std.heap: rework c_allocator

The main goal here was to avoid allocating padding and header space if
`malloc` already guarantees the alignment we need via `max_align_t`.
Previously, the compiler was using `std.heap.raw_c_allocator` as its GPA
in some cases depending on `std.c.max_align_t`, but that's pretty
fragile (it meant we had to encode our alignment requirements into
`src/main.zig`!). Perhaps more importantly, that solution is
unnecessarily restrictive: since Zig's `Allocator` API passes the
`Alignment` not only to `alloc`, but also to `free` etc, we are able to
use a different strategy depending on its value. So `c_allocator` can
simply compare the requested align to `Alignment.of(std.c.max_align_t)`,
and use a raw `malloc` call (no header needed!) if it will guarantee a
suitable alignment (which, in practice, will be true the vast majority
of the time).

So in short, this makes `std.heap.c_allocator` more memory efficient,
and probably removes any incentive to use `std.heap.raw_c_allocator`.

I also refactored the `c_allocator` implementation while doing this,
just to neaten things up a little.

build.zig

@@ -91,6 +91,8 @@ pub fn build(b: *std.Build) !void {
     const skip_libc = b.option(bool, "skip-libc", "Main test suite skips tests that link libc") orelse false;
     const skip_single_threaded = b.option(bool, "skip-single-threaded", "Main test suite skips tests that are single-threaded") orelse false;
     const skip_compile_errors = b.option(bool, "skip-compile-errors", "Main test suite skips compile error tests") orelse false;
+    const skip_spirv = b.option(bool, "skip-spirv", "Main test suite skips targets with spirv32/spirv64 architecture") orelse false;
+    const skip_wasm = b.option(bool, "skip-wasm", "Main test suite skips targets with wasm32/wasm64 architecture") orelse false;
     const skip_freebsd = b.option(bool, "skip-freebsd", "Main test suite skips targets with freebsd OS") orelse false;
     const skip_netbsd = b.option(bool, "skip-netbsd", "Main test suite skips targets with netbsd OS") orelse false;
     const skip_windows = b.option(bool, "skip-windows", "Main test suite skips targets with windows OS") orelse false;
@@ -421,6 +423,8 @@ pub fn build(b: *std.Build) !void {
         .test_target_filters = test_target_filters,
         .skip_compile_errors = skip_compile_errors,
         .skip_non_native = skip_non_native,
+        .skip_spirv = skip_spirv,
+        .skip_wasm = skip_wasm,
         .skip_freebsd = skip_freebsd,
         .skip_netbsd = skip_netbsd,
         .skip_windows = skip_windows,
@@ -452,6 +456,8 @@ pub fn build(b: *std.Build) !void {
         .skip_single_threaded = skip_single_threaded,
         .skip_non_native = skip_non_native,
         .test_default_only = no_matrix,
+        .skip_spirv = skip_spirv,
+        .skip_wasm = skip_wasm,
         .skip_freebsd = skip_freebsd,
         .skip_netbsd = skip_netbsd,
         .skip_windows = skip_windows,
@@ -459,8 +465,29 @@ pub fn build(b: *std.Build) !void {
         .skip_linux = skip_linux,
         .skip_llvm = skip_llvm,
         .skip_libc = skip_libc,
-        // 3888779264 was observed on an x86_64-linux-gnu host.
-        .max_rss = 4000000000,
+        .max_rss = switch (b.graph.host.result.os.tag) {
+            .freebsd => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 1_060_217_241,
+                else => 1_100_000_000,
+            },
+            .linux => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 659_809_075,
+                .loongarch64 => 598_902_374,
+                .riscv64 => 731_258_880,
+                .s390x => 580_596_121,
+                .x86_64 => 3_290_894_745,
+                else => 3_300_000_000,
+            },
+            .macos => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 767_736_217,
+                else => 800_000_000,
+            },
+            .windows => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 603_070_054,
+                else => 700_000_000,
+            },
+            else => 3_300_000_000,
+        },
     }));
 
     test_modules_step.dependOn(tests.addModuleTests(b, .{
@@ -475,6 +502,8 @@ pub fn build(b: *std.Build) !void {
         .skip_single_threaded = true,
         .skip_non_native = skip_non_native,
         .test_default_only = no_matrix,
+        .skip_spirv = skip_spirv,
+        .skip_wasm = skip_wasm,
         .skip_freebsd = skip_freebsd,
         .skip_netbsd = skip_netbsd,
         .skip_windows = skip_windows,
@@ -483,6 +512,29 @@ pub fn build(b: *std.Build) !void {
         .skip_llvm = skip_llvm,
         .skip_libc = true,
         .no_builtin = true,
+        .max_rss = switch (b.graph.host.result.os.tag) {
+            .freebsd => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 743_802_470,
+                else => 800_000_000,
+            },
+            .linux => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 639_565_414,
+                .loongarch64 => 598_884_352,
+                .riscv64 => 636_429_516,
+                .s390x => 574_166_630,
+                .x86_64 => 764_861_644,
+                else => 800_000_000,
+            },
+            .macos => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 701_413_785,
+                else => 800_000_000,
+            },
+            .windows => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 536_414_208,
+                else => 600_000_000,
+            },
+            else => 800_000_000,
+        },
     }));
 
