c: support +r output constraint

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.

.forgejo/ISSUE_TEMPLATE/bug.yml

@@ -1,6 +1,6 @@
 name: Bug Report
 description: File a bug report
-labels: ["kind/bug"]
+labels: ["bug"]
 body:
   - type: markdown
     attributes:

.forgejo/ISSUE_TEMPLATE/config.yml

@@ -1,7 +1,7 @@
 contact_links:
   - name: Language Proposal
     about: "Please do not submit a proposal to change the language"
-    url: https://ziglang.org/code-of-conduct
+    url: https://ziglang.org/code-of-conduct/#no-language-proposals
   - name: Question
     about: "Please use one of the community spaces instead for questions or general discussions."
     url: https://ziglang.org/community
@@ -10,4 +10,4 @@ contact_links:
     url: https://codeberg.org/ziglang/translate-c
   - name: Copilot and Other LLMs
     about: "Please do not use GitHub Copilot or any other LLM to write an issue."
-    url: https://ziglang.org/code-of-conduct
+    url: https://ziglang.org/code-of-conduct/#strict-no-llm-no-ai-policy

.forgejo/ISSUE_TEMPLATE/error_message.yml

@@ -1,6 +1,6 @@
 name: Error message improvement
 description: Compiler produces an unhelpful or misleading error message.
-labels: ["kind/error message"]
+labels: ["error message"]
 body:
   - type: input
     id: version

.forgejo/workflows/ci.yaml

@@ -152,7 +152,7 @@ jobs:
           fetch-depth: 0
       - name: Build and Test
         run: sh ci/x86_64-linux-debug.sh
-        timeout-minutes: 240
+        timeout-minutes: 180
   x86_64-linux-debug-llvm:
     runs-on: [self-hosted, x86_64-linux]
     steps:
@@ -162,7 +162,7 @@ jobs:
           fetch-depth: 0
       - name: Build and Test
         run: sh ci/x86_64-linux-debug-llvm.sh
-        timeout-minutes: 480
+        timeout-minutes: 360
   x86_64-linux-release:
     runs-on: [self-hosted, x86_64-linux]
     steps:
@@ -172,7 +172,7 @@ jobs:
           fetch-depth: 0
       - name: Build and Test
         run: sh ci/x86_64-linux-release.sh
-        timeout-minutes: 480
+        timeout-minutes: 360
 
   x86_64-windows-debug:
     runs-on: [self-hosted, x86_64-windows]

README.md

@@ -128,7 +128,7 @@ while your Zig build has the option to be a Debug build. It also works
 completely independently from MSVC so you don't need it to be installed.
 
 Determine the URL by
-[looking at the CI script](https://github.com/ziglang/zig/blob/master/ci/x86_64-windows-debug.ps1#L1-L4).
+[looking at the CI script](https://codeberg.org/ziglang/zig/src/branch/master/ci/x86_64-windows-debug.ps1#L1-L4).
 It will look something like this (replace `$VERSION` with the one you see by
 following the above link):
 
@@ -485,16 +485,14 @@ interpret your words.
 ### Find a Contributor Friendly Issue
 
 The issue label
-[Contributor Friendly](https://github.com/ziglang/zig/issues?q=is%3Aissue+is%3Aopen+label%3A%22contributor+friendly%22)
+[Contributor Friendly](https://codeberg.org/ziglang/zig/issues?labels=741726&state=open)
 exists to help you find issues that are **limited in scope and/or
 knowledge of Zig internals.**
 
 Please note that issues labeled
-[Proposal](https://github.com/ziglang/zig/issues?q=is%3Aissue+is%3Aopen+label%3Aproposal)
+[Proposal: Proposed](https://codeberg.org/ziglang/zig/issues?labels=746937&state=open)
-but do not also have the
+are still under consideration, and efforts to implement such a proposal have
-[Accepted](https://github.com/ziglang/zig/issues?q=is%3Aissue+is%3Aopen+label%3Aaccepted)
+a high risk of being wasted. If you are interested in a proposal which is
-label are still under consideration, and efforts to implement such a proposal
-have a high risk of being wasted. If you are interested in a proposal which is
 still under consideration, please express your interest in the issue tracker,
 providing extra insights and considerations that others have not yet expressed.
 The most highly regarded argument in such a discussion is a real world use case.
@@ -665,11 +663,11 @@ based on Clang, but is now based on Aro:
 
