Sync Aro sources (#19199)

ref: 02353ad9f17f659e173f68975a442fcec3dd2c94

Author: Vexu

.gitattributes

@@ -12,4 +12,4 @@ lib/libcxx/** linguist-vendored
 lib/libcxxabi/** linguist-vendored
 lib/libunwind/** linguist-vendored
 lib/tsan/** linguist-vendored
-deps/** linguist-vendored
+lib/compiler/aro/** linguist-vendored

lib/compiler/aro/README.md

@@ -20,8 +20,7 @@ int main(void) {
     printf("Hello, world!\n");
     return 0;
 }
-$ zig build run -- hello.c -o hello
+$ zig build && ./zig-out/bin/arocc hello.c -o hello
 $ ./hello
 Hello, world!
-$
 ```

lib/compiler/aro/aro/Attribute/names.zig

@@ -1,4 +1,4 @@
-//! Autogenerated by GenerateDef from deps/aro/aro/Attribute/names.def, do not edit
+//! Autogenerated by GenerateDef from src/aro/Attribute/names.def, do not edit
 // zig fmt: off
 
 const std = @import("std");
@@ -142,34 +142,24 @@ pub fn nameFromUniqueIndex(index: u16, buf: []u8) []u8 {
     return fbs.getWritten();
 }
 
-/// We're 1 bit shy of being able to fit this in a u32:
-/// - char only contains 0-9, a-z, A-Z, and _, so it could use a enum(u6) with a way to convert <-> u8
-///   (note: this would have a performance cost that may make the u32 not worth it)
-/// - number has a max value of > 2047 and < 4095 (the first _ node has the largest number),
-///   so it could fit into a u12
-/// - child_index currently has a max of > 4095 and < 8191, so it could fit into a u13
-///
-/// with the end_of_word/end_of_list 2 bools, that makes 33 bits total
-const Node = packed struct(u64) {
+const Node = packed struct(u32) {
     char: u8,
     /// Nodes are numbered with "an integer which gives the number of words that
     /// would be accepted by the automaton starting from that state." This numbering
     /// allows calculating "a one-to-one correspondence between the integers 1 to L
     /// (L is the number of words accepted by the automaton) and the words themselves."
     ///
     /// Essentially, this allows us to have a minimal perfect hashing scheme such that
-    /// it's possible to store & lookup the properties of each builtin using a separate array.
-    number: u16,
-    /// If true, this node is the end of a valid builtin.
+    /// it's possible to store & lookup the properties of each name using a separate array.
+    number: u8,
+    /// If true, this node is the end of a valid name.
     /// Note: This does not necessarily mean that this node does not have child nodes.
     end_of_word: bool,
     /// If true, this node is the end of a sibling list.
     /// If false, then (index + 1) will contain the next sibling.
     end_of_list: bool,
-    /// Padding bits to get to u64, unsure if there's some way to use these to improve something.
-    _extra: u22 = 0,
     /// Index of the first child of this node.