     test_modules_step.dependOn(tests.addModuleTests(b, .{
@@ -497,6 +549,8 @@ pub fn build(b: *std.Build) !void {
         .skip_single_threaded = true,
         .skip_non_native = skip_non_native,
         .test_default_only = no_matrix,
+        .skip_spirv = skip_spirv,
+        .skip_wasm = skip_wasm,
         .skip_freebsd = skip_freebsd,
         .skip_netbsd = skip_netbsd,
         .skip_windows = skip_windows,
@@ -505,6 +559,29 @@ pub fn build(b: *std.Build) !void {
         .skip_llvm = skip_llvm,
         .skip_libc = true,
         .no_builtin = true,
+        .max_rss = switch (b.graph.host.result.os.tag) {
+            .freebsd => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 557_892_403,
+                else => 600_000_000,
+            },
+            .linux => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 615_302_758,
+                .loongarch64 => 598_974_464,
+                .riscv64 => 382_786_764,
+                .s390x => 395_555_635,
+                .x86_64 => 692_348_518,
+                else => 700_000_000,
+            },
+            .macos => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 451_389_030,
+                else => 500_000_000,
+            },
+            .windows => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 367_747_072,
+                else => 400_000_000,
+            },
+            else => 700_000_000,
+        },
     }));
 
     test_modules_step.dependOn(tests.addModuleTests(b, .{
@@ -519,6 +596,8 @@ pub fn build(b: *std.Build) !void {
         .skip_single_threaded = skip_single_threaded,
         .skip_non_native = skip_non_native,
         .test_default_only = no_matrix,
+        .skip_spirv = skip_spirv,
+        .skip_wasm = skip_wasm,
         .skip_freebsd = skip_freebsd,
         .skip_netbsd = skip_netbsd,
         .skip_windows = skip_windows,
@@ -526,8 +605,29 @@ pub fn build(b: *std.Build) !void {
         .skip_linux = skip_linux,
         .skip_llvm = skip_llvm,
         .skip_libc = skip_libc,
-        // I observed a value of 5605064704 on the M2 CI.
-        .max_rss = 6165571174,
+        .max_rss = switch (b.graph.host.result.os.tag) {
+            .freebsd => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 3_756_422_348,
+                else => 3_800_000_000,
+            },
+            .linux => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 6_732_817_203,
+                .loongarch64 => 3_216_349_593,
+                .riscv64 => 3_570_899_763,
+                .s390x => 3_652_514_201,
+                .x86_64 => 3_249_546_854,
+                else => 6_800_000_000,
+            },
+            .macos => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 8_273_795_481,
+                else => 8_300_000_000,
+            },
+            .windows => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 3_750_236_160,
+                else => 3_800_000_000,
+            },
+            else => 8_300_000_000,
+        },
     }));
 
     const unit_tests_step = b.step("test-unit", "Run the compiler source unit tests");
@@ -543,6 +643,29 @@ pub fn build(b: *std.Build) !void {
         .use_llvm = use_llvm,
         .use_lld = use_llvm,
         .zig_lib_dir = b.path("lib"),
+        .max_rss = switch (b.graph.host.result.os.tag) {
+            .freebsd => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 2_188_099_584,
+                else => 2_200_000_000,
+            },
+            .linux => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 1_991_934_771,
+                .loongarch64 => 1_844_538_572,
+                .riscv64 => 2_459_003_289,
+                .s390x => 1_781_248_409,
+                .x86_64 => 977_192_550,
+                else => 2_500_000_000,
+            },
+            .macos => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 2_062_393_344,
+                else => 2_100_000_000,
+            },
+            .windows => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 1_953_087_488,
+                else => 2_000_000_000,
+            },
+            else => 2_500_000_000,
+        },
     });
     if (link_libc) {
         unit_tests.root_module.link_libc = true;
@@ -560,6 +683,7 @@ pub fn build(b: *std.Build) !void {
     test_step.dependOn(tests.addCAbiTests(b, .{
         .test_target_filters = test_target_filters,
         .skip_non_native = skip_non_native,
+        .skip_wasm = skip_wasm,
         .skip_freebsd = skip_freebsd,
         .skip_netbsd = skip_netbsd,
         .skip_windows = skip_windows,
@@ -567,6 +691,29 @@ pub fn build(b: *std.Build) !void {
         .skip_linux = skip_linux,
         .skip_llvm = skip_llvm,
         .skip_release = skip_release,
+        .max_rss = switch (b.graph.host.result.os.tag) {
+            .freebsd => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 727_221_862,
+                else => 800_000_000,
+            },
+            .linux => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 1_318_185_369,
+                .loongarch64 => 1_422_904_524,
+                .riscv64 => 449_924_710,
+                .s390x => 1_946_743_603,
+                .x86_64 => 2_139_993_292,
+                else => 2_200_000_000,
+            },
+            .macos => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 1_813_612_134,
+                else => 1_900_000_000,
+            },
+            .windows => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 386_287_616,
+                else => 400_000_000,
+            },
+            else => 2_200_000_000,
+        },
     }));
     test_step.dependOn(tests.addLinkTests(b, enable_macos_sdk, enable_ios_sdk, enable_symlinks_windows));
     test_step.dependOn(tests.addStackTraceTests(b, test_filters, skip_non_native));
@@ -616,6 +763,7 @@ pub fn build(b: *std.Build) !void {
         .optimize_modes = optimization_modes,
         .test_filters = test_filters,
         .test_target_filters = test_target_filters,
+        .skip_wasm = skip_wasm,
         // Highest RSS observed in any test case was exactly 1802878976 on x86_64-linux.
         .max_rss = 2253598720,
     })) |test_libc_step| test_step.dependOn(test_libc_step);
@@ -721,7 +869,29 @@ fn addCompilerMod(b: *std.Build, options: AddCompilerModOptions) *std.Build.Modu
 fn addCompilerStep(b: *std.Build, options: AddCompilerModOptions) *std.Build.Step.Compile {
     const exe = b.addExecutable(.{
         .name = "zig",
-        .max_rss = 7_800_000_000,
+        .max_rss = switch (b.graph.host.result.os.tag) {
+            .freebsd => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 6_044_158_771,
+                else => 6_100_000_000,
+            },
+            .linux => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 6_240_805_683,
+                .loongarch64 => 5_024_158_515,
+                .riscv64 => 6_996_309_196,
+                .s390x => 4_997_174_476,
+                .x86_64 => 5_486_090_649,
+                else => 7_000_000_000,
+            },
+            .macos => switch (b.graph.host.result.cpu.arch) {
+                .aarch64 => 6_639_145_779,
+                else => 6_700_000_000,
+            },
+            .windows => switch (b.graph.host.result.cpu.arch) {
+                .x86_64 => 5_770_394_009,
+                else => 5_800_000_000,
+            },
+            else => 7_000_000_000,
+        },
         .root_module = addCompilerMod(b, options),
     });
     exe.stack_size = stack_size;
@@ -798,7 +968,7 @@ fn addCmakeCfgOptionsToExe(
                 };
                 mod.linkSystemLibrary("unwind", .{});
             },
-            .ios, .macos, .watchos, .tvos, .visionos => {
+            .driverkit, .ios, .maccatalyst, .macos, .tvos, .visionos, .watchos => {
                 mod.link_libcpp = true;
             },
             .windows => {