 Test coverage as well as bug reports have been moved to this repository:
 
-[ziglang/translate-c](https://github.com/ziglang/translate-c/)
+[ziglang/translate-c](https://codeberg.org/ziglang/translate-c/)
 
 In the future, [@cImport will move to the build system](https://github.com/ziglang/zig/issues/20630),
 but for now, the translate-c logic is copy-pasted from that project into
-[ziglang/zig](https://github.com/ziglang/zig/), powering both `zig translate-c`
+[ziglang/zig](https://codeberg.org/ziglang/zig/), powering both `zig translate-c`
 and `@cImport`.
 
 Please see the readme of the translate-c project for how to contribute. Once an
@@ -777,7 +775,7 @@ If you will be debugging the Zig compiler itself, or if you will be debugging
 any project compiled with Zig's LLVM backend (not recommended with the LLDB
 fork, prefer vanilla LLDB with a version that matches the version of LLVM that
 Zig is using), you can get a better debugging experience by using
-[`lldb_pretty_printers.py`](https://github.com/ziglang/zig/blob/master/tools/lldb_pretty_printers.py).
+[`lldb_pretty_printers.py`](https://codeberg.org/ziglang/zig/src/branch/master/tools/lldb_pretty_printers.py).
 
 Put this line in `~/.lldbinit`:
 

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 = switch (b.graph.host.result.os.tag) {
-        .max_rss = 4000000000,
+            .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 = switch (b.graph.host.result.os.tag) {
-        .max_rss = 6165571174,
+            .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"
@@ -54,11 +55,11 @@ CheckLastExitCode
 
 Write-Output "Main test suite..."
 & "stage3-debug\bin\zig.exe" build test docs `
+  --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"
@@ -54,6 +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 `

lib/libc/glibc/README.md

@@ -39,7 +39,7 @@ v2.2.5.
 
 The file `lib/libc/glibc/abilist` is a Zig-specific binary blob that
 defines the supported glibc versions and the set of symbols each version
-must define.  See https://github.com/ziglang/glibc-abi-tool for the
+must define.  See https://codeberg.org/ziglang/libc-abi-tools for the
 tooling to generate this blob.  The code in `glibc.zig` parses the abilist
 to build version-specific stub libraries on demand.
 

lib/std/Build.zig

@@ -79,7 +79,7 @@ enable_rosetta: bool = false,
 enable_wasmtime: bool = false,
 /// Use system Wine installation to run cross compiled Windows build artifacts.
 enable_wine: bool = false,
-/// After following the steps in https://github.com/ziglang/zig/wiki/Updating-libc#glibc,
+/// After following the steps in https://codeberg.org/ziglang/infra/src/branch/master/libc-update/glibc.md,
 /// this will be the directory $glibc-build-dir/install/glibcs
 /// Given the example of the aarch64 target, this is the directory
 /// that contains the path `aarch64-linux-gnu/lib/ld-linux-aarch64.so.1`.

lib/std/Io.zig

@@ -620,11 +620,6 @@ pub const VTable = struct {
         result: []u8,
         result_alignment: std.mem.Alignment,
     ) void,
-    /// Returns whether the current thread of execution is known to have
-    /// been requested to cancel.
-    ///
-    /// Thread-safe.
-    cancelRequested: *const fn (?*anyopaque) bool,
 
     /// When this function returns, implementation guarantees that `start` has
     /// either already been called, or a unit of concurrency has been assigned
@@ -895,7 +890,7 @@ pub const Timestamp = struct {
     }
 
     pub fn withClock(t: Timestamp, clock: Clock) Clock.Timestamp {
-        return .{ .nanoseconds = t.nanoseconds, .clock = clock };
+        return .{ .raw = t, .clock = clock };
     }
 
     pub fn fromNanoseconds(x: i96) Timestamp {

lib/std/Io/net.zig

@@ -1090,7 +1090,8 @@ pub const Socket = struct {
     }
 
     pub fn sendMany(s: *const Socket, io: Io, messages: []OutgoingMessage, flags: SendFlags) SendError!void {
-        return io.vtable.netSend(io.userdata, s.handle, messages, flags);
+        const err, const n = io.vtable.netSend(io.userdata, s.handle, messages, flags);
+        if (n != messages.len) return err.?;
     }
 