-    child_index: u16,
+    child_index: u14,
 };
 
 const dafsa = [_]Node{

lib/compiler/aro/aro/Builtins.zig

@@ -99,10 +99,7 @@ fn createType(desc: TypeDescription, it: *TypeDescription.TypeIterator, comp: *c
             }
         },
         .h => builder.combine(undefined, .fp16, 0) catch unreachable,
-        .x => {
-            // Todo: _Float16
-            return .{ .specifier = .invalid };
-        },
+        .x => builder.combine(undefined, .float16, 0) catch unreachable,
         .y => {
             // Todo: __bf16
             return .{ .specifier = .invalid };

lib/compiler/aro/aro/Builtins/Builtin.zig

@@ -1,4 +1,4 @@
-//! Autogenerated by GenerateDef from deps/aro/aro/Builtins/Builtin.def, do not edit
+//! Autogenerated by GenerateDef from src/aro/Builtins/Builtin.def, do not edit
 // zig fmt: off
 
 const std = @import("std");

lib/compiler/aro/aro/Compilation.zig

@@ -241,6 +241,12 @@ pub const SystemDefinesMode = enum {
 fn generateSystemDefines(comp: *Compilation, w: anytype) !void {
     const ptr_width = comp.target.ptrBitWidth();
 
+    if (comp.langopts.gnuc_version > 0) {
+        try w.print("#define __GNUC__ {d}\n", .{comp.langopts.gnuc_version / 10_000});
+        try w.print("#define __GNUC_MINOR__ {d}\n", .{comp.langopts.gnuc_version / 100 % 100});
+        try w.print("#define __GNUC_PATCHLEVEL__ {d}\n", .{comp.langopts.gnuc_version % 100});
+    }
+
     // os macros
     switch (comp.target.os.tag) {
         .linux => try w.writeAll(
@@ -419,6 +425,25 @@ fn generateSystemDefines(comp: *Compilation, w: anytype) !void {
         \\
     );
 
+    // TODO: Set these to target-specific constants depending on backend capabilities
+    // For now they are just set to the "may be lock-free" value
+    try w.writeAll(
+        \\#define __ATOMIC_BOOL_LOCK_FREE 1
+        \\#define __ATOMIC_CHAR_LOCK_FREE 1
+        \\#define __ATOMIC_CHAR16_T_LOCK_FREE 1
+        \\#define __ATOMIC_CHAR32_T_LOCK_FREE 1
+        \\#define __ATOMIC_WCHAR_T_LOCK_FREE 1
+        \\#define __ATOMIC_SHORT_LOCK_FREE 1
+        \\#define __ATOMIC_INT_LOCK_FREE 1
+        \\#define __ATOMIC_LONG_LOCK_FREE 1
+        \\#define __ATOMIC_LLONG_LOCK_FREE 1
+        \\#define __ATOMIC_POINTER_LOCK_FREE 1
+        \\
+    );
+    if (comp.langopts.hasChar8_T()) {
+        try w.writeAll("#define __ATOMIC_CHAR8_T_LOCK_FREE 1\n");
+    }
+
     // types
     if (comp.getCharSignedness() == .unsigned) try w.writeAll("#define __CHAR_UNSIGNED__ 1\n");
     try w.writeAll("#define __CHAR_BIT__ 8\n");
@@ -438,6 +463,7 @@ fn generateSystemDefines(comp: *Compilation, w: anytype) !void {
     try comp.generateIntMaxAndWidth(w, "PTRDIFF", comp.types.ptrdiff);
     try comp.generateIntMaxAndWidth(w, "INTPTR", comp.types.intptr);
     try comp.generateIntMaxAndWidth(w, "UINTPTR", comp.types.intptr.makeIntegerUnsigned());
+    try comp.generateIntMaxAndWidth(w, "SIG_ATOMIC", target_util.sigAtomicType(comp.target));
 
     // int widths
     try w.print("#define __BITINT_MAXWIDTH__ {d}\n", .{bit_int_max_bits});
@@ -474,6 +500,8 @@ fn generateSystemDefines(comp: *Compilation, w: anytype) !void {
     try generateTypeMacro(w, mapper, "__PTRDIFF_TYPE__", comp.types.ptrdiff, comp.langopts);
     try generateTypeMacro(w, mapper, "__SIZE_TYPE__", comp.types.size, comp.langopts);
     try generateTypeMacro(w, mapper, "__WCHAR_TYPE__", comp.types.wchar, comp.langopts);
