ability to use zig cc as a drop-in C compiler

The basics are working

src-self-hosted/clang_options.zig

@@ -0,0 +1,126 @@
+const std = @import("std");
+const mem = std.mem;
+
+pub const list = @import("clang_options_data.zig").data;
+
+pub const CliArg = struct {
+    name: []const u8,
+    syntax: Syntax,
+
+    /// TODO we're going to want to change this when we start shipping self-hosted because this causes
+    /// all the functions in stage2.zig to get exported.
+    zig_equivalent: @import("stage2.zig").ClangArgIterator.ZigEquivalent,
+
+    /// Prefixed by "-"
+    pd1: bool = false,
+
+    /// Prefixed by "--"
+    pd2: bool = false,
+
+    /// Prefixed by "/"
+    psl: bool = false,
+
+    const Syntax = union(enum) {
+        /// A flag with no values.
+        flag,
+
+        /// An option which prefixes its (single) value.
+        joined,
+
+        /// An option which is followed by its value.
+        separate,
+
+        /// An option which is either joined to its (non-empty) value, or followed by its value.
+        joined_or_separate,
+
+        /// An option which is both joined to its (first) value, and followed by its (second) value.
+        joined_and_separate,
+
+        /// An option followed by its values, which are separated by commas.
+        comma_joined,
+
+        /// An option which consumes an optional joined argument and any other remaining arguments.
+        remaining_args_joined,
+
+        /// An option which is which takes multiple (separate) arguments.
+        multi_arg: u8,
+    };
+
+    fn matchEql(self: CliArg, arg: []const u8) bool {
+        if (self.pd1 and arg.len >= self.name.len + 1 and
+            mem.startsWith(u8, arg, "-") and mem.eql(u8, arg[1..], self.name))
+        {
+            return true;
+        }
+        if (self.pd2 and arg.len >= self.name.len + 2 and
+            mem.startsWith(u8, arg, "--") and mem.eql(u8, arg[2..], self.name))
+        {
+            return true;
+        }
+        if (self.psl and arg.len >= self.name.len + 1 and
+            mem.startsWith(u8, arg, "/") and mem.eql(u8, arg[1..], self.name))
+        {
+            return true;
+        }
+        return false;
+    }
+
+    fn matchStartsWith(self: CliArg, arg: []const u8) usize {
+        if (self.pd1 and arg.len >= self.name.len + 1 and
+            mem.startsWith(u8, arg, "-") and mem.startsWith(u8, arg[1..], self.name))
+        {
+            return self.name.len + 1;
+        }
+        if (self.pd2 and arg.len >= self.name.len + 2 and
+            mem.startsWith(u8, arg, "--") and mem.startsWith(u8, arg[2..], self.name))
+        {
+            return self.name.len + 2;
+        }
+        if (self.psl and arg.len >= self.name.len + 1 and
+            mem.startsWith(u8, arg, "/") and mem.startsWith(u8, arg[1..], self.name))
+        {
+            return self.name.len + 1;
+        }
+        return 0;
+    }
+};
+
+/// Shortcut function for initializing a `CliArg`
+pub fn flagpd1(name: []const u8) CliArg {
+    return .{
+        .name = name,
+        .syntax = .flag,
+        .zig_equivalent = .other,
+        .pd1 = true,
+    };
+}
+
+/// Shortcut function for initializing a `CliArg`
+pub fn joinpd1(name: []const u8) CliArg {
+    return .{
+        .name = name,
+        .syntax = .joined,
+        .zig_equivalent = .other,
+        .pd1 = true,
+    };
+}
+
+/// Shortcut function for initializing a `CliArg`
+pub fn jspd1(name: []const u8) CliArg {
+    return .{
+        .name = name,
+        .syntax = .joined_or_separate,
+        .zig_equivalent = .other,
+        .pd1 = true,
+    };
+}
+
+/// Shortcut function for initializing a `CliArg`
+pub fn sepd1(name: []const u8) CliArg {
+    return .{
+        .name = name,
+        .syntax = .separate,
+        .zig_equivalent = .other,
+        .pd1 = true,
+    };
+}

src-self-hosted/stage2.zig

@@ -113,6 +113,7 @@ const Error = extern enum {
     TargetHasNoDynamicLinker,
     InvalidAbiVersion,
     InvalidOperatingSystemVersion,
+    UnknownClangOption,
 };
 
