parse.nix

# Define the list of system with their properties.
#
# See https://clang.llvm.org/docs/CrossCompilation.html and
# http://llvm.org/docs/doxygen/html/Triple_8cpp_source.html especially
# Triple::normalize. Parsing should essentially act as a more conservative
# version of that last function.
#
# Most of the types below come in "open" and "closed" pairs. The open ones
# specify what information we need to know about systems in general, and the
# closed ones are sub-types representing the whitelist of systems we support in
# practice.
#
# Code in the remainder of nixpkgs shouldn't rely on the closed ones in
# e.g. exhaustive cases. Its more a sanity check to make sure nobody defines
# systems that overlap with existing ones and won't notice something amiss.
#
{ lib }:
with lib.lists;
with lib.types;
with lib.attrsets;
with lib.strings;
with (import ./inspect.nix { inherit lib; }).predicates;

let
  inherit (lib.options) mergeOneOption;

  setTypes = type:
    mapAttrs (name: value:
      assert type.check value;
      setType type.name ({ inherit name; } // value));

  # gnu-config will ignore the portion of a triple matching the
  # regex `e?abi.*$` when determining the validity of a triple.  In
  # other words, `i386-linuxabichickenlips` is a valid triple.
  removeAbiSuffix = x:
    let match = builtins.match "(.*)e?abi.*" x;
    in if match==null
       then x
       else lib.elemAt match 0;

in

rec {

  ################################################################################

  types.openSignificantByte = mkOptionType {
    name = "significant-byte";
    description = "Endianness";
    merge = mergeOneOption;
  };

  types.significantByte = enum (attrValues significantBytes);

  significantBytes = setTypes types.openSignificantByte {
    bigEndian = {};
    littleEndian = {};
  };

  ################################################################################

  # Reasonable power of 2
  types.bitWidth = enum [ 8 16 32 64 128 ];

  ################################################################################

  types.openCpuType = mkOptionType {
    name = "cpu-type";
    description = "instruction set architecture name and information";
    merge = mergeOneOption;
    check = x: types.bitWidth.check x.bits
      && (if 8 < x.bits
          then types.significantByte.check x.significantByte
          else !(x ? significantByte));
  };

  types.cpuType = enum (attrValues cpuTypes);

  cpuTypes = with significantBytes; setTypes types.openCpuType {
    arm      = { bits = 32; significantByte = littleEndian; family = "arm"; };
    armv5tel = { bits = 32; significantByte = littleEndian; family = "arm"; version = "5"; arch = "armv5t"; };
    armv6m   = { bits = 32; significantByte = littleEndian; family = "arm"; version = "6"; arch = "armv6-m"; };
    armv6l   = { bits = 32; significantByte = littleEndian; family = "arm"; version = "6"; arch = "armv6"; };
    armv7a   = { bits = 32; significantByte = littleEndian; family = "arm"; version = "7"; arch = "armv7-a"; };
    armv7r   = { bits = 32; significantByte = littleEndian; family = "arm"; version = "7"; arch = "armv7-r"; };
    armv7m   = { bits = 32; significantByte = littleEndian; family = "arm"; version = "7"; arch = "armv7-m"; };
    armv7l   = { bits = 32; significantByte = littleEndian; family = "arm"; version = "7"; arch = "armv7"; };
    armv8a   = { bits = 32; significantByte = littleEndian; family = "arm"; version = "8"; arch = "armv8-a"; };
    armv8r   = { bits = 32; significantByte = littleEndian; family = "arm"; version = "8"; arch = "armv8-a"; };
    armv8m   = { bits = 32; significantByte = littleEndian; family = "arm"; version = "8"; arch = "armv8-m"; };
    aarch64  = { bits = 64; significantByte = littleEndian; family = "arm"; version = "8"; arch = "armv8-a"; };
    aarch64_be = { bits = 64; significantByte = bigEndian; family = "arm"; version = "8";  arch = "armv8-a"; };

