
+            var parent_ni = ni;
+            while (true) {
+                const parent = parent_ni.get(mf);
+                if (set_has_content) parent.flags.has_content = true;
+                if (parent_ni == .none) break;
+                parent_ni = parent.parent;
+                offset += parent_ni.location(mf).resolve(mf)[0];
+            }
+            return .{ .offset = offset, .size = size };
+        }
+
+        pub fn slice(ni: Node.Index, mf: *const MappedFile) []u8 {
+            const file_loc = ni.fileLocation(mf, true);
+            return mf.contents[@intCast(file_loc.offset)..][0..@intCast(file_loc.size)];
+        }
+
+        pub fn sliceConst(ni: Node.Index, mf: *const MappedFile) []const u8 {
+            const file_loc = ni.fileLocation(mf, false);
+            return mf.contents[@intCast(file_loc.offset)..][0..@intCast(file_loc.size)];
+        }
+
+        pub fn resize(ni: Node.Index, mf: *MappedFile, gpa: std.mem.Allocator, size: u64) !void {
+            try mf.resizeNode(gpa, ni, size);
+            var writers_it = mf.writers.first;
+            while (writers_it) |writer_node| : (writers_it = writer_node.next) {
+                const w: *Node.Writer = @fieldParentPtr("writer_node", writer_node);
+                w.interface.buffer = w.ni.slice(mf);
+            }
+        }
+
+        pub fn writer(ni: Node.Index, mf: *MappedFile, gpa: std.mem.Allocator, w: *Writer) void {
+            w.* = .{
+                .gpa = gpa,
+                .mf = mf,
+                .writer_node = .{},
+                .ni = ni,
+                .interface = .{
+                    .buffer = ni.slice(mf),
+                    .vtable = &Writer.vtable,
+                },
+                .err = null,
+            };
+            mf.writers.prepend(&w.writer_node);
+        }
+    };
+
+    pub fn location(node: *const Node) Location {
+        return switch (node.flags.location_tag) {
+            inline else => |tag| @unionInit(
+                Location,
+                @tagName(tag),
+                @field(node.location_payload, @tagName(tag)),
+            ),
+        };
+    }
+
+    pub const Writer = struct {
+        gpa: std.mem.Allocator,
+        mf: *MappedFile,
+        writer_node: std.SinglyLinkedList.Node,
+        ni: Node.Index,
+        interface: std.Io.Writer,
+        err: ?Error,
+
+        pub fn deinit(w: *Writer) void {
+            assert(w.mf.writers.popFirst() == &w.writer_node);
+            w.* = undefined;
+        }
+
+        const vtable: std.Io.Writer.VTable = .{
+            .drain = drain,
+            .sendFile = sendFile,
+            .flush = std.Io.Writer.noopFlush,
+            .rebase = growingRebase,
+        };
+
+        fn drain(
+            interface: *std.Io.Writer,
+            data: []const []const u8,
+            splat: usize,
+        ) std.Io.Writer.Error!usize {
+            const pattern = data[data.len - 1];
+            const splat_len = pattern.len * splat;
+            const start_len = interface.end;
+            assert(data.len != 0);
+            for (data) |bytes| {
+                try growingRebase(interface, interface.end, bytes.len + splat_len + 1);
+                @memcpy(interface.buffer[interface.end..][0..bytes.len], bytes);
+                interface.end += bytes.len;
+            }
+            if (splat == 0) {
+                interface.end -= pattern.len;
+            } else switch (pattern.len) {
+                0 => {},
+                1 => {
+                    @memset(interface.buffer[interface.end..][0 .. splat - 1], pattern[0]);
+                    interface.end += splat - 1;
+                },
+                else => for (0..splat - 1) |_| {
+                    @memcpy(interface.buffer[interface.end..][0..pattern.len], pattern);
+                    interface.end += pattern.len;
+                },
+            }
+            return interface.end - start_len;
+        }
+
+        fn sendFile(
+            interface: *std.Io.Writer,
+            file_reader: *std.fs.File.Reader,
+            limit: std.Io.Limit,
+        ) std.Io.Writer.FileError!usize {
+            if (limit == .nothing) return 0;
+            const pos = file_reader.logicalPos();
+            const additional = if (file_reader.getSize()) |size| size - pos else |_| std.atomic.cache_line;
+            if (additional == 0) return error.EndOfStream;
+            try growingRebase(interface, interface.end, limit.minInt64(additional));
+            switch (file_reader.mode) {
+                .positional => {
+                    const fr_buf = file_reader.interface.buffered();
+                    const buf_copy_size = interface.write(fr_buf) catch unreachable;
+                    file_reader.interface.toss(buf_copy_size);
+                    if (buf_copy_size < fr_buf.len) return buf_copy_size;
+                    assert(file_reader.logicalPos() == file_reader.pos);
+
+                    const w: *Writer = @fieldParentPtr("interface", interface);
+                    const copy_size: usize = @intCast(w.mf.copyFileRange(
+                        file_reader.file,
+                        file_reader.pos,
+                        w.ni.fileLocation(w.mf, true).offset + interface.end,
+                        limit.minInt(interface.unusedCapacityLen()),
+                    ) catch |err| {
+                        w.err = err;
+                        return error.WriteFailed;
+                    });
+                    interface.end += copy_size;