ci/aarch64-linux-debug.sh

@@ -44,7 +44,7 @@ ninja install
 
 # No -fqemu and -fwasmtime here as they're covered by the x86_64-linux scripts.
 stage3-debug/bin/zig build test docs \
-  --maxrss 44918199637 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dstatic-llvm \
   -Dskip-non-native \
   -Dtarget=native-native-musl \

ci/aarch64-linux-release.sh

@@ -44,7 +44,7 @@ ninja install
 
 # No -fqemu and -fwasmtime here as they're covered by the x86_64-linux scripts.
 stage3-release/bin/zig build test docs \
-  --maxrss 44918199637 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dstatic-llvm \
   -Dskip-non-native \
   -Dtarget=native-native-musl \

ci/aarch64-macos-debug.sh

@@ -42,6 +42,7 @@ cmake .. \
 ninja install
 
 stage3-debug/bin/zig build test docs \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   --zig-lib-dir "$PWD/../lib" \
   -Denable-macos-sdk \
   -Dstatic-llvm \

ci/aarch64-macos-release.sh

@@ -51,6 +51,7 @@ stage3-release/bin/zig build test docs \
 
 # Ensure that stage3 and stage4 are byte-for-byte identical.
 stage3-release/bin/zig build \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   --prefix stage4-release \
   -Denable-llvm \
   -Dno-lib \

ci/aarch64-windows.ps1

@@ -5,6 +5,7 @@ $ZIG_LLVM_CLANG_LLD_URL = "https://ziglang.org/deps/$ZIG_LLVM_CLANG_LLD_NAME.zip
 $PREFIX_PATH = "$(Get-Location)\..\$ZIG_LLVM_CLANG_LLD_NAME"
 $ZIG = "$PREFIX_PATH\bin\zig.exe"
 $ZIG_LIB_DIR = "$(Get-Location)\lib"
+$ZSF_MAX_RSS = if ($Env:ZSF_MAX_RSS) { $Env:ZSF_MAX_RSS } else { 0 }
 
 if (!(Test-Path "..\$ZIG_LLVM_CLANG_LLD_NAME.zip")) {
     Write-Output "Downloading $ZIG_LLVM_CLANG_LLD_URL"
@@ -54,6 +55,7 @@ CheckLastExitCode
 
 Write-Output "Main test suite..."
 & "stage3-release\bin\zig.exe" build test docs `
+  --maxrss $ZSF_MAX_RSS `
   --zig-lib-dir "$ZIG_LIB_DIR" `
   --search-prefix "$PREFIX_PATH" `
   -Dstatic-llvm `

ci/loongarch64-linux-debug.sh

@@ -45,7 +45,7 @@ ninja install
 
 # No -fqemu and -fwasmtime here as they're covered by the x86_64-linux scripts.
 stage3-debug/bin/zig build test docs \
-  --maxrss 60129542144 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dstatic-llvm \
   -Dskip-non-native \
   -Dtarget=native-native-musl \

ci/loongarch64-linux-release.sh

@@ -45,7 +45,7 @@ ninja install
 
 # No -fqemu and -fwasmtime here as they're covered by the x86_64-linux scripts.
 stage3-release/bin/zig build test docs \
-  --maxrss 60129542144 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dstatic-llvm \
   -Dskip-non-native \
   -Dtarget=native-native-musl \

ci/riscv64-linux-debug.sh

@@ -44,7 +44,7 @@ ninja install
 
 # No -fqemu and -fwasmtime here as they're covered by the x86_64-linux scripts.
 stage3-debug/bin/zig build test-cases test-modules test-unit test-c-abi test-stack-traces test-error-traces test-llvm-ir \
-  --maxrss 68719476736 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dstatic-llvm \
   -Dskip-non-native \
   -Dskip-single-threaded \

ci/riscv64-linux-release.sh

@@ -44,7 +44,7 @@ ninja install
 
 # No -fqemu and -fwasmtime here as they're covered by the x86_64-linux scripts.
 stage3-release/bin/zig build test-cases test-modules test-unit test-c-abi test-stack-traces test-error-traces test-llvm-ir \
-  --maxrss 68719476736 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dstatic-llvm \
   -Dskip-non-native \
   -Dskip-single-threaded \

ci/s390x-linux-debug.sh

@@ -45,7 +45,7 @@ ninja install
 
 # No -fqemu and -fwasmtime here as they're covered by the x86_64-linux scripts.
 stage3-debug/bin/zig build test docs \
-  --maxrss 30064771072 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dstatic-llvm \
   -Dskip-non-native \
   -Dtarget=native-native-musl \