    i386     = { bits = 32; significantByte = littleEndian; family = "x86"; arch = "i386"; };
    i486     = { bits = 32; significantByte = littleEndian; family = "x86"; arch = "i486"; };
    i586     = { bits = 32; significantByte = littleEndian; family = "x86"; arch = "i586"; };
    i686     = { bits = 32; significantByte = littleEndian; family = "x86"; arch = "i686"; };
    x86_64   = { bits = 64; significantByte = littleEndian; family = "x86"; arch = "x86-64"; };

    microblaze   = { bits = 32; significantByte = bigEndian;    family = "microblaze"; };
    microblazeel = { bits = 32; significantByte = littleEndian; family = "microblaze"; };

    mips     = { bits = 32; significantByte = bigEndian;    family = "mips"; };
    mipsel   = { bits = 32; significantByte = littleEndian; family = "mips"; };
    mips64   = { bits = 64; significantByte = bigEndian;    family = "mips"; };
    mips64el = { bits = 64; significantByte = littleEndian; family = "mips"; };

    mmix     = { bits = 64; significantByte = bigEndian;    family = "mmix"; };

    m68k     = { bits = 32; significantByte = bigEndian; family = "m68k"; };

    powerpc  = { bits = 32; significantByte = bigEndian;    family = "power"; };
    powerpc64 = { bits = 64; significantByte = bigEndian; family = "power"; };
    powerpc64le = { bits = 64; significantByte = littleEndian; family = "power"; };
    powerpcle = { bits = 32; significantByte = littleEndian; family = "power"; };

    riscv32  = { bits = 32; significantByte = littleEndian; family = "riscv"; };
    riscv64  = { bits = 64; significantByte = littleEndian; family = "riscv"; };

    s390     = { bits = 32; significantByte = bigEndian; family = "s390"; };
    s390x    = { bits = 64; significantByte = bigEndian; family = "s390"; };

    sparc    = { bits = 32; significantByte = bigEndian;    family = "sparc"; };
    sparc64  = { bits = 64; significantByte = bigEndian;    family = "sparc"; };

    wasm32   = { bits = 32; significantByte = littleEndian; family = "wasm"; };
    wasm64   = { bits = 64; significantByte = littleEndian; family = "wasm"; };

    alpha    = { bits = 64; significantByte = littleEndian; family = "alpha"; };

    rx       = { bits = 32; significantByte = littleEndian; family = "rx"; };
    msp430   = { bits = 16; significantByte = littleEndian; family = "msp430"; };
    avr      = { bits = 8; family = "avr"; };

    vc4      = { bits = 32; significantByte = littleEndian; family = "vc4"; };

    or1k     = { bits = 32; significantByte = bigEndian; family = "or1k"; };

    loongarch64 = { bits = 64; significantByte = littleEndian; family = "loongarch"; };

    javascript = { bits = 32; significantByte = littleEndian; family = "javascript"; };
  };

  # GNU build systems assume that older NetBSD architectures are using a.out.
  gnuNetBSDDefaultExecFormat = cpu:
    if (cpu.family == "arm" && cpu.bits == 32) ||
       (cpu.family == "sparc" && cpu.bits == 32) ||
       (cpu.family == "m68k" && cpu.bits == 32) ||
       (cpu.family == "x86" && cpu.bits == 32)
    then execFormats.aout
    else execFormats.elf;

  # Determine when two CPUs are compatible with each other. That is,
  # can code built for system B run on system A? For that to happen,
  # the programs that system B accepts must be a subset of the
  # programs that system A accepts.
  #
  # We have the following properties of the compatibility relation,
  # which must be preserved when adding compatibility information for
  # additional CPUs.
  # - (reflexivity)
  #   Every CPU is compatible with itself.
  # - (transitivity)
  #   If A is compatible with B and B is compatible with C then A is compatible with C.
  #
  # Note: Since 22.11 the archs of a mode switching CPU are no longer considered
  # pairwise compatible. Mode switching implies that binaries built for A
  # and B respectively can't be executed at the same time.
  isCompatible = a: b: with cpuTypes; lib.any lib.id [
    # x86
    (b == i386 && isCompatible a i486)
    (b == i486 && isCompatible a i586)
    (b == i586 && isCompatible a i686)