 const FILE = std.c.FILE;
@@ -1215,3 +1216,157 @@ fn convertSlice(slice: [][:0]u8, ptr: *[*][*:0]u8, len: *usize) !void {
     }
     ptr.* = new_slice.ptr;
 }
+
+const clang_args = @import("clang_options.zig").list;
+
+// ABI warning
+pub const ClangArgIterator = extern struct {
+    has_next: bool,
+    zig_equivalent: ZigEquivalent,
+    only_arg: [*:0]const u8,
+    second_arg: [*:0]const u8,
+    other_args_ptr: [*]const [*:0]const u8,
+    other_args_len: usize,
+    argv_ptr: [*]const [*:0]const u8,
+    argv_len: usize,
+    next_index: usize,
+
+    // ABI warning
+    pub const ZigEquivalent = extern enum {
+        target,
+        o,
+        c,
+        other,
+        positional,
+        l,
+    };
+
+    fn init(argv: []const [*:0]const u8) ClangArgIterator {
+        return .{
+            .next_index = 2, // `zig cc foo` this points to `foo`
+            .has_next = argv.len > 2,
+            .zig_equivalent = undefined,
+            .only_arg = undefined,
+            .second_arg = undefined,
+            .other_args_ptr = undefined,
+            .other_args_len = undefined,
+            .argv_ptr = argv.ptr,
+            .argv_len = argv.len,
+        };
+    }
+
+    fn next(self: *ClangArgIterator) !void {
+        assert(self.has_next);
+        assert(self.next_index < self.argv_len);
+        // In this state we know that the parameter we are looking at is a root parameter
+        // rather than an argument to a parameter.
+        self.other_args_ptr = self.argv_ptr + self.next_index;
+        self.other_args_len = 1; // We adjust this value below when necessary.
+        const arg = mem.span(self.argv_ptr[self.next_index]);
+        self.next_index += 1;
+        defer {
+            if (self.next_index >= self.argv_len) self.has_next = false;
+        }
+
+        if (!mem.startsWith(u8, arg, "-")) {
+            self.zig_equivalent = .positional;
+            self.only_arg = arg.ptr;
+            return;
+        }
+
+        find_clang_arg: for (clang_args) |clang_arg| switch (clang_arg.syntax) {
+            .flag => if (clang_arg.matchEql(arg)) {
+                self.zig_equivalent = clang_arg.zig_equivalent;
+                self.only_arg = arg.ptr;
+
+                break :find_clang_arg;
+            },
+            .joined, .comma_joined => {
+                // Example: --target=foo
+                const prefix_len = clang_arg.matchStartsWith(arg);
+                if (prefix_len != 0) {
+                    self.zig_equivalent = clang_arg.zig_equivalent;
+                    self.only_arg = arg.ptr + prefix_len; // This will skip over the "--target=" part.
+
+                    break :find_clang_arg;
+                }
+            },
+            .joined_or_separate => {
+                // Examples: `-lfoo`, `-l foo`
+                const prefix_len = clang_arg.matchStartsWith(arg);
+                if (prefix_len == arg.len) {
+                    if (self.next_index >= self.argv_len) {
+                        std.debug.warn("Expected parameter after '{}'\n", .{arg});
+                        process.exit(1);
+                    }
+                    self.only_arg = self.argv_ptr[self.next_index];
+                    self.next_index += 1;
+                    self.other_args_len += 1;
+                    self.zig_equivalent = clang_arg.zig_equivalent;
+
+                    break :find_clang_arg;
+                } else if (prefix_len != 0) {
+                    self.zig_equivalent = clang_arg.zig_equivalent;
+                    self.only_arg = arg.ptr + prefix_len;
+
+                    break :find_clang_arg;
+                }
+            },
+            .joined_and_separate => {
+                // Example: `-Xopenmp-target=riscv64-linux-unknown foo`
+                const prefix_len = clang_arg.matchStartsWith(arg);
+                if (prefix_len != 0) {
+                    self.only_arg = arg.ptr + prefix_len;
+                    if (self.next_index >= self.argv_len) {
+                        std.debug.warn("Expected parameter after '{}'\n", .{arg});
+                        process.exit(1);
+                    }
+                    self.second_arg = self.argv_ptr[self.next_index];
+                    self.next_index += 1;
+                    self.other_args_len += 1;
+                    self.zig_equivalent = clang_arg.zig_equivalent;
+                    break :find_clang_arg;
+                }
+            },
+            .separate => if (clang_arg.matchEql(arg)) {
+                if (self.next_index >= self.argv_len) {
+                    std.debug.warn("Expected parameter after '{}'\n", .{arg});
+                    process.exit(1);
+                }
+                self.only_arg = self.argv_ptr[self.next_index];
+                self.next_index += 1;
+                self.other_args_len += 1;
+                self.zig_equivalent = clang_arg.zig_equivalent;
+                break :find_clang_arg;
+            },
+            .remaining_args_joined => {
+                const prefix_len = clang_arg.matchStartsWith(arg);
+                if (prefix_len != 0) {
+                    @panic("TODO");
+                }
+            },
+            .multi_arg => if (clang_arg.matchEql(arg)) {
+                @panic("TODO");
+            },
+        }
+        else {
+            std.debug.warn("Unknown Clang option: '{}'\n", .{arg});
+            process.exit(1);
+        }
+    }
+};
+
+export fn stage2_clang_arg_iterator(
+    result: *ClangArgIterator,
+    argc: usize,
+    argv: [*]const [*:0]const u8,
+) void {
+    result.* = ClangArgIterator.init(argv[0..argc]);
+}
+
+export fn stage2_clang_arg_next(it: *ClangArgIterator) Error {
+    it.next() catch |err| switch (err) {
+        error.UnknownClangOption => return .UnknownClangOption,
+    };
+    return .None;
+}