     pub const ReceiveError = error{
@@ -1333,6 +1334,10 @@ pub const Server = struct {
         /// Not enough free memory. This often means that the memory allocation is limited
         /// by the socket buffer limits, not by the system memory.
         SystemResources,
+        /// Either `listen` was never called, or `shutdown` was called (possibly while
+        /// this call was blocking). This allows `shutdown` to be used as a concurrent
+        /// cancellation mechanism.
+        SocketNotListening,
         /// The network subsystem has failed.
         NetworkDown,
         /// No connection is already queued and ready to be accepted, and

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,
 };
 
@@ -10881,6 +10902,23 @@ pub extern "c" fn pthread_create(
     start_routine: *const fn (?*anyopaque) callconv(.c) ?*anyopaque,
     noalias arg: ?*anyopaque,
 ) E;
+pub const pthread_cancelstate = switch (native_os) {
+    .ios, .maccatalyst, .macos, .tvos, .visionos, .watchos => enum(c_int) {
+        ENABLE = 1,
+        DISABLE = 0,
+    },
+    .linux => if (native_abi.isMusl()) enum(c_int) {
+        ENABLE = 0,
+        DISABLE = 1,
+        MASKED = 2,
+    } else if (native_abi.isGnu()) enum(c_int) {
+        ENABLE = 0,
+        DISABLE = 1,
+    },
+    else => void,
+};
+pub extern "c" fn pthread_setcancelstate(pthread_cancelstate, ?*pthread_cancelstate) E;
+pub extern "c" fn pthread_cancel(pthread_t) E;
 pub extern "c" fn pthread_attr_init(attr: *pthread_attr_t) E;
 pub extern "c" fn pthread_attr_setstack(attr: *pthread_attr_t, stackaddr: *anyopaque, stacksize: usize) E;
 pub extern "c" fn pthread_attr_setstacksize(attr: *pthread_attr_t, stacksize: usize) E;

lib/std/crypto.zig

@@ -116,6 +116,7 @@ pub const dh = struct {
 
 /// Key Encapsulation Mechanisms.
 pub const kem = struct {
+    pub const hybrid = @import("crypto/hybrid_kem.zig");
     pub const kyber_d00 = @import("crypto/ml_kem.zig").d00;
     pub const ml_kem = @import("crypto/ml_kem.zig").nist;
 };

lib/std/crypto/aes.zig

@@ -108,6 +108,36 @@ test "expand 128-bit key" {
     }
 }
 
+test "invMixColumns" {
+    const key = [_]u8{ 0x2b, 0x7e, 0x15, 0x16, 0x28, 0xae, 0xd2, 0xa6, 0xab, 0xf7, 0x15, 0x88, 0x09, 0xcf, 0x4f, 0x3c };
+    const enc_ctx = Aes128.initEnc(key);
+    const dec_ctx = Aes128.initDec(key);
+
+    for (1..10) |i| {
+        const enc_rk = enc_ctx.key_schedule.round_keys[10 - i];
+        const dec_rk = dec_ctx.key_schedule.round_keys[i];
+        const computed = enc_rk.invMixColumns();
+        try testing.expectEqualSlices(u8, &dec_rk.toBytes(), &computed.toBytes());
+    }
+}
+
+test "BlockVec invMixColumns" {
+    const input = [_]u8{
+        0x5f, 0x57, 0xf7, 0x1d, 0x72, 0xf5, 0xbe, 0xb9, 0x64, 0xbc, 0x3b, 0xf9, 0x15, 0x92, 0x29, 0x1a,
+        0x2b, 0x7e, 0x15, 0x16, 0x28, 0xae, 0xd2, 0xa6, 0xab, 0xf7, 0x15, 0x88, 0x09, 0xcf, 0x4f, 0x3c,
+    };
+
+    const vec2 = BlockVec(2).fromBytes(&input);
+    const result_vec = vec2.invMixColumns();
+    const result_bytes = result_vec.toBytes();
+
+    for (0..2) |i| {
+        const block = Block.fromBytes(input[i * 16 ..][0..16]);
+        const expected = block.invMixColumns().toBytes();
+        try testing.expectEqualSlices(u8, &expected, result_bytes[i * 16 ..][0..16]);
+    }
+}
+
 test "expand 256-bit key" {
     const key = [_]u8{
         0x60, 0x3d, 0xeb, 0x10,

lib/std/crypto/aes/aesni.zig

@@ -96,6 +96,17 @@ pub const Block = struct {
         return Block{ .repr = block1.repr | block2.repr };
     }
 