ci/s390x-linux-release.sh

@@ -45,7 +45,7 @@ ninja install
 
 # No -fqemu and -fwasmtime here as they're covered by the x86_64-linux scripts.
 stage3-release/bin/zig build test docs \
-  --maxrss 30064771072 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dstatic-llvm \
   -Dskip-non-native \
   -Dtarget=native-native-musl \

ci/x86_64-freebsd-debug.sh

@@ -44,8 +44,10 @@ unset CXX
 ninja install
 
 stage3-debug/bin/zig build test docs \
-  --maxrss 42949672960 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dstatic-llvm \
+  -Dskip-spirv \
+  -Dskip-wasm \
   -Dskip-linux \
   -Dskip-netbsd \
   -Dskip-windows \

ci/x86_64-freebsd-release.sh

@@ -44,8 +44,10 @@ unset CXX
 ninja install
 
 stage3-release/bin/zig build test docs \
-  --maxrss 42949672960 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dstatic-llvm \
+  -Dskip-spirv \
+  -Dskip-wasm \
   -Dskip-linux \
   -Dskip-netbsd \
   -Dskip-windows \

ci/x86_64-linux-debug-llvm.sh

@@ -49,7 +49,7 @@ stage3-debug/bin/zig build \
   -Dno-lib
 
 stage3-debug/bin/zig build test docs \
-  --maxrss 21000000000 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dlldb=$HOME/deps/lldb-zig/Debug-e0a42bb34/bin/lldb \
   -fqemu \
   -fwasmtime \

ci/x86_64-linux-debug.sh

@@ -48,7 +48,7 @@ stage3-debug/bin/zig build \
   -Dno-lib
 
 stage3-debug/bin/zig build test docs \
-  --maxrss 21000000000 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dlldb=$HOME/deps/lldb-zig/Debug-e0a42bb34/bin/lldb \
   -fqemu \
   -fwasmtime \

ci/x86_64-linux-release.sh

@@ -54,7 +54,7 @@ stage3-release/bin/zig build \
   -Dno-lib
 
 stage3-release/bin/zig build test docs \
-  --maxrss 21000000000 \
+  --maxrss ${ZSF_MAX_RSS:-0} \
   -Dlldb=$HOME/deps/lldb-zig/Release-e0a42bb34/bin/lldb \
   -fqemu \
   -fwasmtime \

ci/x86_64-windows-debug.ps1

@@ -5,6 +5,7 @@ $ZIG_LLVM_CLANG_LLD_URL = "https://ziglang.org/deps/$ZIG_LLVM_CLANG_LLD_NAME.zip
 $PREFIX_PATH = "$($Env:USERPROFILE)\$ZIG_LLVM_CLANG_LLD_NAME"
 $ZIG = "$PREFIX_PATH\bin\zig.exe"
 $ZIG_LIB_DIR = "$(Get-Location)\lib"
+$ZSF_MAX_RSS = if ($Env:ZSF_MAX_RSS) { $Env:ZSF_MAX_RSS } else { 0 }
 
 if (!(Test-Path "$PREFIX_PATH.zip")) {
     Write-Output "Downloading $ZIG_LLVM_CLANG_LLD_URL"
@@ -54,12 +55,11 @@ CheckLastExitCode
 
 Write-Output "Main test suite..."
 & "stage3-debug\bin\zig.exe" build test docs `
-  --maxrss 32212254720 `
+  --maxrss $ZSF_MAX_RSS `
   --zig-lib-dir "$ZIG_LIB_DIR" `
   --search-prefix "$PREFIX_PATH" `
   -Dstatic-llvm `
   -Dskip-non-native `
-  -Dskip-release `
   -Dskip-test-incremental `
   -Denable-symlinks-windows `
   --test-timeout 30m

ci/x86_64-windows-release.ps1

@@ -5,6 +5,7 @@ $ZIG_LLVM_CLANG_LLD_URL = "https://ziglang.org/deps/$ZIG_LLVM_CLANG_LLD_NAME.zip
 $PREFIX_PATH = "$($Env:USERPROFILE)\$ZIG_LLVM_CLANG_LLD_NAME"
 $ZIG = "$PREFIX_PATH\bin\zig.exe"
 $ZIG_LIB_DIR = "$(Get-Location)\lib"
+$ZSF_MAX_RSS = if ($Env:ZSF_MAX_RSS) { $Env:ZSF_MAX_RSS } else { 0 }
 
 if (!(Test-Path "$PREFIX_PATH.zip")) {
     Write-Output "Downloading $ZIG_LLVM_CLANG_LLD_URL"
@@ -54,7 +55,7 @@ CheckLastExitCode
 
 Write-Output "Main test suite..."
 & "stage3-release\bin\zig.exe" build test docs `
-  --maxrss 32212254720 `
+  --maxrss $ZSF_MAX_RSS `
   --zig-lib-dir "$ZIG_LIB_DIR" `
   --search-prefix "$PREFIX_PATH" `
   -Dstatic-llvm `

lib/std/c.zig

@@ -4471,7 +4471,7 @@ pub const rusage = switch (native_os) {
         pub const SELF = 1;
         pub const CHILDREN = 2;
     },
-    .freebsd => extern struct {
+    .freebsd, .openbsd => extern struct {
         utime: timeval,
         stime: timeval,
         maxrss: c_long,
@@ -4493,6 +4493,27 @@ pub const rusage = switch (native_os) {
         pub const CHILDREN = -1;
         pub const THREAD = 1;
     },
+    .dragonfly, .netbsd => extern struct {
+        utime: timeval,
+        stime: timeval,
+        maxrss: c_long,
+        ixrss: c_long,
+        idrss: c_long,
+        isrss: c_long,
+        minflt: c_long,
+        majflt: c_long,
+        nswap: c_long,
+        inblock: c_long,
+        oublock: c_long,
+        msgsnd: c_long,
+        msgrcv: c_long,
+        nsignals: c_long,
+        nvcsw: c_long,
+        nivcsw: c_long,
+
+        pub const SELF = 0;
+        pub const CHILDREN = -1;
+    },
     else => void,
 };
 

lib/std/heap.zig

@@ -141,7 +141,19 @@ test defaultQueryPageSize {
     assert(std.math.isPowerOfTwo(defaultQueryPageSize()));
 }
 