+                    return copy_size;
+                },
+                .streaming,
+                .streaming_reading,
+                .positional_reading,
+                .failure,
+                => {
+                    const dest = limit.slice(interface.unusedCapacitySlice());
+                    const n = try file_reader.read(dest);
+                    interface.end += n;
+                    return n;
+                },
+            }
+        }
+
+        fn growingRebase(
+            interface: *std.Io.Writer,
+            preserve: usize,
+            unused_capacity: usize,
+        ) std.Io.Writer.Error!void {
+            _ = preserve;
+            const total_capacity = interface.end + unused_capacity;
+            if (interface.buffer.len >= total_capacity) return;
+            const w: *Writer = @fieldParentPtr("interface", interface);
+            w.ni.resize(w.mf, w.gpa, total_capacity +| total_capacity / 2) catch |err| {
+                w.err = err;
+                return error.WriteFailed;
+            };
+        }
+    };
+
+    comptime {
+        if (!std.debug.runtime_safety) std.debug.assert(@sizeOf(Node) == 32);
+    }
+};
+
+fn addNode(mf: *MappedFile, gpa: std.mem.Allocator, opts: struct {
+    parent: Node.Index = .none,
+    prev: Node.Index = .none,
+    next: Node.Index = .none,
+    offset: u64 = 0,
+    add_node: AddNodeOptions,
+}) !Node.Index {
+    if (opts.add_node.moved or opts.add_node.resized) try mf.updates.ensureUnusedCapacity(gpa, 1);
+    const offset = opts.add_node.alignment.forward(@intCast(opts.offset));
+    const location_tag: Node.Location.Tag, const location_payload: Node.Location.Payload = location: {
+        if (std.math.cast(u32, offset)) |small_offset| break :location .{ .small, .{
+            .small = .{ .offset = small_offset, .size = 0 },
+        } };
+        try mf.large.ensureUnusedCapacity(gpa, 2);
+        defer mf.large.appendSliceAssumeCapacity(&.{ offset, 0 });
+        break :location .{ .large, .{ .large = .{ .index = mf.large.items.len } } };
+    };
+    const free_ni: Node.Index, const free_node = free: switch (mf.free_ni) {
+        .none => .{ @enumFromInt(mf.nodes.items.len), mf.nodes.addOneAssumeCapacity() },
+        else => |free_ni| {
+            const free_node = free_ni.get(mf);
+            mf.free_ni = free_node.next;
+            break :free .{ free_ni, free_node };
+        },
+    };
+    free_node.* = .{
+        .parent = opts.parent,
+        .prev = opts.prev,
+        .next = opts.next,
+        .first = .none,
+        .last = .none,
+        .flags = .{
+            .location_tag = location_tag,
+            .alignment = opts.add_node.alignment,
+            .fixed = opts.add_node.fixed,
+            .moved = true,
+            .resized = true,
+            .has_content = false,
+        },
+        .location_payload = location_payload,
+    };
+    {
+        defer {
+            free_node.flags.moved = false;
+            free_node.flags.resized = false;
+        }
+        if (offset > opts.parent.location(mf).resolve(mf)[1]) try opts.parent.resize(mf, gpa, offset);
+        try free_ni.resize(mf, gpa, opts.add_node.size);
+    }
+    if (opts.add_node.moved) free_ni.movedAssumeCapacity(mf);
+    if (opts.add_node.resized) free_ni.resizedAssumeCapacity(mf);
+    return free_ni;
+}
+
+pub const AddNodeOptions = struct {
+    size: u64 = 0,
+    alignment: std.mem.Alignment = .@"1",
+    fixed: bool = false,
+    moved: bool = false,
+    resized: bool = false,
+};
+
+pub fn addOnlyChildNode(
+    mf: *MappedFile,
+    gpa: std.mem.Allocator,
+    parent_ni: Node.Index,
+    opts: AddNodeOptions,
+) !Node.Index {
+    try mf.nodes.ensureUnusedCapacity(gpa, 1);
+    const parent = parent_ni.get(mf);
+    assert(parent.first == .none and parent.last == .none);
+    const ni = try mf.addNode(gpa, .{
+        .parent = parent_ni,
+        .add_node = opts,
+    });
+    parent.first = ni;
+    parent.last = ni;
+    return ni;
+}
+
+pub fn addLastChildNode(
+    mf: *MappedFile,
+    gpa: std.mem.Allocator,
+    parent_ni: Node.Index,
+    opts: AddNodeOptions,
+) !Node.Index {
+    try mf.nodes.ensureUnusedCapacity(gpa, 1);
+    const parent = parent_ni.get(mf);
+    const ni = try mf.addNode(gpa, .{
+        .parent = parent_ni,
+        .prev = parent.last,
+        .offset = offset: switch (parent.last) {
+            .none => 0,
+            else => |last_ni| {
+                const last_offset, const last_size = last_ni.location(mf).resolve(mf);
+                break :offset last_offset + last_size;
+            },
+        },
+        .add_node = opts,
+    });
+    switch (parent.last) {
+        .none => parent.first = ni,
+        else => |last_ni| last_ni.get(mf).next = ni,
+    }
+    parent.last = ni;
+    return ni;
+}
+
+pub fn addNodeAfter(
+    mf: *MappedFile,
+    gpa: std.mem.Allocator,
+    prev_ni: Node.Index,
+    opts: AddNodeOptions,
+) !Node.Index {
+    assert(prev_ni != .none);
+    try mf.nodes.ensureUnusedCapacity(gpa, 1);
+    const prev = prev_ni.get(mf);
+    const prev_offset, const prev_size = prev.location().resolve(mf);
+    const ni = try mf.addNode(gpa, .{
+        .parent = prev.parent,
+        .prev = prev_ni,
+        .next = prev.next,
+        .offset = prev_offset + prev_size,
+        .add_node = opts,
+    });
+    switch (prev.next) {
+        .none => prev.parent.get(mf).last = ni,
+        else => |next_ni| next_ni.get(mf).prev = ni,
+    }
+    prev.next = ni;
+    return ni;
+}
+