    # XXX: Not true in some cases. Like in WSL mode.
    (b == i686 && isCompatible a x86_64)

    # ARMv4
    (b == arm && isCompatible a armv5tel)

    # ARMv5
    (b == armv5tel && isCompatible a armv6l)

    # ARMv6
    (b == armv6l && isCompatible a armv6m)
    (b == armv6m && isCompatible a armv7l)

    # ARMv7
    (b == armv7l && isCompatible a armv7a)
    (b == armv7l && isCompatible a armv7r)
    (b == armv7l && isCompatible a armv7m)

    # ARMv8
    (b == aarch64 && a == armv8a)
    (b == armv8a && isCompatible a aarch64)
    (b == armv8r && isCompatible a armv8a)
    (b == armv8m && isCompatible a armv8a)

    # PowerPC
    (b == powerpc && isCompatible a powerpc64)
    (b == powerpcle && isCompatible a powerpc64le)

    # MIPS
    (b == mips && isCompatible a mips64)
    (b == mipsel && isCompatible a mips64el)

    # RISCV
    (b == riscv32 && isCompatible a riscv64)

    # SPARC
    (b == sparc && isCompatible a sparc64)

    # WASM
    (b == wasm32 && isCompatible a wasm64)

    # identity
    (b == a)
  ];

  ################################################################################

  types.openVendor = mkOptionType {
    name = "vendor";
    description = "vendor for the platform";
    merge = mergeOneOption;
  };

  types.vendor = enum (attrValues vendors);

  vendors = setTypes types.openVendor {
    apple = {};
    pc = {};
    knuth = {};

    # Actually matters, unlocking some MinGW-w64-specific options in GCC. See
    # bottom of https://sourceforge.net/p/mingw-w64/wiki2/Unicode%20apps/
    w64 = {};

    none = {};
    unknown = {};
  };

  ################################################################################

  types.openExecFormat = mkOptionType {
    name = "exec-format";
    description = "executable container used by the kernel";
    merge = mergeOneOption;
  };

  types.execFormat = enum (attrValues execFormats);

  execFormats = setTypes types.openExecFormat {
    aout = {}; # a.out
    elf = {};
    macho = {};
    pe = {};
    wasm = {};

    unknown = {};
  };

  ################################################################################

  types.openKernelFamily = mkOptionType {
    name = "exec-format";
    description = "executable container used by the kernel";
    merge = mergeOneOption;
  };

  types.kernelFamily = enum (attrValues kernelFamilies);

  kernelFamilies = setTypes types.openKernelFamily {
    bsd = {};
    darwin = {};
  };

  ################################################################################

  types.openKernel = mkOptionType {
    name = "kernel";
    description = "kernel name and information";
    merge = mergeOneOption;
    check = x: types.execFormat.check x.execFormat
        && all types.kernelFamily.check (attrValues x.families);
  };

  types.kernel = enum (attrValues kernels);

  kernels = with execFormats; with kernelFamilies; setTypes types.openKernel {
    # TODO(@Ericson2314): Don't want to mass-rebuild yet to keeping 'darwin' as
    # the normalized name for macOS.
    macos    = { execFormat = macho;   families = { inherit darwin; }; name = "darwin"; };
    ios      = { execFormat = macho;   families = { inherit darwin; }; };
    # A tricky thing about FreeBSD is that there is no stable ABI across
    # versions. That means that putting in the version as part of the
    # config string is paramount.
    freebsd12 = { execFormat = elf;     families = { inherit bsd; }; name = "freebsd"; version = 12; };
    freebsd13 = { execFormat = elf;     families = { inherit bsd; }; name = "freebsd"; version = 13; };
    linux    = { execFormat = elf;     families = { }; };
    netbsd   = { execFormat = elf;     families = { inherit bsd; }; };
    none     = { execFormat = unknown; families = { }; };
    openbsd  = { execFormat = elf;     families = { inherit bsd; }; };
    solaris  = { execFormat = elf;     families = { }; };
    wasi     = { execFormat = wasm;    families = { }; };
    redox    = { execFormat = elf;     families = { }; };
    windows  = { execFormat = pe;      families = { }; };
    ghcjs    = { execFormat = unknown; families = { }; };
    genode   = { execFormat = elf;     families = { }; };
    mmixware = { execFormat = unknown; families = { }; };
  } // { # aliases
    # 'darwin' is the kernel for all of them. We choose macOS by default.
    darwin = kernels.macos;
    watchos = kernels.ios;
    tvos = kernels.ios;
    win32 = kernels.windows;
  };