src/codegen.cpp

@@ -9778,7 +9778,7 @@ static void gen_c_object(CodeGen *g, Buf *self_exe_path, CFile *c_file) {
         Termination term;
         ZigList<const char *> args = {};
         args.append(buf_ptr(self_exe_path));
-        args.append("cc");
+        args.append("clang");
 
         Buf *out_dep_path = buf_sprintf("%s.d", buf_ptr(out_obj_path));
         add_cc_args(g, args, buf_ptr(out_dep_path), false);

src/error.cpp

@@ -83,6 +83,7 @@ const char *err_str(Error err) {
         case ErrorTargetHasNoDynamicLinker: return "target has no dynamic linker";
         case ErrorInvalidAbiVersion: return "invalid C ABI version";
         case ErrorInvalidOperatingSystemVersion: return "invalid operating system version";
+        case ErrorUnknownClangOption: return "unknown Clang option";
     }
     return "(invalid error)";
 }

src/link.cpp

@@ -2008,7 +2008,7 @@ static const char *get_def_lib(CodeGen *parent, const char *name, Buf *def_in_fi
 
         ZigList<const char *> args = {};
         args.append(buf_ptr(self_exe_path));
-        args.append("cc");
+        args.append("clang");
         args.append("-x");
         args.append("c");
         args.append(buf_ptr(def_in_file));

src/main.cpp

@@ -36,7 +36,7 @@ static int print_full_usage(const char *arg0, FILE *file, int return_code) {
         "  build-lib [source]           create library from source or object files\n"
         "  build-obj [source]           create object from source or assembly\n"
         "  builtin                      show the source code of @import(\"builtin\")\n"
-        "  cc                           C compiler\n"
+        "  cc                           use Zig as a drop-in C compiler\n"
         "  fmt                          parse files and render in canonical zig format\n"
         "  id                           print the base64-encoded compiler id\n"
         "  init-exe                     initialize a `zig build` application in the cwd\n"
@@ -271,7 +271,7 @@ static int main0(int argc, char **argv) {
         return 0;
     }
 