-const CAllocator = struct {
+/// A wrapper around the C memory allocation API which supports the full `Allocator`
+/// interface, including arbitrary alignment. Simple `malloc` calls are used when
+/// possible, but large requested alignments may require larger buffers in order to
+/// satisfy the request. As well as `malloc`, `realloc`, and `free`, the extension
+/// functions `malloc_usable_size` and `posix_memalign` are used when available.
+///
+/// For an allocator that directly calls `malloc`/`realloc`/`free`, with no padding
+/// or special handling, see `raw_c_allocator`.
+pub const c_allocator: Allocator = .{
+    .ptr = undefined,
+    .vtable = &c_allocator_impl.vtable,
+};
+const c_allocator_impl = struct {
     comptime {
         if (!builtin.link_libc) {
             @compileError("C allocator is only available when linking against libc");
@@ -155,67 +167,55 @@ const CAllocator = struct {
         .free = free,
     };
 
-    pub const supports_malloc_size = @TypeOf(malloc_size) != void;
-    pub const malloc_size = if (@TypeOf(c.malloc_size) != void)
-        c.malloc_size
-    else if (@TypeOf(c.malloc_usable_size) != void)
-        c.malloc_usable_size
-    else if (@TypeOf(c._msize) != void)
-        c._msize
-    else {};
-
-    pub const supports_posix_memalign = switch (builtin.os.tag) {
-        .dragonfly, .netbsd, .freebsd, .illumos, .openbsd, .linux, .driverkit, .ios, .maccatalyst, .macos, .tvos, .visionos, .watchos, .serenity => true,
+    const have_posix_memalign = switch (builtin.os.tag) {
+        .dragonfly,
+        .netbsd,
+        .freebsd,
+        .illumos,
+        .openbsd,
+        .linux,
+        .driverkit,
+        .ios,
+        .maccatalyst,
+        .macos,
+        .tvos,
+        .visionos,
+        .watchos,
+        .serenity,
+        => true,
         else => false,
     };
 
-    fn getHeader(ptr: [*]u8) *[*]u8 {
-        return @ptrCast(@alignCast(ptr - @sizeOf(usize)));
+    fn allocStrat(need_align: Alignment) union(enum) {
+        raw,
+        posix_memalign: if (have_posix_memalign) void else noreturn,
+        manual_align: if (have_posix_memalign) noreturn else void,
+    } {
+        // If `malloc` guarantees `need_align`, always prefer a raw allocation.
+        if (Alignment.compare(need_align, .lte, .of(c.max_align_t))) {
+            return .raw;
+        }
+        // Use `posix_memalign` if available. Otherwise, we must manually align the allocation.
+        return if (have_posix_memalign) .posix_memalign else .manual_align;
     }
 
-    fn alignedAlloc(len: usize, alignment: Alignment) ?[*]u8 {
-        const alignment_bytes = alignment.toByteUnits();
-        if (supports_posix_memalign) {
-            // The posix_memalign only accepts alignment values that are a
-            // multiple of the pointer size
-            const effective_alignment = @max(alignment_bytes, @sizeOf(usize));
-
-            var aligned_ptr: ?*anyopaque = undefined;
-            if (c.posix_memalign(&aligned_ptr, effective_alignment, len) != 0)
-                return null;
-
-            return @ptrCast(aligned_ptr);
-        }
-
-        // Thin wrapper around regular malloc, overallocate to account for
-        // alignment padding and store the original malloc()'ed pointer before
-        // the aligned address.
-        const unaligned_ptr = @as([*]u8, @ptrCast(c.malloc(len + alignment_bytes - 1 + @sizeOf(usize)) orelse return null));
-        const unaligned_addr = @intFromPtr(unaligned_ptr);
-        const aligned_addr = mem.alignForward(usize, unaligned_addr + @sizeOf(usize), alignment_bytes);
-        const aligned_ptr = unaligned_ptr + (aligned_addr - unaligned_addr);
-        getHeader(aligned_ptr).* = unaligned_ptr;
-
-        return aligned_ptr;
-    }
-
-    fn alignedFree(ptr: [*]u8) void {
-        if (supports_posix_memalign) {
-            return c.free(ptr);
-        }
-
-        const unaligned_ptr = getHeader(ptr).*;
-        c.free(unaligned_ptr);
-    }
-
-    fn alignedAllocSize(ptr: [*]u8) usize {
-        if (supports_posix_memalign) {
-            return CAllocator.malloc_size(ptr);
-        }
-
-        const unaligned_ptr = getHeader(ptr).*;
-        const delta = @intFromPtr(ptr) - @intFromPtr(unaligned_ptr);
-        return CAllocator.malloc_size(unaligned_ptr) - delta;
+    /// If `allocStrat(a) == .manual_align`, an allocation looks like this:
+    ///
+    /// unaligned_ptr   hdr_ptr  aligned_ptr
+    /// v               v        v
+    /// +---------------+--------+--------------+
+    /// |    padding    | header | usable bytes |
+    /// +---------------+--------+--------------+
+    ///
+    /// * `unaligned_ptr` is the raw return value of `malloc`.
+    /// * `aligned_ptr` is computed by aligning `unaligned_ptr` forward; it is what `alloc` returns.
+    /// * `hdr_ptr` points to a pointer-sized header directly before the usable space. This header
+    ///   contains the value `unaligned_ptr`, so that we can pass it to `free` later. This is
+    ///   necessary because the width of the padding is unknown.
+    ///
+    /// This function accepts `aligned_ptr` and offsets it backwards to return `hdr_ptr`.
+    fn manualAlignHeader(aligned_ptr: [*]u8) *[*]u8 {
+        return @ptrCast(@alignCast(aligned_ptr - @sizeOf(usize)));
     }
 