+fn resizeNode(mf: *MappedFile, gpa: std.mem.Allocator, ni: Node.Index, requested_size: u64) !void {
+    const node = ni.get(mf);
+    var old_offset, const old_size = node.location().resolve(mf);
+    const new_size = node.flags.alignment.forward(@intCast(requested_size));
+    // Resize the entire file
+    if (ni == Node.Index.root) {
+        try mf.file.setEndPos(new_size);
+        try mf.ensureTotalCapacity(@intCast(new_size));
+        try mf.ensureCapacityForSetLocation(gpa);
+        ni.setLocationAssumeCapacity(mf, old_offset, new_size);
+        return;
+    }
+    while (true) {
+        const parent = node.parent.get(mf);
+        _, const old_parent_size = parent.location().resolve(mf);
+        const trailing_end = switch (node.next) {
+            .none => parent.location().resolve(mf)[1],
+            else => |next_ni| next_ni.location(mf).resolve(mf)[0],
+        };
+        assert(old_offset + old_size <= trailing_end);
+        // Expand the node into available trailing free space
+        if (old_offset + new_size <= trailing_end) {
+            try mf.ensureCapacityForSetLocation(gpa);
+            ni.setLocationAssumeCapacity(mf, old_offset, new_size);
+            return;
+        }
+        // Ask the filesystem driver to insert an extent into the file without copying any data
+        if (is_linux and !mf.flags.fallocate_insert_range_unsupported and
+            node.flags.alignment.order(mf.flags.block_size).compare(.gte))
+        insert_range: {
+            const last_offset, const last_size = parent.last.location(mf).resolve(mf);
+            const last_end = last_offset + last_size;
+            assert(last_end <= old_parent_size);
+            const range_size =
+                node.flags.alignment.forward(@intCast(requested_size +| requested_size / 2)) - old_size;
+            const new_parent_size = last_end + range_size;
+            if (new_parent_size > old_parent_size) {
+                try mf.resizeNode(gpa, node.parent, new_parent_size +| new_parent_size / 2);
+                continue;
+            }
+            const range_file_offset = ni.fileLocation(mf, false).offset + old_size;
+            retry: while (true) {
+                switch (linux.E.init(linux.fallocate(
+                    mf.file.handle,
+                    linux.FALLOC.FL_INSERT_RANGE,
+                    @intCast(range_file_offset),
+                    @intCast(range_size),
+                ))) {
+                    .SUCCESS => {
+                        var enclosing_ni = ni;
+                        while (enclosing_ni != .none) {
+                            try mf.ensureCapacityForSetLocation(gpa);
+                            const enclosing = enclosing_ni.get(mf);
+                            const enclosing_offset, const enclosing_size =
+                                enclosing.location().resolve(mf);
+                            enclosing_ni.setLocationAssumeCapacity(
+                                mf,
+                                enclosing_offset,
+                                enclosing_size + range_size,
+                            );
+                            var after_ni = enclosing.next;
+                            while (after_ni != .none) {
+                                try mf.ensureCapacityForSetLocation(gpa);
+                                const after = after_ni.get(mf);
+                                const after_offset, const after_size = after.location().resolve(mf);
+                                after_ni.setLocationAssumeCapacity(
+                                    mf,
+                                    range_size + after_offset,
+                                    after_size,
+                                );
+                                after_ni = after.next;
+                            }
+                            enclosing_ni = enclosing.parent;
+                        }
+                        return;
+                    },
+                    .INTR => continue :retry,
+                    .BADF, .FBIG, .INVAL => unreachable,
+                    .IO => return error.InputOutput,
+                    .NODEV => return error.NotFile,
+                    .NOSPC => return error.NoSpaceLeft,
+                    .NOSYS, .OPNOTSUPP => {
+                        mf.flags.fallocate_insert_range_unsupported = true;
+                        break :insert_range;
+                    },
+                    .PERM => return error.PermissionDenied,
+                    .SPIPE => return error.Unseekable,
+                    .TXTBSY => return error.FileBusy,
+                    else => |e| return std.posix.unexpectedErrno(e),
+                }
+            }
+        }
+        switch (node.next) {
+            .none => {
+                // As this is the last node, we simply need more space in the parent
+                const new_parent_size = old_offset + new_size;
+                try mf.resizeNode(gpa, node.parent, new_parent_size +| new_parent_size / 2);
+            },
+            else => |*next_ni_ptr| switch (node.flags.fixed) {
+                false => {
+                    // Make space at the end of the parent for this floating node
+                    const last = parent.last.get(mf);
+                    const last_offset, const last_size = last.location().resolve(mf);
+                    const new_offset = node.flags.alignment.forward(@intCast(last_offset + last_size));
+                    const new_parent_size = new_offset + new_size;