  ################################################################################

  types.openAbi = mkOptionType {
    name = "abi";
    description = "binary interface for compiled code and syscalls";
    merge = mergeOneOption;
  };

  types.abi = enum (attrValues abis);

  abis = setTypes types.openAbi {
    cygnus       = {};
    msvc         = {};

    # Note: eabi is specific to ARM and PowerPC.
    # On PowerPC, this corresponds to PPCEABI.
    # On ARM, this corresponds to ARMEABI.
    eabi         = { float = "soft"; };
    eabihf       = { float = "hard"; };

    # Other architectures should use ELF in embedded situations.
    elf          = {};

    androideabi  = {};
    android      = {
      assertions = [
        { assertion = platform: !platform.isAarch32;
          message = ''
            The "android" ABI is not for 32-bit ARM. Use "androideabi" instead.
          '';
        }
      ];
    };

    gnueabi      = { float = "soft"; };
    gnueabihf    = { float = "hard"; };
    gnu          = {
      assertions = [
        { assertion = platform: !platform.isAarch32;
          message = ''
            The "gnu" ABI is ambiguous on 32-bit ARM. Use "gnueabi" or "gnueabihf" instead.
          '';
        }
        { assertion = platform: with platform; !(isPower64 && isBigEndian);
          message = ''
            The "gnu" ABI is ambiguous on big-endian 64-bit PowerPC. Use "gnuabielfv2" or "gnuabielfv1" instead.
          '';
        }
      ];
    };
    gnuabi64     = { abi = "64"; };
    muslabi64    = { abi = "64"; };

    # NOTE: abi=n32 requires a 64-bit MIPS chip!  That is not a typo.
    # It is basically the 64-bit abi with 32-bit pointers.  Details:
    # https://www.linux-mips.org/pub/linux/mips/doc/ABI/MIPS-N32-ABI-Handbook.pdf
    gnuabin32    = { abi = "n32"; };
    muslabin32   = { abi = "n32"; };

    gnuabielfv2  = { abi = "elfv2"; };
    gnuabielfv1  = { abi = "elfv1"; };

    musleabi     = { float = "soft"; };
    musleabihf   = { float = "hard"; };
    musl         = {};

    uclibceabi   = { float = "soft"; };
    uclibceabihf = { float = "hard"; };
    uclibc       = {};

    unknown = {};
  };

  ################################################################################

  types.parsedPlatform = mkOptionType {
    name = "system";
    description = "fully parsed representation of llvm- or nix-style platform tuple";
    merge = mergeOneOption;
    check = { cpu, vendor, kernel, abi }:
           types.cpuType.check cpu
        && types.vendor.check vendor
        && types.kernel.check kernel
        && types.abi.check abi;
  };

  isSystem = isType "system";

  mkSystem = components:
    assert types.parsedPlatform.check components;
    setType "system" components;