+    /// Apply the inverse MixColumns operation to a block.
+    pub fn invMixColumns(block: Block) Block {
+        return Block{
+            .repr = asm (
+                \\ vaesimc %[in], %[out]
+                : [out] "=x" (-> Repr),
+                : [in] "x" (block.repr),
+            ),
+        };
+    }
+
     /// Perform operations on multiple blocks in parallel.
     pub const parallel = struct {
         const cpu = std.Target.x86.cpu;
@@ -308,6 +319,17 @@ pub fn BlockVec(comptime blocks_count: comptime_int) type {
             }
             return out;
         }
+
+        /// Apply the inverse MixColumns operation to each block in the vector.
+        pub fn invMixColumns(block_vec: Self) Self {
+            var out_bytes: [blocks_count * 16]u8 = undefined;
+            const in_bytes = block_vec.toBytes();
+            inline for (0..blocks_count) |i| {
+                const block = Block.fromBytes(in_bytes[i * 16 ..][0..16]);
+                out_bytes[i * 16 ..][0..16].* = block.invMixColumns().toBytes();
+            }
+            return fromBytes(&out_bytes);
+        }
     };
 }
 

lib/std/crypto/aes/armcrypto.zig

@@ -99,6 +99,17 @@ pub const Block = struct {
         return Block{ .repr = block1.repr | block2.repr };
     }
 
+    /// Apply the inverse MixColumns operation to a block.
+    pub fn invMixColumns(block: Block) Block {
+        return Block{
+            .repr = asm (
+                \\ aesimc %[out].16b, %[in].16b
+                : [out] "=x" (-> Repr),
+                : [in] "x" (block.repr),
+            ),
+        };
+    }
+
     /// Perform operations on multiple blocks in parallel.
     pub const parallel = struct {
         /// The recommended number of AES encryption/decryption to perform in parallel for the chosen implementation.
@@ -275,6 +286,15 @@ pub fn BlockVec(comptime blocks_count: comptime_int) type {
             }
             return out;
         }
+
+        /// Apply the inverse MixColumns operation to each block in the vector.
+        pub fn invMixColumns(block_vec: Self) Self {
+            var out: Self = undefined;
+            inline for (0..native_words) |i| {
+                out.repr[i] = block_vec.repr[i].invMixColumns();
+            }
+            return out;
+        }
     };
 }
 

lib/std/crypto/aes/soft.zig

@@ -265,6 +265,26 @@ pub const Block = struct {
         return Block{ .repr = x };
     }
 
+    /// Apply the inverse MixColumns operation to a block.
+    pub fn invMixColumns(block: Block) Block {
+        var out: Repr = undefined;
+        inline for (0..4) |i| {
+            const col = block.repr[i];
+            const b0: u8 = @truncate(col);
+            const b1: u8 = @truncate(col >> 8);
+            const b2: u8 = @truncate(col >> 16);
+            const b3: u8 = @truncate(col >> 24);
+
+            const r0 = mul(0x0e, b0) ^ mul(0x0b, b1) ^ mul(0x0d, b2) ^ mul(0x09, b3);
+            const r1 = mul(0x09, b0) ^ mul(0x0e, b1) ^ mul(0x0b, b2) ^ mul(0x0d, b3);
+            const r2 = mul(0x0d, b0) ^ mul(0x09, b1) ^ mul(0x0e, b2) ^ mul(0x0b, b3);
+            const r3 = mul(0x0b, b0) ^ mul(0x0d, b1) ^ mul(0x09, b2) ^ mul(0x0e, b3);
+
+            out[i] = @as(u32, r0) | (@as(u32, r1) << 8) | (@as(u32, r2) << 16) | (@as(u32, r3) << 24);
+        }
+        return Block{ .repr = out };
+    }
+
     /// Perform operations on multiple blocks in parallel.
     pub const parallel = struct {
         /// The recommended number of AES encryption/decryption to perform in parallel for the chosen implementation.
@@ -441,6 +461,15 @@ pub fn BlockVec(comptime blocks_count: comptime_int) type {
             }
             return out;
         }
+
+        /// Apply the inverse MixColumns operation to each block in the vector.
+        pub fn invMixColumns(block_vec: Self) Self {
+            var out: Self = undefined;
+            for (0..native_words) |i| {
+                out.repr[i] = block_vec.repr[i].invMixColumns();
+            }
+            return out;
+        }
     };
 }
 