+    try generateTypeMacro(w, mapper, "__CHAR16_TYPE__", comp.types.uint_least16_t, comp.langopts);
+    try generateTypeMacro(w, mapper, "__CHAR32_TYPE__", comp.types.uint_least32_t, comp.langopts);
 
     try comp.generateExactWidthTypes(w, mapper);
     try comp.generateFastAndLeastWidthTypes(w, mapper);
@@ -518,7 +546,6 @@ pub fn generateBuiltinMacros(comp: *Compilation, system_defines_mode: SystemDefi
 
     // standard macros
     try buf.appendSlice(
-        \\#define __STDC_NO_ATOMICS__ 1
         \\#define __STDC_NO_COMPLEX__ 1
         \\#define __STDC_NO_THREADS__ 1
         \\#define __STDC_NO_VLA__ 1
@@ -1030,9 +1057,8 @@ pub fn getCharSignedness(comp: *const Compilation) std.builtin.Signedness {
     return comp.langopts.char_signedness_override orelse comp.target.charSignedness();
 }
 
-pub fn defineSystemIncludes(comp: *Compilation, aro_dir: []const u8) !void {
-    var stack_fallback = std.heap.stackFallback(path_buf_stack_limit, comp.gpa);
-    const allocator = stack_fallback.get();
+/// Add built-in aro headers directory to system include paths
+pub fn addBuiltinIncludeDir(comp: *Compilation, aro_dir: []const u8) !void {
     var search_path = aro_dir;
     while (std.fs.path.dirname(search_path)) |dirname| : (search_path = dirname) {
         var base_dir = std.fs.cwd().openDir(dirname, .{}) catch continue;
@@ -1044,23 +1070,12 @@ pub fn defineSystemIncludes(comp: *Compilation, aro_dir: []const u8) !void {
         try comp.system_include_dirs.append(comp.gpa, path);
         break;
     } else return error.AroIncludeNotFound;
+}
 
-    if (comp.target.os.tag == .linux) {
-        const triple_str = try comp.target.linuxTriple(allocator);
-        defer allocator.free(triple_str);
-
-        const multiarch_path = try std.fs.path.join(allocator, &.{ "/usr/include", triple_str });
-        defer allocator.free(multiarch_path);
-
-        if (!std.meta.isError(std.fs.accessAbsolute(multiarch_path, .{}))) {
-            const duped = try comp.gpa.dupe(u8, multiarch_path);
-            errdefer comp.gpa.free(duped);
-            try comp.system_include_dirs.append(comp.gpa, duped);
-        }
-    }
-    const usr_include = try comp.gpa.dupe(u8, "/usr/include");
-    errdefer comp.gpa.free(usr_include);
-    try comp.system_include_dirs.append(comp.gpa, usr_include);
+pub fn addSystemIncludeDir(comp: *Compilation, path: []const u8) !void {
+    const duped = try comp.gpa.dupe(u8, path);
+    errdefer comp.gpa.free(duped);
+    try comp.system_include_dirs.append(comp.gpa, duped);
 }
 
 pub fn getSource(comp: *const Compilation, id: Source.Id) Source {
@@ -1331,6 +1346,10 @@ pub fn hasInclude(
     /// __has_include vs __has_include_next
     which: WhichInclude,
 ) !bool {
+    if (mem.indexOfScalar(u8, filename, 0) != null) {
+        return false;
+    }
+
     const cwd = std.fs.cwd();
     if (std.fs.path.isAbsolute(filename)) {
         if (which == .next) return false;

lib/compiler/aro/aro/Diagnostics.zig

@@ -208,6 +208,8 @@ pub const Options = struct {
     @"unsupported-embed-param": Kind = .default,
     @"unused-result": Kind = .default,
     normalized: Kind = .default,
+    @"shift-count-negative": Kind = .default,
+    @"shift-count-overflow": Kind = .default,
 };
 
 const Diagnostics = @This();
@@ -291,7 +293,7 @@ pub fn addExtra(
                 .kind = .note,
                 .extra = .{ .unsigned = expansion_locs.len - d.macro_backtrace_limit },
             });