  mkSkeletonFromList = l: {
    "1" = if elemAt l 0 == "avr"
      then { cpu = elemAt l 0; kernel = "none"; abi = "unknown"; }
      else throw "Target specification with 1 components is ambiguous";
    "2" = # We only do 2-part hacks for things Nix already supports
      if elemAt l 1 == "cygwin"
        then { cpu = elemAt l 0;                      kernel = "windows";  abi = "cygnus";   }
      # MSVC ought to be the default ABI so this case isn't needed. But then it
      # becomes difficult to handle the gnu* variants for Aarch32 correctly for
      # minGW. So it's easier to make gnu* the default for the MinGW, but
      # hack-in MSVC for the non-MinGW case right here.
      else if elemAt l 1 == "windows"
        then { cpu = elemAt l 0;                      kernel = "windows";  abi = "msvc";     }
      else if (elemAt l 1) == "elf"
        then { cpu = elemAt l 0; vendor = "unknown";  kernel = "none";     abi = elemAt l 1; }
      else   { cpu = elemAt l 0;                      kernel = elemAt l 1;                   };
    "3" =
      # cpu-kernel-environment
      if elemAt l 1 == "linux" ||
         elem (elemAt l 2) ["eabi" "eabihf" "elf" "gnu"]
      then {
        cpu    = elemAt l 0;
        kernel = elemAt l 1;
        abi    = elemAt l 2;
        vendor = "unknown";
      }
      # cpu-vendor-os
      else if elemAt l 1 == "apple" ||
              elem (elemAt l 2) [ "wasi" "redox" "mmixware" "ghcjs" "mingw32" ] ||
              hasPrefix "freebsd" (elemAt l 2) ||
              hasPrefix "netbsd" (elemAt l 2) ||
              hasPrefix "genode" (elemAt l 2)
      then {
        cpu    = elemAt l 0;
        vendor = elemAt l 1;
        kernel = if elemAt l 2 == "mingw32"
                 then "windows"  # autotools breaks on -gnu for window
                 else elemAt l 2;
      }
      else throw "Target specification with 3 components is ambiguous";
    "4" =    { cpu = elemAt l 0; vendor = elemAt l 1; kernel = elemAt l 2; abi = elemAt l 3; };
  }.${toString (length l)}
    or (throw "system string has invalid number of hyphen-separated components");

  # This should revert the job done by config.guess from the gcc compiler.
  mkSystemFromSkeleton = { cpu
                         , # Optional, but fallback too complex for here.
                           # Inferred below instead.
                           vendor ? assert false; null
                         , kernel
                         , # Also inferred below
                           abi    ? assert false; null
                         } @ args: let
    getCpu    = name: cpuTypes.${name} or (throw "Unknown CPU type: ${name}");
    getVendor = name:  vendors.${name} or (throw "Unknown vendor: ${name}");
    getKernel = name:  kernels.${name} or (throw "Unknown kernel: ${name}");
    getAbi    = name:     abis.${name} or (throw "Unknown ABI: ${name}");

    parsed = {
      cpu = getCpu args.cpu;
      vendor =
        /**/ if args ? vendor    then getVendor args.vendor
        else if isDarwin  parsed then vendors.apple
        else if isWindows parsed then vendors.pc
        else                     vendors.unknown;
      kernel = if hasPrefix "darwin" args.kernel      then getKernel "darwin"
               else if hasPrefix "netbsd" args.kernel then getKernel "netbsd"
               else                                   getKernel (removeAbiSuffix args.kernel);
      abi =
        /**/ if args ? abi       then getAbi args.abi
        else if isLinux parsed || isWindows parsed then
          if isAarch32 parsed then
            if lib.versionAtLeast (parsed.cpu.version or "0") "6"
            then abis.gnueabihf
            else abis.gnueabi
          # Default ppc64 BE to ELFv2
          else if isPower64 parsed && isBigEndian parsed then abis.gnuabielfv2
          else abis.gnu
        else                     abis.unknown;
    };

  in mkSystem parsed;

  mkSystemFromString = s: mkSystemFromSkeleton (mkSkeletonFromList (lib.splitString "-" s));

  kernelName = kernel:
    kernel.name + toString (kernel.version or "");

  doubleFromSystem = { cpu, kernel, abi, ... }:
    /**/ if abi == abis.cygnus       then "${cpu.name}-cygwin"
    else if kernel.families ? darwin then "${cpu.name}-darwin"
    else "${cpu.name}-${kernelName kernel}";

  tripleFromSystem = { cpu, vendor, kernel, abi, ... } @ sys: assert isSystem sys; let
    optExecFormat =
      lib.optionalString (kernel.name == "netbsd" &&
                          gnuNetBSDDefaultExecFormat cpu != kernel.execFormat)
        kernel.execFormat.name;
    optAbi = lib.optionalString (abi != abis.unknown) "-${abi.name}";
  in "${cpu.name}-${vendor.name}-${kernelName kernel}${optExecFormat}${optAbi}";

  ################################################################################

}