-    if (argc >= 2 && (strcmp(argv[1], "cc") == 0 ||
+    if (argc >= 2 && (strcmp(argv[1], "clang") == 0 ||
             strcmp(argv[1], "-cc1") == 0 || strcmp(argv[1], "-cc1as") == 0))
     {
         return ZigClang_main(argc, argv);
@@ -575,9 +575,55 @@ static int main0(int argc, char **argv) {
         return (term.how == TerminationIdClean) ? term.code : -1;
     } else if (argc >= 2 && strcmp(argv[1], "fmt") == 0) {
         return stage2_fmt(argc, argv);
+    } else if (argc >= 2 && strcmp(argv[1], "cc") == 0) {
+        const char *o_arg = nullptr;
+        bool c_arg = false;
+        Stage2ClangArgIterator it;
+        stage2_clang_arg_iterator(&it, argc, argv);
+        while (it.has_next) {
+            if ((err = stage2_clang_arg_next(&it))) {
+                fprintf(stderr, "unable to parse command line parameters: %s\n", err_str(err));
+                return EXIT_FAILURE;
             }
-
-    for (int i = 1; i < argc; i += 1) {
+            switch (it.kind) {
+                case Stage2ClangArgTarget: // example: -target riscv64-linux-unknown
+                    target_string = it.only_arg;
+                    break;
+                case Stage2ClangArgO: // -o
+                    o_arg = it.only_arg;
+                    break;
+                case Stage2ClangArgC: // -c
+                    c_arg = true;
+                    break;
+                case Stage2ClangArgOther:
+                    for (size_t i = 0; i < it.other_args_len; i += 1) {
+                        clang_argv.append(it.other_args_ptr[i]);
+                    }
+                    break;
+                case Stage2ClangArgPositional: {
+                    CFile *c_file = heap::c_allocator.create<CFile>();
+                    c_file->source_path = it.only_arg;
+                    c_source_files.append(c_file);
+                    break;
+                }
+                case Stage2ClangArgL: // -l
+                    if (strcmp(it.only_arg, "c") == 0)
+                        have_libc = true;
+                    link_libs.append(it.only_arg);
+                    break;
+            }
+        }
+        if (!c_arg) {
+            cmd = CmdBuild;
+            out_type = OutTypeExe;
+            if (o_arg == nullptr) {
+                zig_panic("TODO set out name to a.out");
+            }
+        } else {
+            cmd = CmdBuild;
+            out_type = OutTypeObj;
+        }
+    } else for (int i = 1; i < argc; i += 1) {
         char *arg = argv[i];
 
         if (arg[0] == '-') {

src/stage2.cpp

@@ -304,3 +304,15 @@ enum Error stage2_detect_native_paths(struct Stage2NativePaths *native_paths) {
 
     return ErrorNone;
 }
+
+void stage2_clang_arg_iterator(struct Stage2ClangArgIterator *it,
+        size_t argc, char **argv)
+{
+    const char *msg = "stage0 called stage2_clang_arg_iterator";
+    stage2_panic(msg, strlen(msg));
+}
+
+enum Error stage2_clang_arg_next(struct Stage2ClangArgIterator *it) {
+    const char *msg = "stage0 called stage2_clang_arg_next";
+    stage2_panic(msg, strlen(msg));
+}

src/stage2.h

@@ -105,6 +105,7 @@ enum Error {
     ErrorTargetHasNoDynamicLinker,
     ErrorInvalidAbiVersion,
     ErrorInvalidOperatingSystemVersion,
+    ErrorUnknownClangOption,
 };
 
 // ABI warning
@@ -316,4 +317,34 @@ struct Stage2NativePaths {
 // ABI warning
 ZIG_EXTERN_C enum Error stage2_detect_native_paths(struct Stage2NativePaths *native_paths);
 
+// ABI warning
+enum Stage2ClangArg {
+    Stage2ClangArgTarget,
+    Stage2ClangArgO,
+    Stage2ClangArgC,
+    Stage2ClangArgOther,
+    Stage2ClangArgPositional,
+    Stage2ClangArgL,
+};
+
+// ABI warning
+struct Stage2ClangArgIterator {
+    bool has_next;
+    enum Stage2ClangArg kind;
+    const char *only_arg;
+    const char *second_arg;
+    const char **other_args_ptr;
+    size_t other_args_len;
+    const char **argv_ptr;
+    size_t argv_len;
+    size_t next_index;
+};
+
+// ABI warning
+ZIG_EXTERN_C void stage2_clang_arg_iterator(struct Stage2ClangArgIterator *it,
+        size_t argc, char **argv);
+
+// ABI warning
+ZIG_EXTERN_C enum Error stage2_clang_arg_next(struct Stage2ClangArgIterator *it);
+
 #endif