+                    if (new_parent_size > old_parent_size) {
+                        try mf.resizeNode(
+                            gpa,
+                            node.parent,
+                            new_parent_size +| new_parent_size / 2,
+                        );
+                        continue;
+                    }
+                    const next_ni = next_ni_ptr.*;
+                    next_ni.get(mf).prev = node.prev;
+                    switch (node.prev) {
+                        .none => parent.first = next_ni,
+                        else => |prev_ni| prev_ni.get(mf).next = next_ni,
+                    }
+                    last.next = ni;
+                    node.prev = parent.last;
+                    next_ni_ptr.* = .none;
+                    parent.last = ni;
+                    if (node.flags.has_content) {
+                        const parent_file_offset = node.parent.fileLocation(mf, false).offset;
+                        try mf.moveRange(
+                            parent_file_offset + old_offset,
+                            parent_file_offset + new_offset,
+                            old_size,
+                        );
+                    }
+                    old_offset = new_offset;
+                },
+                true => {
+                    // Move the next floating node to make space for this fixed node
+                    const next_ni = next_ni_ptr.*;
+                    const next = next_ni.get(mf);
+                    assert(!next.flags.fixed);
+                    const next_offset, const next_size = next.location().resolve(mf);
+                    const last = parent.last.get(mf);
+                    const last_offset, const last_size = last.location().resolve(mf);
+                    const new_offset = next.flags.alignment.forward(@intCast(
+                        @max(old_offset + new_size, last_offset + last_size),
+                    ));
+                    const new_parent_size = new_offset + next_size;
+                    if (new_parent_size > old_parent_size) {
+                        try mf.resizeNode(
+                            gpa,
+                            node.parent,
+                            new_parent_size +| new_parent_size / 2,
+                        );
+                        continue;
+                    }
+                    try mf.ensureCapacityForSetLocation(gpa);
+                    next.prev = parent.last;
+                    parent.last = next_ni;
+                    last.next = next_ni;
+                    next_ni_ptr.* = next.next;
+                    switch (next.next) {
+                        .none => {},
+                        else => |next_next_ni| next_next_ni.get(mf).prev = ni,
+                    }
+                    next.next = .none;
+                    if (node.flags.has_content) {
+                        const parent_file_offset = node.parent.fileLocation(mf, false).offset;
+                        try mf.moveRange(
+                            parent_file_offset + next_offset,
+                            parent_file_offset + new_offset,
+                            next_size,
+                        );
+                    }
+                    next_ni.setLocationAssumeCapacity(mf, new_offset, next_size);
+                },
+            },
+        }
+    }
+}
+
+fn moveRange(mf: *MappedFile, old_file_offset: u64, new_file_offset: u64, size: u64) !void {
+    // make a copy of this node at the new location
+    try mf.copyRange(old_file_offset, new_file_offset, size);
+    // delete the copy of this node at the old location
+    if (is_linux and !mf.flags.fallocate_punch_hole_unsupported and
+        size >= mf.flags.block_size.toByteUnits() * 2 - 1) while (true)
+    {
+        switch (linux.E.init(linux.fallocate(
+            mf.file.handle,
+            linux.FALLOC.FL_PUNCH_HOLE | linux.FALLOC.FL_KEEP_SIZE,
+            @intCast(old_file_offset),
+            @intCast(size),
+        ))) {
+            .SUCCESS => return,
+            .INTR => continue,
+            .BADF, .FBIG, .INVAL => unreachable,
+            .IO => return error.InputOutput,
+            .NODEV => return error.NotFile,
+            .NOSPC => return error.NoSpaceLeft,
+            .NOSYS, .OPNOTSUPP => {
+                mf.flags.fallocate_punch_hole_unsupported = true;
+                break;
+            },
+            .PERM => return error.PermissionDenied,
+            .SPIPE => return error.Unseekable,
+            .TXTBSY => return error.FileBusy,
+            else => |e| return std.posix.unexpectedErrno(e),
+        }
+    };
+    @memset(mf.contents[@intCast(old_file_offset)..][0..@intCast(size)], 0);
+}
+
+fn copyRange(mf: *MappedFile, old_file_offset: u64, new_file_offset: u64, size: u64) !void {
+    const copy_size = try mf.copyFileRange(mf.file, old_file_offset, new_file_offset, size);
+    if (copy_size < size) @memcpy(
+        mf.contents[@intCast(new_file_offset + copy_size)..][0..@intCast(size - copy_size)],
+        mf.contents[@intCast(old_file_offset + copy_size)..][0..@intCast(size - copy_size)],
+    );
+}
+
+fn copyFileRange(
+    mf: *MappedFile,
+    old_file: std.fs.File,
+    old_file_offset: u64,
+    new_file_offset: u64,
+    size: u64,
+) !u64 {
+    var remaining_size = size;
+    if (is_linux and !mf.flags.copy_file_range_unsupported) {
+        var old_file_offset_mut: i64 = @intCast(old_file_offset);
+        var new_file_offset_mut: i64 = @intCast(new_file_offset);