lib/std/crypto/argon2.zig

@@ -7,12 +7,11 @@ const builtin = @import("builtin");
 
 const blake2 = crypto.hash.blake2;
 const crypto = std.crypto;
+const Io = std.Io;
 const math = std.math;
 const mem = std.mem;
 const phc_format = pwhash.phc_format;
 const pwhash = crypto.pwhash;
-
-const Thread = std.Thread;
 const Blake2b512 = blake2.Blake2b512;
 const Blocks = std.array_list.AlignedManaged([block_length]u64, .@"16");
 const H0 = [Blake2b512.digest_length + 8]u8;
@@ -204,20 +203,20 @@ fn initBlocks(
 }
 
 fn processBlocks(
-    allocator: mem.Allocator,
     blocks: *Blocks,
     time: u32,
     memory: u32,
     threads: u24,
     mode: Mode,
-) KdfError!void {
+    io: Io,
+) void {
     const lanes = memory / threads;
     const segments = lanes / sync_points;
 
     if (builtin.single_threaded or threads == 1) {
         processBlocksSt(blocks, time, memory, threads, mode, lanes, segments);
     } else {
-        try processBlocksMt(allocator, blocks, time, memory, threads, mode, lanes, segments);
+        processBlocksMt(blocks, time, memory, threads, mode, lanes, segments, io);
     }
 }
 
@@ -243,7 +242,6 @@ fn processBlocksSt(
 }
 
 fn processBlocksMt(
-    allocator: mem.Allocator,
     blocks: *Blocks,
     time: u32,
     memory: u32,
@@ -251,26 +249,20 @@ fn processBlocksMt(
     mode: Mode,
     lanes: u32,
     segments: u32,
-) KdfError!void {
+    io: Io,
-    var threads_list = try std.array_list.Managed(Thread).initCapacity(allocator, threads);
+) void {
-    defer threads_list.deinit();
-
     var n: u32 = 0;
     while (n < time) : (n += 1) {
         var slice: u32 = 0;
         while (slice < sync_points) : (slice += 1) {
+            var group: Io.Group = .init;
             var lane: u24 = 0;
             while (lane < threads) : (lane += 1) {
-                const thread = try Thread.spawn(.{}, processSegment, .{
+                group.async(io, processSegment, .{
                     blocks, time, memory, threads, mode, lanes, segments, n, slice, lane,
                 });
-                threads_list.appendAssumeCapacity(thread);
             }
-            lane = 0;
+            group.wait(io);
-            while (lane < threads) : (lane += 1) {
-                threads_list.items[lane].join();
-            }
-            threads_list.clearRetainingCapacity();
         }
     }
 }
@@ -489,6 +481,7 @@ pub fn kdf(
     salt: []const u8,
     params: Params,
     mode: Mode,
+    io: Io,
 ) KdfError!void {
     if (derived_key.len < 4) return KdfError.WeakParameters;
     if (derived_key.len > max_int) return KdfError.OutputTooLong;
@@ -510,7 +503,7 @@ pub fn kdf(
     blocks.appendNTimesAssumeCapacity(@splat(0), memory);
 
     initBlocks(&blocks, &h0, memory, params.p);
-    try processBlocks(allocator, &blocks, params.t, memory, params.p, mode);
+    processBlocks(&blocks, params.t, memory, params.p, mode, io);
     finalize(&blocks, memory, params.p, derived_key);
 }
 