     fn alloc(
@@ -226,67 +226,120 @@ const CAllocator = struct {
     ) ?[*]u8 {
         _ = return_address;
         assert(len > 0);
-        return alignedAlloc(len, alignment);
+        switch (allocStrat(alignment)) {
+            .raw => {
+                // C only needs to respect `max_align_t` up to the allocation size due to object
+                // alignment rules. If necessary, extend the allocation size.
+                const actual_len = @max(len, @alignOf(std.c.max_align_t));
+                const ptr = c.malloc(actual_len) orelse return null;
+                assert(alignment.check(@intFromPtr(ptr)));
+                return @ptrCast(ptr);
+            },
+            .posix_memalign => {
+                // The posix_memalign only accepts alignment values that are a
+                // multiple of the pointer size
+                const effective_alignment = @max(alignment.toByteUnits(), @sizeOf(usize));
+                var aligned_ptr: ?*anyopaque = undefined;
+                if (c.posix_memalign(&aligned_ptr, effective_alignment, len) != 0) {
+                    return null;
+                }
+                assert(alignment.check(@intFromPtr(aligned_ptr)));
+                return @ptrCast(aligned_ptr);
+            },
+            .manual_align => {
+                // Overallocate to account for alignment padding and store the original pointer
+                // returned by `malloc` before the aligned address.
+                const padded_len = len + @sizeOf(usize) + alignment.toByteUnits() - 1;
+                const unaligned_ptr: [*]u8 = @ptrCast(c.malloc(padded_len) orelse return null);
+                const unaligned_addr = @intFromPtr(unaligned_ptr);
+                const aligned_addr = alignment.forward(unaligned_addr + @sizeOf(usize));
+                const aligned_ptr = unaligned_ptr + (aligned_addr - unaligned_addr);
+                manualAlignHeader(aligned_ptr).* = unaligned_ptr;
+                return aligned_ptr;
+            },
+        }
     }
 
     fn resize(
         _: *anyopaque,
-        buf: []u8,
+        memory: []u8,
         alignment: Alignment,
         new_len: usize,
         return_address: usize,
     ) bool {
-        _ = alignment;
         _ = return_address;
-        if (new_len <= buf.len) {
-            return true;
+        assert(new_len > 0);
+        if (new_len <= memory.len) {
+            return true; // in-place shrink always works
         }
-        if (CAllocator.supports_malloc_size) {
-            const full_len = alignedAllocSize(buf.ptr);
-            if (new_len <= full_len) {
-                return true;
-            }
-        }
-        return false;
+        const mallocSize = func: {
+            if (@TypeOf(c.malloc_size) != void) break :func c.malloc_size;
+            if (@TypeOf(c.malloc_usable_size) != void) break :func c.malloc_usable_size;
+            if (@TypeOf(c._msize) != void) break :func c._msize;
+            return false; // we don't know how much space is actually available
+        };
+        const usable_len: usize = switch (allocStrat(alignment)) {
+            .raw, .posix_memalign => mallocSize(memory.ptr),
+            .manual_align => usable_len: {
+                const unaligned_ptr = manualAlignHeader(memory.ptr).*;
+                const full_len = mallocSize(unaligned_ptr);
+                const padding = @intFromPtr(memory.ptr) - @intFromPtr(unaligned_ptr);
+                break :usable_len full_len - padding;
+            },
+        };
+        return new_len <= usable_len;
     }
 
     fn remap(
-        context: *anyopaque,
+        ctx: *anyopaque,
         memory: []u8,
         alignment: Alignment,
         new_len: usize,
         return_address: usize,
     ) ?[*]u8 {
-        // realloc would potentially return a new allocation that does not
-        // respect the original alignment.
-        return if (resize(context, memory, alignment, new_len, return_address)) memory.ptr else null;
+        assert(new_len > 0);
+        // Prefer resizing in-place if possible, since `realloc` could be expensive even if legal.
+        if (resize(ctx, memory, alignment, new_len, return_address)) {
+            return memory.ptr;
+        }
+        switch (allocStrat(alignment)) {
+            .raw => {
+                // `malloc` and friends guarantee the required alignment, so we can try `realloc`.
+                // C only needs to respect `max_align_t` up to the allocation size due to object
+                // alignment rules. If necessary, extend the allocation size.
+                const actual_len = @max(new_len, @alignOf(std.c.max_align_t));
+                const new_ptr = c.realloc(memory.ptr, actual_len) orelse return null;
+                assert(alignment.check(@intFromPtr(new_ptr)));
+                return @ptrCast(new_ptr);
+            },
+            .posix_memalign, .manual_align => {
+                // `realloc` would potentially return a new allocation which does not respect
+                // the original alignment, so we can't do anything more.
+                return null;
+            },
+        }
     }
 
     fn free(
         _: *anyopaque,
-        buf: []u8,
+        memory: []u8,
         alignment: Alignment,
         return_address: usize,
     ) void {
-        _ = alignment;
         _ = return_address;
-        alignedFree(buf.ptr);
+        switch (allocStrat(alignment)) {
+            .raw, .posix_memalign => c.free(memory.ptr),
+            .manual_align => c.free(manualAlignHeader(memory.ptr).*),
+        }
     }
 };
 