+        while (remaining_size >= mf.flags.block_size.toByteUnits() * 2 - 1) {
+            const copy_len = linux.copy_file_range(
+                old_file.handle,
+                &old_file_offset_mut,
+                mf.file.handle,
+                &new_file_offset_mut,
+                @intCast(remaining_size),
+                0,
+            );
+            switch (linux.E.init(copy_len)) {
+                .SUCCESS => {
+                    if (copy_len == 0) break;
+                    remaining_size -= copy_len;
+                    if (remaining_size == 0) break;
+                },
+                .INTR => continue,
+                .BADF, .FBIG, .INVAL, .OVERFLOW => unreachable,
+                .IO => return error.InputOutput,
+                .ISDIR => return error.IsDir,
+                .NOMEM => return error.SystemResources,
+                .NOSPC => return error.NoSpaceLeft,
+                .NOSYS, .OPNOTSUPP, .XDEV => {
+                    mf.flags.copy_file_range_unsupported = true;
+                    break;
+                },
+                .PERM => return error.PermissionDenied,
+                .TXTBSY => return error.FileBusy,
+                else => |e| return std.posix.unexpectedErrno(e),
+            }
+        }
+    }
+    return size - remaining_size;
+}
+
+fn ensureCapacityForSetLocation(mf: *MappedFile, gpa: std.mem.Allocator) !void {
+    try mf.large.ensureUnusedCapacity(gpa, 2);
+    try mf.updates.ensureUnusedCapacity(gpa, 1);
+}
+
+pub fn ensureTotalCapacity(mf: *MappedFile, new_capacity: usize) !void {
+    if (mf.contents.len >= new_capacity) return;
+    try mf.ensureTotalCapacityPrecise(new_capacity +| new_capacity / 2);
+}
+
+pub fn ensureTotalCapacityPrecise(mf: *MappedFile, new_capacity: usize) !void {
+    if (mf.contents.len >= new_capacity) return;
+    const aligned_capacity = mf.flags.block_size.forward(new_capacity);
+    if (!is_linux) mf.unmap() else if (mf.contents.len > 0) {
+        mf.contents = try std.posix.mremap(
+            mf.contents.ptr,
+            mf.contents.len,
+            aligned_capacity,
+            .{ .MAYMOVE = true },
+            null,
+        );
+        return;
+    }
+    if (is_windows) {
+        if (mf.section == windows.INVALID_HANDLE_VALUE) switch (windows.ntdll.NtCreateSection(
+            &mf.section,
+            windows.STANDARD_RIGHTS_REQUIRED | windows.SECTION_QUERY |
+                windows.SECTION_MAP_WRITE | windows.SECTION_MAP_READ | windows.SECTION_EXTEND_SIZE,
+            null,
+            @constCast(&@as(i64, @intCast(aligned_capacity))),
+            windows.PAGE_READWRITE,
+            windows.SEC_COMMIT,
+            mf.file.handle,
+        )) {
+            .SUCCESS => {},
+            else => return error.MemoryMappingNotSupported,
+        };
+        var contents_ptr: ?[*]align(std.heap.page_size_min) u8 = null;
+        var contents_len = aligned_capacity;
+        switch (windows.ntdll.NtMapViewOfSection(
+            mf.section,
+            windows.GetCurrentProcess(),
+            @ptrCast(&contents_ptr),
+            null,
+            0,
+            null,
+            &contents_len,
+            .ViewUnmap,
+            0,
+            windows.PAGE_READWRITE,
+        )) {
+            .SUCCESS => mf.contents = contents_ptr.?[0..contents_len],
+            else => return error.MemoryMappingNotSupported,
+        }
+    } else mf.contents = try std.posix.mmap(
+        null,
+        aligned_capacity,
+        std.posix.PROT.READ | std.posix.PROT.WRITE,
+        .{ .TYPE = if (is_linux) .SHARED_VALIDATE else .SHARED },
+        mf.file.handle,
+        0,
+    );
+}
+
+pub fn unmap(mf: *MappedFile) void {
+    if (mf.contents.len == 0) return;
+    if (is_windows)
+        _ = windows.ntdll.NtUnmapViewOfSection(windows.GetCurrentProcess(), mf.contents.ptr)
+    else
+        std.posix.munmap(mf.contents);
+    mf.contents = &.{};
+    if (is_windows and mf.section != windows.INVALID_HANDLE_VALUE) {
+        windows.CloseHandle(mf.section);
+        mf.section = windows.INVALID_HANDLE_VALUE;
+    }
+}
+
+fn verify(mf: *MappedFile) void {
+    const root = Node.Index.root.get(mf);
+    assert(root.parent == .none);
+    assert(root.prev == .none);
+    assert(root.next == .none);
+    mf.verifyNode(Node.Index.root);
+}
+
+fn verifyNode(mf: *MappedFile, parent_ni: Node.Index) void {
+    const parent = parent_ni.get(mf);
+    const parent_offset, const parent_size = parent.location().resolve(mf);
+    var prev_ni: Node.Index = .none;
+    var prev_end: u64 = 0;
+    var ni = parent.first;
+    while (true) {
+        if (ni == .none) {
+            assert(parent.last == prev_ni);
+            return;
+        }
+        const node = ni.get(mf);
+        assert(node.parent == parent_ni);
+        const offset, const size = node.location().resolve(mf);
+        assert(node.flags.alignment.check(@intCast(offset)));
+        assert(node.flags.alignment.check(@intCast(size)));
+        const end = offset + size;
+        assert(end <= parent_offset + parent_size);
+        assert(offset >= prev_end);
+        assert(node.prev == prev_ni);
+        mf.verifyNode(ni);
+        prev_ni = ni;
+        prev_end = end;
+        ni = node.next;
+    }
+}
+
+const assert = std.debug.assert;
+const builtin = @import("builtin");
+const is_linux = builtin.os.tag == .linux;
+const is_windows = builtin.os.tag == .windows;
+const linux = std.os.linux;
+const MappedFile = @This();
+const std = @import("std");
+const windows = std.os.windows;