tools/process_headers.zig

@@ -1,14 +1,14 @@
-// To get started, run this tool with no args and read the help message.
-//
-// The build systems of musl-libc and glibc require specifying a single target
-// architecture. Meanwhile, Zig supports out-of-the-box cross compilation for
-// every target. So the process to create libc headers that Zig ships is to use
-// this tool.
-// First, use the musl/glibc build systems to create installations of all the
-// targets in the `glibc_targets`/`musl_targets` variables.
-// Next, run this tool to create a new directory which puts .h files into
-// <arch> subdirectories, with `generic` being files that apply to all architectures.
-// You'll then have to manually update Zig source repo with these new files.
+//! To get started, run this tool with no args and read the help message.
+//!
+//! The build systems of musl-libc and glibc require specifying a single target
+//! architecture. Meanwhile, Zig supports out-of-the-box cross compilation for
+//! every target. So the process to create libc headers that Zig ships is to use
+//! this tool.
+//! First, use the musl/glibc build systems to create installations of all the
+//! targets in the `glibc_targets`/`musl_targets` variables.
+//! Next, run this tool to create a new directory which puts .h files into
+//! <arch> subdirectories, with `generic` being files that apply to all architectures.
+//! You'll then have to manually update Zig source repo with these new files.
 
 const std = @import("std");
 const Arch = std.Target.Cpu.Arch;