-/// Supports the full Allocator interface, including alignment, and exploiting
-/// `malloc_usable_size` if available. For an allocator that directly calls
-/// `malloc`/`free`, see `raw_c_allocator`.
-pub const c_allocator: Allocator = .{
-    .ptr = undefined,
-    .vtable = &CAllocator.vtable,
-};
-
-/// Asserts allocations are within `@alignOf(std.c.max_align_t)` and directly
-/// calls `malloc`/`free`. Does not attempt to utilize `malloc_usable_size`.
-/// This allocator is safe to use as the backing allocator with
-/// `ArenaAllocator` for example and is more optimal in such a case than
-/// `c_allocator`.
+/// Asserts that allocations have alignments which `malloc` can satisfy. This means that
+/// the requested alignment is no greater than `@min(@alignOf(std.c.max_align_t), size)`.
+///
+/// This allocator is rarely appropriate to use. In general, prefer `c_allocator`, which
+/// does not have any special requirements of its input, but is still highly efficient for
+/// allocation requests which obey `malloc` alignment rules.
 pub const raw_c_allocator: Allocator = .{
     .ptr = undefined,
     .vtable = &raw_c_allocator_vtable,
@@ -306,13 +359,20 @@ fn rawCAlloc(
 ) ?[*]u8 {
     _ = context;
     _ = return_address;
-    assert(alignment.compare(.lte, .of(std.c.max_align_t)));
-    // Note that this pointer cannot be aligncasted to max_align_t because if
-    // len is < max_align_t then the alignment can be smaller. For example, if
-    // max_align_t is 16, but the user requests 8 bytes, there is no built-in
-    // type in C that is size 8 and has 16 byte alignment, so the alignment may
-    // be 8 bytes rather than 16. Similarly if only 1 byte is requested, malloc
-    // is allowed to return a 1-byte aligned pointer.
+    // `std.c.max_align_t` isn't the whole story, because if `len` is smaller than
+    // every C type with alignment `max_align_t`, the allocation can be less-aligned.
+    // The implementation need only guarantee that any type of length `len` would be
+    // suitably aligned.
+    //
+    // For instance, if `len == 8` and `alignment == .@"16"`, then `malloc` may not
+    // fulfil this request, because there is necessarily no C type with 8-byte size
+    // but 16-byte alignment.
+    //
+    // In theory, the resulting rule here would be target-specific, but in practice,
+    // the smallest type with an alignment of `max_align_t` has the same size (it's
+    // usually `c_longdouble`), so we can just check that `alignment <= len`.
+    assert(alignment.toByteUnits() <= len);
+    assert(Alignment.compare(alignment, .lte, .of(std.c.max_align_t)));
     return @ptrCast(c.malloc(len));
 }
 
@@ -339,8 +399,9 @@ fn rawCRemap(
     return_address: usize,
 ) ?[*]u8 {
     _ = context;
-    _ = alignment;
     _ = return_address;
+    // See `rawCMalloc` for an explanation of this `assert` call.
+    assert(alignment.toByteUnits() <= new_len);
     return @ptrCast(c.realloc(memory.ptr, new_len));
 }
 

lib/std/process/Child.zig

@@ -122,7 +122,7 @@ pub const ResourceUsageStatistics = struct {
     /// if available.
     pub inline fn getMaxRss(rus: ResourceUsageStatistics) ?usize {
         switch (native_os) {
-            .linux => {
+            .dragonfly, .freebsd, .netbsd, .openbsd, .illumos, .linux, .serenity => {
                 if (rus.rusage) |ru| {
                     return @as(usize, @intCast(ru.maxrss)) * 1024;
                 } else {
@@ -149,7 +149,21 @@ pub const ResourceUsageStatistics = struct {
     }
 
     const rusage_init = switch (native_os) {
-        .linux, .driverkit, .ios, .maccatalyst, .macos, .tvos, .visionos, .watchos => @as(?posix.rusage, null),
+        .dragonfly,
+        .freebsd,
+        .netbsd,
+        .openbsd,
+        .illumos,
+        .linux,
+        .serenity,
+        .driverkit,
+        .ios,
+        .maccatalyst,
+        .macos,
+        .tvos,
+        .visionos,
+        .watchos,
+        => @as(?posix.rusage, null),
         .windows => @as(?windows.VM_COUNTERS, null),
         else => {},
     };
@@ -486,7 +500,21 @@ fn waitUnwrappedPosix(self: *ChildProcess) void {
     const res: posix.WaitPidResult = res: {
         if (self.request_resource_usage_statistics) {
             switch (native_os) {
-                .linux, .driverkit, .ios, .maccatalyst, .macos, .tvos, .visionos, .watchos => {
+                .dragonfly,
+                .freebsd,
+                .netbsd,
+                .openbsd,
+                .illumos,
+                .linux,
+                .serenity,
+                .driverkit,
+                .ios,
+                .maccatalyst,
+                .macos,
+                .tvos,
+                .visionos,
+                .watchos,
+                => {
                     var ru: posix.rusage = undefined;
                     const res = posix.wait4(self.id, 0, &ru);
                     self.resource_usage_statistics.rusage = ru;

src/main.zig

@@ -167,14 +167,7 @@ pub fn main() anyerror!void {
     const gpa, const is_debug = gpa: {
         if (build_options.debug_gpa) break :gpa .{ debug_allocator.allocator(), true };
         if (native_os == .wasi) break :gpa .{ std.heap.wasm_allocator, false };
-        if (builtin.link_libc) {
-            // We would prefer to use raw libc allocator here, but cannot use
-            // it if it won't support the alignment we need.
-            if (@alignOf(std.c.max_align_t) < @max(@alignOf(i128), std.atomic.cache_line)) {
-                break :gpa .{ std.heap.c_allocator, false };
-            }
-            break :gpa .{ std.heap.raw_c_allocator, false };
-        }
+        if (builtin.link_libc) break :gpa .{ std.heap.c_allocator, false };
         break :gpa switch (builtin.mode) {
             .Debug, .ReleaseSafe => .{ debug_allocator.allocator(), true },
             .ReleaseFast, .ReleaseSmall => .{ std.heap.smp_allocator, false },

test/src/Cases.zig

@@ -444,6 +444,8 @@ pub const CaseTestOptions = struct {
     test_target_filters: []const []const u8,
     skip_compile_errors: bool,
     skip_non_native: bool,
+    skip_spirv: bool,
+    skip_wasm: bool,
     skip_freebsd: bool,
     skip_netbsd: bool,
     skip_windows: bool,
@@ -472,6 +474,9 @@ pub fn lowerToBuildSteps(
         if (options.skip_non_native and !case.target.query.isNative())
             continue;
 