src/link/Queue.zig

@@ -22,17 +22,17 @@ prelink_wait_count: u32,
 
 /// Prelink tasks which have been enqueued and are not yet owned by the worker thread.
 /// Allocated into `gpa`, guarded by `mutex`.
-queued_prelink: std.ArrayListUnmanaged(PrelinkTask),
+queued_prelink: std.ArrayList(PrelinkTask),
 /// The worker thread moves items from `queued_prelink` into this array in order to process them.
 /// Allocated into `gpa`, accessed only by the worker thread.
-wip_prelink: std.ArrayListUnmanaged(PrelinkTask),
+wip_prelink: std.ArrayList(PrelinkTask),
 
 /// Like `queued_prelink`, but for ZCU tasks.
 /// Allocated into `gpa`, guarded by `mutex`.
-queued_zcu: std.ArrayListUnmanaged(ZcuTask),
+queued_zcu: std.ArrayList(ZcuTask),
 /// Like `wip_prelink`, but for ZCU tasks.
 /// Allocated into `gpa`, accessed only by the worker thread.
-wip_zcu: std.ArrayListUnmanaged(ZcuTask),
+wip_zcu: std.ArrayList(ZcuTask),
 
 /// When processing ZCU link tasks, we might have to block due to unpopulated MIR. When this
 /// happens, some tasks in `wip_zcu` have been run, and some are still pending. This is the
@@ -213,32 +213,41 @@ pub fn enqueueZcu(q: *Queue, comp: *Compilation, task: ZcuTask) Allocator.Error!
 