@@ -533,6 +526,7 @@ const PhcFormatHasher = struct {
         params: Params,
         mode: Mode,
         buf: []u8,
+        io: Io,
     ) HasherError![]const u8 {
         if (params.secret != null or params.ad != null) return HasherError.InvalidEncoding;
 
@@ -540,7 +534,7 @@ const PhcFormatHasher = struct {
         crypto.random.bytes(&salt);
 
         var hash: [default_hash_len]u8 = undefined;
-        try kdf(allocator, &hash, password, &salt, params, mode);
+        try kdf(allocator, &hash, password, &salt, params, mode, io);
 
         return phc_format.serialize(HashResult{
             .alg_id = @tagName(mode),
@@ -557,6 +551,7 @@ const PhcFormatHasher = struct {
         allocator: mem.Allocator,
         str: []const u8,
         password: []const u8,
+        io: Io,
     ) HasherError!void {
         const hash_result = try phc_format.deserialize(HashResult, str);
 
@@ -572,7 +567,7 @@ const PhcFormatHasher = struct {
         if (expected_hash.len > hash_buf.len) return HasherError.InvalidEncoding;
         const hash = hash_buf[0..expected_hash.len];
 
-        try kdf(allocator, hash, password, hash_result.salt.constSlice(), params, mode);
+        try kdf(allocator, hash, password, hash_result.salt.constSlice(), params, mode, io);
         if (!mem.eql(u8, hash, expected_hash)) return HasherError.PasswordVerificationFailed;
     }
 };
@@ -595,6 +590,7 @@ pub fn strHash(
     password: []const u8,
     options: HashOptions,
     out: []u8,
+    io: Io,
 ) Error![]const u8 {
     const allocator = options.allocator orelse return Error.AllocatorRequired;
     switch (options.encoding) {
@@ -604,6 +600,7 @@ pub fn strHash(
             options.params,
             options.mode,
             out,
+            io,
         ),
         .crypt => return Error.InvalidEncoding,
     }
@@ -621,9 +618,10 @@ pub fn strVerify(
     str: []const u8,
     password: []const u8,
     options: VerifyOptions,
+    io: Io,
 ) Error!void {
     const allocator = options.allocator orelse return Error.AllocatorRequired;
-    return PhcFormatHasher.verify(allocator, str, password);
+    return PhcFormatHasher.verify(allocator, str, password, io);
 }
 
 test "argon2d" {
@@ -640,6 +638,7 @@ test "argon2d" {
         &salt,
         .{ .t = 3, .m = 32, .p = 4, .secret = &secret, .ad = &ad },
         .argon2d,
+        std.testing.io,
     );
 
     const want = [_]u8{
@@ -665,6 +664,7 @@ test "argon2i" {
         &salt,
         .{ .t = 3, .m = 32, .p = 4, .secret = &secret, .ad = &ad },
         .argon2i,
+        std.testing.io,
     );
 
     const want = [_]u8{
@@ -690,6 +690,7 @@ test "argon2id" {
         &salt,
         .{ .t = 3, .m = 32, .p = 4, .secret = &secret, .ad = &ad },
         .argon2id,
+        std.testing.io,
     );
 