-            i = half - 1;
+            i = half -| 1;
             while (i > 0) {
                 i -= 1;
                 d.list.appendAssumeCapacity(.{

lib/compiler/aro/aro/Diagnostics/messages.zig

@@ -1,4 +1,4 @@
-//! Autogenerated by GenerateDef from deps/aro/aro/Diagnostics/messages.def, do not edit
+//! Autogenerated by GenerateDef from src/aro/Diagnostics/messages.def, do not edit
 // zig fmt: off
 
 const std = @import("std");
@@ -504,6 +504,11 @@ pub const Tag = enum {
     c23_auto_single_declarator,
     c32_auto_requires_initializer,
     c23_auto_scalar_init,
+    negative_shift_count,
+    too_big_shift_count,
+    complex_conj,
+    overflow_builtin_requires_int,
+    overflow_result_requires_ptr,
 
     pub fn property(tag: Tag) Properties {
         return named_data[@intFromEnum(tag)];
@@ -1005,6 +1010,11 @@ pub const Tag = enum {
         .{ .msg = "'auto' can only be used with a single declarator", .kind = .@"error" },
         .{ .msg = "'auto' requires an initializer", .kind = .@"error" },
         .{ .msg = "'auto' requires a scalar initializer", .kind = .@"error" },
+        .{ .msg = "shift count is negative", .opt = W("shift-count-negative"), .kind = .warning, .all = true },
+        .{ .msg = "shift count >= width of type", .opt = W("shift-count-overflow"), .kind = .warning, .all = true },
+        .{ .msg = "ISO C does not support '~' for complex conjugation of '{s}'", .opt = W("pedantic"), .extra = .str, .kind = .off },
+        .{ .msg = "operand argument to overflow builtin must be an integer ('{s}' invalid)", .extra = .str, .kind = .@"error" },
+        .{ .msg = "result argument to overflow builtin must be a pointer to a non-const integer ('{s}' invalid)", .extra = .str, .kind = .@"error" },
     };
 };
 };

lib/compiler/aro/aro/Driver.zig

@@ -12,6 +12,7 @@ const Preprocessor = @import("Preprocessor.zig");
 const Source = @import("Source.zig");
 const Toolchain = @import("Toolchain.zig");
 const target_util = @import("target.zig");
+const GCCVersion = @import("Driver/GCCVersion.zig");
 
 pub const Linker = enum {
     ld,
@@ -43,6 +44,9 @@ verbose_pp: bool = false,
 verbose_ir: bool = false,
 verbose_linker_args: bool = false,
 color: ?bool = null,
+nobuiltininc: bool = false,
+nostdinc: bool = false,
+nostdlibinc: bool = false,
 
 /// Full path to the aro executable
 aro_name: []const u8 = "",
@@ -95,6 +99,7 @@ pub const usage =
     \\  -fcolor-diagnostics     Enable colors in diagnostics
     \\  -fno-color-diagnostics  Disable colors in diagnostics
     \\  -fdeclspec              Enable support for __declspec attributes
+    \\  -fgnuc-version=<value>  Controls value of __GNUC__ and related macros. Set to 0 or empty to disable them.
     \\  -fno-declspec           Disable support for __declspec attributes
     \\  -ffp-eval-method=[source|double|extended]
     \\                          Evaluation method to use for floating-point arithmetic
@@ -127,6 +132,10 @@ pub const usage =
     \\  -isystem                Add directory to SYSTEM include search path
     \\  --emulate=[clang|gcc|msvc]
     \\                          Select which C compiler to emulate (default clang)
+    \\  -nobuiltininc           Do not search the compiler's builtin directory for include files
+    \\  -nostdinc, --no-standard-includes
+    \\                          Do not search the standard system directories or compiler builtin directories for include files.