+        if (options.skip_spirv and case.target.query.cpu_arch != null and case.target.query.cpu_arch.?.isSpirV()) continue;
+        if (options.skip_wasm and case.target.query.cpu_arch != null and case.target.query.cpu_arch.?.isWasm()) continue;
+
         if (options.skip_freebsd and case.target.query.os_tag == .freebsd) continue;
         if (options.skip_netbsd and case.target.query.os_tag == .netbsd) continue;
         if (options.skip_windows and case.target.query.os_tag == .windows) continue;

test/src/Libc.zig

@@ -10,6 +10,7 @@ pub const Options = struct {
     optimize_modes: []const std.builtin.OptimizeMode,
     test_filters: []const []const u8,
     test_target_filters: []const []const u8,
+    skip_wasm: bool,
     max_rss: usize,
 };
 
@@ -41,6 +42,8 @@ pub fn addLibcTestCase(
 }
 
 pub fn addTarget(libc: *const Libc, target: std.Build.ResolvedTarget) void {
+    if (libc.options.skip_wasm and target.query.cpu_arch != null and target.query.cpu_arch.?.isWasm()) return;
+
     if (libc.options.test_target_filters.len > 0) {
         const triple_txt = target.query.zigTriple(libc.b.allocator) catch @panic("OOM");
         for (libc.options.test_target_filters) |filter| {

test/tests.zig

@@ -2246,6 +2246,8 @@ const ModuleTestOptions = struct {
     test_default_only: bool,
     skip_single_threaded: bool,
     skip_non_native: bool,
+    skip_spirv: bool,
+    skip_wasm: bool,
     skip_freebsd: bool,
     skip_netbsd: bool,
     skip_windows: bool,
@@ -2281,6 +2283,9 @@ pub fn addModuleTests(b: *std.Build, options: ModuleTestOptions) *Step {
         if (options.skip_non_native and !test_target.target.isNative())
             continue;
 
+        if (options.skip_spirv and test_target.target.cpu_arch != null and test_target.target.cpu_arch.?.isSpirV()) continue;
+        if (options.skip_wasm and test_target.target.cpu_arch != null and test_target.target.cpu_arch.?.isWasm()) continue;
+
         if (options.skip_freebsd and test_target.target.os_tag == .freebsd) continue;
         if (options.skip_netbsd and test_target.target.os_tag == .netbsd) continue;
         if (options.skip_windows and test_target.target.os_tag == .windows) continue;
@@ -2339,8 +2344,11 @@ fn addOneModuleTest(
     const libc_suffix = if (test_target.link_libc == true) "-libc" else "";
     const model_txt = target.cpu.model.name;
 
-    // wasm32-wasi builds need more RAM, idk why
-    const max_rss = if (target.os.tag == .wasi)
+    // These emulated targets need a lot more RAM for unknown reasons.
+    const max_rss = if (mem.eql(u8, options.name, "std") and
+        (target.cpu.arch == .hexagon or
+            (target.cpu.arch.isRISCV() and !resolved_target.query.isNative()) or
+            target.cpu.arch.isWasm()))
         options.max_rss * 2
     else
         options.max_rss;
@@ -2519,6 +2527,7 @@ pub fn wouldUseLlvm(use_llvm: ?bool, query: std.Target.Query, optimize_mode: Opt
 const CAbiTestOptions = struct {
     test_target_filters: []const []const u8,
     skip_non_native: bool,
+    skip_wasm: bool,
     skip_freebsd: bool,
     skip_netbsd: bool,
     skip_windows: bool,
@@ -2526,6 +2535,7 @@ const CAbiTestOptions = struct {
     skip_linux: bool,
     skip_llvm: bool,
     skip_release: bool,
+    max_rss: usize = 0,
 };
 
 pub fn addCAbiTests(b: *std.Build, options: CAbiTestOptions) *Step {
@@ -2538,6 +2548,9 @@ pub fn addCAbiTests(b: *std.Build, options: CAbiTestOptions) *Step {
 
         for (c_abi_targets) |c_abi_target| {
             if (options.skip_non_native and !c_abi_target.target.isNative()) continue;
+
+            if (options.skip_wasm and c_abi_target.target.cpu_arch != null and c_abi_target.target.cpu_arch.?.isWasm()) continue;
+
             if (options.skip_freebsd and c_abi_target.target.os_tag == .freebsd) continue;
             if (options.skip_netbsd and c_abi_target.target.os_tag == .netbsd) continue;
             if (options.skip_windows and c_abi_target.target.os_tag == .windows) continue;
@@ -2595,6 +2608,7 @@ pub fn addCAbiTests(b: *std.Build, options: CAbiTestOptions) *Step {
                 .root_module = test_mod,
                 .use_llvm = c_abi_target.use_llvm,
                 .use_lld = c_abi_target.use_lld,
+                .max_rss = options.max_rss,
             });
 
             // This test is intentionally trying to check if the external ABI is