 fn flushTaskQueue(tid: usize, q: *Queue, comp: *Compilation) void {
     q.flush_safety.lock(); // every `return` site should unlock this before unlocking `q.mutex`
-
     if (std.debug.runtime_safety) {
         q.mutex.lock();
         defer q.mutex.unlock();
         assert(q.state == .running);
     }
+
+    var have_idle_tasks = true;
     prelink: while (true) {
         assert(q.wip_prelink.items.len == 0);
+        swap_queues: while (true) {
             {
                 q.mutex.lock();
                 defer q.mutex.unlock();
-            std.mem.swap(std.ArrayListUnmanaged(PrelinkTask), &q.queued_prelink, &q.wip_prelink);
+                std.mem.swap(std.ArrayList(PrelinkTask), &q.queued_prelink, &q.wip_prelink);
-            if (q.wip_prelink.items.len == 0) {
+                if (q.wip_prelink.items.len > 0) break :swap_queues;
-                if (q.prelink_wait_count == 0) {
+                if (q.prelink_wait_count == 0) break :prelink; // prelink is done
-                    break :prelink; // prelink is done
+                if (!have_idle_tasks) {
-                } else {
                     // We're expecting more prelink tasks so can't move on to ZCU tasks.
                     q.state = .finished;
                     q.flush_safety.unlock();
                     return;
                 }
             }
+            have_idle_tasks = link.doIdleTask(comp, tid) catch |err| switch (err) {
+                error.OutOfMemory => have_idle_tasks: {
+                    comp.link_diags.setAllocFailure();
+                    break :have_idle_tasks false;
+                },
+                error.LinkFailure => false,
+            };
         }
         for (q.wip_prelink.items) |task| {
             link.doPrelinkTask(comp, task);
         }
+        have_idle_tasks = true;
         q.wip_prelink.clearRetainingCapacity();
     }
 
@@ -256,26 +265,48 @@ fn flushTaskQueue(tid: usize, q: *Queue, comp: *Compilation) void {
 
     // Now we can run ZCU tasks.
     while (true) {
-        if (q.wip_zcu.items.len == q.wip_zcu_idx) {
+        if (q.wip_zcu.items.len == q.wip_zcu_idx) swap_queues: {
             q.wip_zcu.clearRetainingCapacity();
             q.wip_zcu_idx = 0;
+            while (true) {
+                {
                     q.mutex.lock();
                     defer q.mutex.unlock();
-            std.mem.swap(std.ArrayListUnmanaged(ZcuTask), &q.queued_zcu, &q.wip_zcu);
+                    std.mem.swap(std.ArrayList(ZcuTask), &q.queued_zcu, &q.wip_zcu);
-            if (q.wip_zcu.items.len == 0) {
+                    if (q.wip_zcu.items.len > 0) break :swap_queues;
+                    if (!have_idle_tasks) {
                         // We've exhausted all available tasks.
                         q.state = .finished;
                         q.flush_safety.unlock();
                         return;
                     }
                 }
+                have_idle_tasks = link.doIdleTask(comp, tid) catch |err| switch (err) {
+                    error.OutOfMemory => have_idle_tasks: {
+                        comp.link_diags.setAllocFailure();
+                        break :have_idle_tasks false;
+                    },
+                    error.LinkFailure => false,
+                };
+            }
+        }
         const task = q.wip_zcu.items[q.wip_zcu_idx];
         // If the task is a `link_func`, we might have to stop until its MIR is populated.
         pending: {
             if (task != .link_func) break :pending;
             const status_ptr = &task.link_func.mir.status;
+            while (true) {
                 // First check without the mutex to optimize for the common case where MIR is ready.
                 if (status_ptr.load(.acquire) != .pending) break :pending;
+                if (have_idle_tasks) have_idle_tasks = link.doIdleTask(comp, tid) catch |err| switch (err) {
+                    error.OutOfMemory => have_idle_tasks: {
+                        comp.link_diags.setAllocFailure();
+                        break :have_idle_tasks false;
+                    },
+                    error.LinkFailure => false,
+                };
+                if (!have_idle_tasks) break;
+            }
             q.mutex.lock();
             defer q.mutex.unlock();
             if (status_ptr.load(.acquire) != .pending) break :pending;
@@ -298,6 +329,7 @@ fn flushTaskQueue(tid: usize, q: *Queue, comp: *Compilation) void {
             }
         }
         q.wip_zcu_idx += 1;
+        have_idle_tasks = true;
     }
 }
 

src/link/Wasm.zig

@@ -4257,7 +4257,14 @@ fn lowerZcuData(wasm: *Wasm, pt: Zcu.PerThread, ip_index: InternPool.Index) !Zcu
     const func_table_fixups_start: u32 = @intCast(wasm.func_table_fixups.items.len);
     wasm.string_bytes_lock.lock();
 