+    \\  -nostdlibinc            Do not search the standard system directories for include files, but do search compiler builtin include directories
     \\  -o <file>               Write output to <file>
     \\  -P, --no-line-commands  Disable linemarker output in -E mode
     \\  -pedantic               Warn on language extensions
@@ -180,6 +189,7 @@ pub fn parseArgs(
     var i: usize = 1;
     var comment_arg: []const u8 = "";
     var hosted: ?bool = null;
+    var gnuc_version: []const u8 = "4.2.1"; // default value set by clang
     while (i < args.len) : (i += 1) {
         const arg = args[i];
         if (mem.startsWith(u8, arg, "-") and arg.len > 1) {
@@ -303,6 +313,10 @@ pub fn parseArgs(
                 d.only_syntax = true;
             } else if (mem.startsWith(u8, arg, "-fno-syntax-only")) {
                 d.only_syntax = false;
+            } else if (mem.eql(u8, arg, "-fgnuc-version=")) {
+                gnuc_version = "0";
+            } else if (option(arg, "-fgnuc-version=")) |version| {
+                gnuc_version = version;
             } else if (mem.startsWith(u8, arg, "-isystem")) {
                 var path = arg["-isystem".len..];
                 if (path.len == 0) {
@@ -421,6 +435,12 @@ pub fn parseArgs(
                 d.nodefaultlibs = true;
             } else if (mem.eql(u8, arg, "-nolibc")) {
                 d.nolibc = true;
+            } else if (mem.eql(u8, arg, "-nobuiltininc")) {
+                d.nobuiltininc = true;
+            } else if (mem.eql(u8, arg, "-nostdinc") or mem.eql(u8, arg, "--no-standard-includes")) {
+                d.nostdinc = true;
+            } else if (mem.eql(u8, arg, "-nostdlibinc")) {
+                d.nostdlibinc = true;
             } else if (mem.eql(u8, arg, "-nostdlib")) {
                 d.nostdlib = true;
             } else if (mem.eql(u8, arg, "-nostartfiles")) {
@@ -459,6 +479,11 @@ pub fn parseArgs(
             d.comp.target.os.tag = .freestanding;
         }
     }
+    const version = GCCVersion.parse(gnuc_version);
+    if (version.major == -1) {
+        return d.fatal("invalid value '{0s}' in '-fgnuc-version={0s}'", .{gnuc_version});
+    }
+    d.comp.langopts.gnuc_version = version.toUnsigned();
     return false;
 }
 
@@ -558,7 +583,8 @@ pub fn main(d: *Driver, tc: *Toolchain, args: []const []const u8, comptime fast_
         try d.comp.addDiagnostic(.{ .tag = .cli_unused_link_object, .extra = .{ .str = obj } }, &.{});
     };
 
-    d.comp.defineSystemIncludes(d.aro_name) catch |er| switch (er) {
+    try tc.discover();
+    tc.defineSystemIncludes() catch |er| switch (er) {
         error.OutOfMemory => return error.OutOfMemory,
         error.AroIncludeNotFound => return d.fatal("unable to find Aro builtin headers", .{}),
     };
@@ -763,8 +789,6 @@ fn dumpLinkerArgs(items: []const []const u8) !void {
 /// The entry point of the Aro compiler.
 /// **MAY call `exit` if `fast_exit` is set.**
 pub fn invokeLinker(d: *Driver, tc: *Toolchain, comptime fast_exit: bool) !void {
-    try tc.discover();
-
     var argv = std.ArrayList([]const u8).init(d.comp.gpa);
     defer argv.deinit();
 

lib/compiler/aro/aro/Driver/GCCVersion.zig

@@ -98,6 +98,16 @@ pub fn order(a: GCCVersion, b: GCCVersion) Order {
     return .eq;
 }
 
+/// Used for determining __GNUC__ macro values
+/// This matches clang's logic for overflowing values
+pub fn toUnsigned(self: GCCVersion) u32 {
+    var result: u32 = 0;
+    if (self.major > 0) result = @as(u32, @intCast(self.major)) *% 10_000;
+    if (self.minor > 0) result +%= @as(u32, @intCast(self.minor)) *% 100;
+    if (self.patch > 0) result +%= @as(u32, @intCast(self.patch));
+    return result;
+}
+
 test parse {
     const versions = [10]GCCVersion{
         parse("5"),