tools/update_clang_options.zig

@@ -0,0 +1,274 @@
+//! To get started, run this tool with no args and read the help message.
+//!
+//! Clang has a file "options.td" which describes all of its command line parameter options.
+//! When using `zig cc`, Zig acts as a proxy between the user and Clang. It does not need
+//! to understand all the parameters, but it does need to understand some of them, such as
+//! the target. This means that Zig must understand when a C command line parameter expects
+//! to "consume" the next parameter on the command line.
+//!
+//! For example, `-z -target` would mean to pass `-target` to the linker, whereas `-E -target`
+//! would mean that the next parameter specifies the target.
+
+const std = @import("std");
+const fs = std.fs;
+const assert = std.debug.assert;
+const json = std.json;
+
+const KnownOpt = struct {
+    name: []const u8,
+
+    /// Corresponds to stage.zig ClangArgIterator.Kind
+    ident: []const u8,
+};
+
+const known_options = [_]KnownOpt{
+    .{
+        .name = "target",
+        .ident = "target",
+    },
+    .{
+        .name = "o",
+        .ident = "o",
+    },
+    .{
+        .name = "c",
+        .ident = "c",
+    },
+    .{
+        .name = "l",
+        .ident = "l",
+    },
+};
+
+const blacklisted_options = [_][]const u8{};
+
+fn knownOption(name: []const u8) ?[]const u8 {
+    const chopped_name = if (std.mem.endsWith(u8, name, "=")) name[0 .. name.len - 1] else name;
+    for (known_options) |item| {
+        if (std.mem.eql(u8, chopped_name, item.name)) {
+            return item.ident;
+        }
+    }
+    return null;
+}
+
+pub fn main() anyerror!void {
+    var arena = std.heap.ArenaAllocator.init(std.heap.page_allocator);
+    defer arena.deinit();
+
+    const allocator = &arena.allocator;
+    const args = try std.process.argsAlloc(allocator);
+
+    if (args.len <= 1) {
+        usageAndExit(std.io.getStdErr(), args[0], 1);
+    }
+    if (std.mem.eql(u8, args[1], "--help")) {
+        usageAndExit(std.io.getStdOut(), args[0], 0);
+    }
+    if (args.len < 3) {
+        usageAndExit(std.io.getStdErr(), args[0], 1);
+    }
+
+    const llvm_tblgen_exe = args[1];
+    if (std.mem.startsWith(u8, llvm_tblgen_exe, "-")) {
+        usageAndExit(std.io.getStdErr(), args[0], 1);
+    }
+
+    const llvm_src_root = args[2];
+    if (std.mem.startsWith(u8, llvm_src_root, "-")) {
+        usageAndExit(std.io.getStdErr(), args[0], 1);
+    }
+
+    const child_args = [_][]const u8{
+        llvm_tblgen_exe,
+        "--dump-json",
+        try std.fmt.allocPrint(allocator, "{}/clang/include/clang/Driver/Options.td", .{llvm_src_root}),
+        try std.fmt.allocPrint(allocator, "-I={}/llvm/include", .{llvm_src_root}),
+        try std.fmt.allocPrint(allocator, "-I={}/clang/include/clang/Driver", .{llvm_src_root}),
+    };
+
+    const child_result = try std.ChildProcess.exec2(.{
+        .allocator = allocator,
+        .argv = &child_args,
+        .max_output_bytes = 100 * 1024 * 1024,
+    });
+
+    std.debug.warn("{}\n", .{child_result.stderr});
+
+    const json_text = switch (child_result.term) {
+        .Exited => |code| if (code == 0) child_result.stdout else {
+            std.debug.warn("llvm-tblgen exited with code {}\n", .{code});
+            std.process.exit(1);
+        },
+        else => {
+            std.debug.warn("llvm-tblgen crashed\n", .{});
+            std.process.exit(1);
+        },
+    };
+
+    var parser = json.Parser.init(allocator, false);
+    const tree = try parser.parse(json_text);
+    const root_map = &tree.root.Object;
+
+    var all_names = std.ArrayList([]const u8).init(allocator);
+    {
+        var it = root_map.iterator();
+        it_map: while (it.next()) |kv| {
+            if (kv.key.len == 0) continue;
+            if (kv.key[0] == '!') continue;
+            if (kv.value != .Object) continue;
+            if (!kv.value.Object.contains("NumArgs")) continue;
+            if (!kv.value.Object.contains("Name")) continue;
+            for (blacklisted_options) |blacklisted_key| {
+                if (std.mem.eql(u8, blacklisted_key, kv.key)) continue :it_map;
+            }
+            if (kv.value.Object.get("Name").?.value.String.len == 0) continue;
+            try all_names.append(kv.key);
+        }
+    }
+    std.sort.sort([]const u8, all_names.span(), nameLessThan);
+
+    var stdout_bos = std.io.bufferedOutStream(std.io.getStdOut().outStream());
+    const stdout = stdout_bos.outStream();
+    try stdout.writeAll(
+        \\// This file is generated by tools/update_clang_options.zig.
+        \\// zig fmt: off
+        \\usingnamespace @import("clang_options.zig");
+        \\pub const data = blk: { @setEvalBranchQuota(6000); break :blk &[_]CliArg{
+        \\
+    );
+
+    for (all_names.span()) |key| {
+        const obj = &root_map.get(key).?.value.Object;
+        const name = obj.get("Name").?.value.String;
+        var pd1 = false;
+        var pd2 = false;
+        var pslash = false;
+        for (obj.get("Prefixes").?.value.Array.span()) |prefix_json| {
+            const prefix = prefix_json.String;
+            if (std.mem.eql(u8, prefix, "-")) {
+                pd1 = true;
+            } else if (std.mem.eql(u8, prefix, "--")) {
+                pd2 = true;