-    try codegen.generateSymbol(&wasm.base, pt, .unneeded, .fromInterned(ip_index), &wasm.string_bytes, .none);
+    {
+        var aw: std.Io.Writer.Allocating = .fromArrayList(wasm.base.comp.gpa, &wasm.string_bytes);
+        defer wasm.string_bytes = aw.toArrayList();
+        codegen.generateSymbol(&wasm.base, pt, .unneeded, .fromInterned(ip_index), &aw.writer, .none) catch |err| switch (err) {
+            error.WriteFailed => return error.OutOfMemory,
+            else => |e| return e,
+        };
+    }
 
     const code_len: u32 = @intCast(wasm.string_bytes.items.len - code_start);
     const relocs_len: u32 = @intCast(wasm.out_relocs.len - relocs_start);

src/main.zig

@@ -904,7 +904,6 @@ fn buildOutputType(
     var mingw_unicode_entry_point: bool = false;
     var enable_link_snapshots: bool = false;
     var debug_compiler_runtime_libs = false;
-    var opt_incremental: ?bool = null;
     var install_name: ?[]const u8 = null;
     var hash_style: link.File.Lld.Elf.HashStyle = .both;
     var entitlements: ?[]const u8 = null;
@@ -1374,9 +1373,9 @@ fn buildOutputType(
                         }
                     } else if (mem.eql(u8, arg, "-fincremental")) {
                         dev.check(.incremental);
-                        opt_incremental = true;
+                        create_module.opts.incremental = true;
                     } else if (mem.eql(u8, arg, "-fno-incremental")) {
-                        opt_incremental = false;
+                        create_module.opts.incremental = false;
                     } else if (mem.eql(u8, arg, "--entitlements")) {
                         entitlements = args_iter.nextOrFatal();
                     } else if (mem.eql(u8, arg, "-fcompiler-rt")) {
@@ -1479,6 +1478,10 @@ fn buildOutputType(
                         create_module.opts.use_lld = true;
                     } else if (mem.eql(u8, arg, "-fno-lld")) {
                         create_module.opts.use_lld = false;
+                    } else if (mem.eql(u8, arg, "-fnew-linker")) {
+                        create_module.opts.use_new_linker = true;
+                    } else if (mem.eql(u8, arg, "-fno-new-linker")) {
+                        create_module.opts.use_new_linker = false;
                     } else if (mem.eql(u8, arg, "-fclang")) {
                         create_module.opts.use_clang = true;
                     } else if (mem.eql(u8, arg, "-fno-clang")) {
@@ -3371,7 +3374,7 @@ fn buildOutputType(
         else => false,
     };
 
-    const incremental = opt_incremental orelse false;
+    const incremental = create_module.resolved_options.incremental;
     if (debug_incremental and !incremental) {
         fatal("--debug-incremental requires -fincremental", .{});
     }
@@ -3502,7 +3505,6 @@ fn buildOutputType(
         .subsystem = subsystem,
         .debug_compile_errors = debug_compile_errors,
         .debug_incremental = debug_incremental,
-        .incremental = incremental,
         .enable_link_snapshots = enable_link_snapshots,
         .install_name = install_name,
         .entitlements = entitlements,
@@ -4016,6 +4018,8 @@ fn createModule(
             error.LldUnavailable => fatal("zig was compiled without LLD libraries", .{}),
             error.ClangUnavailable => fatal("zig was compiled without Clang libraries", .{}),
             error.DllExportFnsRequiresWindows => fatal("only Windows OS targets support DLLs", .{}),
+            error.NewLinkerIncompatibleObjectFormat => fatal("using the new linker to link {s} files is unsupported", .{@tagName(target.ofmt)}),
+            error.NewLinkerIncompatibleWithLld => fatal("using the new linker is incompatible with using lld", .{}),
         };
     }
 

src/target.zig

@@ -231,6 +231,13 @@ pub fn hasLldSupport(ofmt: std.Target.ObjectFormat) bool {
     };
 }
 
+pub fn hasNewLinkerSupport(ofmt: std.Target.ObjectFormat) bool {
+    return switch (ofmt) {
+        .elf => true,
+        else => false,
+    };
+}
+
 /// The set of targets that our own self-hosted backends have robust support for.
 /// Used to select between LLVM backend and self-hosted backend when compiling in
 /// debug mode. A given target should only return true here if it is passing greater

test/incremental/change_exports

@@ -1,4 +1,4 @@
-//#target=x86_64-linux-selfhosted
+#target=x86_64-linux-selfhosted
 #target=x86_64-linux-cbe
 #target=x86_64-windows-cbe
 

test/incremental/change_panic_handler

@@ -1,3 +1,4 @@
+#target=x86_64-linux-selfhosted
 #target=x86_64-linux-cbe
 #target=x86_64-windows-cbe
 #update=initial version

test/incremental/change_panic_handler_explicit

@@ -1,3 +1,4 @@
+#target=x86_64-linux-selfhosted
 #target=x86_64-linux-cbe
 #target=x86_64-windows-cbe
 #update=initial version

test/incremental/change_struct_same_fields

@@ -1,4 +1,4 @@
-//#target=x86_64-linux-selfhosted
+#target=x86_64-linux-selfhosted
 #target=x86_64-linux-cbe
 #target=x86_64-windows-cbe
 #target=wasm32-wasi-selfhosted

test/incremental/type_becomes_comptime_only

@@ -1,3 +1,4 @@
+#target=x86_64-linux-selfhosted
 #target=x86_64-linux-cbe
 #target=x86_64-windows-cbe
 #target=wasm32-wasi-selfhosted