+            } else if (std.mem.eql(u8, prefix, "/")) {
+                pslash = true;
+            } else {
+                std.debug.warn("{} (key {}) has unrecognized prefix '{}'\n", .{ name, key, prefix });
+                std.process.exit(1);
+            }
+        }
+        const num_args = @intCast(u8, obj.get("NumArgs").?.value.Integer);
+        const syntax_str: []const u8 = blk: {
+            for (obj.get("!superclasses").?.value.Array.span()) |superclass_json| {
+                const superclass = superclass_json.String;
+                if (std.mem.eql(u8, superclass, "Joined")) {
+                    break :blk ".joined";
+                } else if (std.mem.eql(u8, superclass, "CLJoined")) {
+                    break :blk ".joined";
+                } else if (std.mem.eql(u8, superclass, "CLIgnoredJoined")) {
+                    break :blk ".joined";
+                } else if (std.mem.eql(u8, superclass, "CLCompileJoined")) {
+                    break :blk ".joined";
+                } else if (std.mem.eql(u8, superclass, "JoinedOrSeparate")) {
+                    break :blk ".joined_or_separate";
+                } else if (std.mem.eql(u8, superclass, "CLJoinedOrSeparate")) {
+                    break :blk ".joined_or_separate";
+                } else if (std.mem.eql(u8, superclass, "CLCompileJoinedOrSeparate")) {
+                    break :blk ".joined_or_separate";
+                } else if (std.mem.eql(u8, superclass, "Flag")) {
+                    break :blk ".flag";
+                } else if (std.mem.eql(u8, superclass, "CLFlag")) {
+                    break :blk ".flag";
+                } else if (std.mem.eql(u8, superclass, "CLIgnoredFlag")) {
+                    break :blk ".flag";
+                } else if (std.mem.eql(u8, superclass, "Separate")) {
+                    break :blk ".separate";
+                } else if (std.mem.eql(u8, superclass, "JoinedAndSeparate")) {
+                    break :blk ".joined_and_separate";
+                } else if (std.mem.eql(u8, superclass, "CommaJoined")) {
+                    break :blk ".comma_joined";
+                } else if (std.mem.eql(u8, superclass, "CLRemainingArgsJoined")) {
+                    break :blk ".remaining_args_joined";
+                } else if (std.mem.eql(u8, superclass, "MultiArg")) {
+                    break :blk try std.fmt.allocPrint(allocator, ".{{ .multi_arg = {} }}", .{num_args});
+                }
+            }
+            if (std.mem.eql(u8, name, "<input>")) {
+                break :blk ".flag";
+            } else if (std.mem.eql(u8, name, "<unknown>")) {
+                break :blk ".flag";
+            }
+            const kind_def = obj.get("Kind").?.value.Object.get("def").?.value.String;
+            if (std.mem.eql(u8, kind_def, "KIND_FLAG")) {
+                break :blk ".flag";
+            }
+            std.debug.warn("{} (key {}) has unrecognized superclasses:\n", .{ name, key });
+            for (obj.get("!superclasses").?.value.Array.span()) |superclass_json| {
+                std.debug.warn(" {}\n", .{superclass_json.String});
+            }
+            std.process.exit(1);
+        };
+        if (knownOption(name)) |ident| {
+            try stdout.print(
+                \\.{{
+                \\    .name = "{}",
+                \\    .syntax = {},
+                \\    .zig_equivalent = .{},
+                \\    .pd1 = {},
+                \\    .pd2 = {},
+                \\    .psl = {},
+                \\}},
+                \\
+            , .{ name, syntax_str, ident, pd1, pd2, pslash });
+        } else if (pd1 and !pd2 and !pslash and
+            std.mem.eql(u8, syntax_str, ".flag"))
+        {
+            try stdout.print("flagpd1(\"{}\"),\n", .{name});
+        } else if (pd1 and !pd2 and !pslash and
+            std.mem.eql(u8, syntax_str, ".joined"))
+        {
+            try stdout.print("joinpd1(\"{}\"),\n", .{name});
+        } else if (pd1 and !pd2 and !pslash and
+            std.mem.eql(u8, syntax_str, ".joined_or_separate"))
+        {
+            try stdout.print("jspd1(\"{}\"),\n", .{name});
+        } else if (pd1 and !pd2 and !pslash and
+            std.mem.eql(u8, syntax_str, ".separate"))
+        {
+            try stdout.print("sepd1(\"{}\"),\n", .{name});
+        } else {
+            try stdout.print(
+                \\.{{
+                \\    .name = "{}",
+                \\    .syntax = {},
+                \\    .zig_equivalent = .other,
+                \\    .pd1 = {},
+                \\    .pd2 = {},
+                \\    .psl = {},
+                \\}},
+                \\
+            , .{ name, syntax_str, pd1, pd2, pslash });
+        }
+    }
+
+    try stdout.writeAll(
+        \\};};
+        \\
+    );
+
+    try stdout_bos.flush();
+}
+
+fn nameLessThan(a: []const u8, b: []const u8) bool {
+    return std.mem.lessThan(u8, a, b);
+}
+
+fn usageAndExit(file: fs.File, arg0: []const u8, code: u8) noreturn {
+    file.outStream().print(
+        \\Usage: {} /path/to/llvm-tblgen /path/to/git/llvm/llvm-project
+        \\
+        \\Prints to stdout Zig code which you can use to replace the file src-self-hosted/clang_options_data.zig.
+        \\
+    , .{arg0}) catch std.process.exit(1);
+    std.process.exit(code);
+}