     const want = [_]u8{
@@ -800,44 +801,44 @@ test "kdf" {
         .{
             .mode = .argon2i,
             .time = 4,
-            .memory = 4096,
+            .memory = 256,
             .threads = 4,
-            .hash = "a11f7b7f3f93f02ad4bddb59ab62d121e278369288a0d0e7",
+            .hash = "f7dbbacbf16999e3700817a7e06f65a8db2e9fa9504ede4c",
         },
         .{
             .mode = .argon2d,
             .time = 4,
-            .memory = 4096,
+            .memory = 256,
             .threads = 4,
-            .hash = "935598181aa8dc2b720914aa6435ac8d3e3a4210c5b0fb2d",
+            .hash = "ea2970501cf49faa5ba1d2e6370204e9b57ca90a8fea937b",
         },
         .{
             .mode = .argon2id,
             .time = 4,
-            .memory = 4096,
+            .memory = 256,
             .threads = 4,
-            .hash = "145db9733a9f4ee43edf33c509be96b934d505a4efb33c5a",
+            .hash = "fbd40d5a8cb92f88c20bda4b3cdb1f9d5af1efa937032410",
         },
         .{
             .mode = .argon2i,
             .time = 4,
-            .memory = 1024,
+            .memory = 256,
             .threads = 8,
-            .hash = "0cdd3956aa35e6b475a7b0c63488822f774f15b43f6e6e17",
+            .hash = "15d3c398364e53f68fd12d19baf3f21432d964254fe27467",
         },
         .{
             .mode = .argon2d,
             .time = 4,
-            .memory = 1024,
+            .memory = 256,
             .threads = 8,
-            .hash = "83604fc2ad0589b9d055578f4d3cc55bc616df3578a896e9",
+            .hash = "23c9adc06f06e21e4612c1466a1be02627690932b02c0df0",
         },
         .{
             .mode = .argon2id,
             .time = 4,
-            .memory = 1024,
+            .memory = 256,
             .threads = 8,
-            .hash = "8dafa8e004f8ea96bf7c0f93eecf67a6047476143d15577f",
+            .hash = "f22802f8ca47be93f9954e4ce20c1e944e938fbd4a125d9d",
         },
         .{
             .mode = .argon2i,
@@ -863,23 +864,23 @@ test "kdf" {
         .{
             .mode = .argon2i,
             .time = 3,
-            .memory = 1024,
+            .memory = 256,
             .threads = 6,
-            .hash = "d236b29c2b2a09babee842b0dec6aa1e83ccbdea8023dced",
+            .hash = "ebc8f91964abd8ceab49a12963b0a9e57d635bfa2aad2884",
         },
         .{
             .mode = .argon2d,
             .time = 3,
-            .memory = 1024,
+            .memory = 256,
             .threads = 6,
-            .hash = "a3351b0319a53229152023d9206902f4ef59661cdca89481",
+            .hash = "1dd7202fd68da6675f769f4034b7a1db30d8785331954117",
         },
         .{
             .mode = .argon2id,
             .time = 3,
-            .memory = 1024,
+            .memory = 256,
             .threads = 6,
-            .hash = "1640b932f4b60e272f5d2207b9a9c626ffa1bd88d2349016",
+            .hash = "424436b6ee22a66b04b9d0cf78f190305c5c166bae8baa09",
         },
     };
     for (test_vectors) |v| {
@@ -894,6 +895,7 @@ test "kdf" {
             salt,
             .{ .t = v.time, .m = v.memory, .p = v.threads },
             v.mode,
+            std.testing.io,
         );
 
         try std.testing.expectEqualSlices(u8, &dk, &want);
@@ -903,6 +905,7 @@ test "kdf" {
 test "phc format hasher" {
     const allocator = std.testing.allocator;
     const password = "testpass";
+    const io = std.testing.io;
 
     var buf: [128]u8 = undefined;
     const hash = try PhcFormatHasher.create(
@@ -911,25 +914,29 @@ test "phc format hasher" {
         .{ .t = 3, .m = 32, .p = 4 },
         .argon2id,
         &buf,
+        io,
     );
-    try PhcFormatHasher.verify(allocator, hash, password);
+    try PhcFormatHasher.verify(allocator, hash, password, io);
 }
 
 test "password hash and password verify" {
     const allocator = std.testing.allocator;
     const password = "testpass";
+    const io = std.testing.io;
 
     var buf: [128]u8 = undefined;
     const hash = try strHash(
         password,
         .{ .allocator = allocator, .params = .{ .t = 3, .m = 32, .p = 4 } },
         &buf,
+        io,
     );
-    try strVerify(hash, password, .{ .allocator = allocator });
+    try strVerify(hash, password, .{ .allocator = allocator }, io);
 }
 
 test "kdf derived key length" {
     const allocator = std.testing.allocator;
+    const io = std.testing.io;
 
     const password = "testpass";
     const salt = "saltsalt";
@@ -937,11 +944,11 @@ test "kdf derived key length" {
     const mode = Mode.argon2id;
 
     var dk1: [11]u8 = undefined;
-    try kdf(allocator, &dk1, password, salt, params, mode);
+    try kdf(allocator, &dk1, password, salt, params, mode, io);
 
     var dk2: [77]u8 = undefined;
-    try kdf(allocator, &dk2, password, salt, params, mode);
+    try kdf(allocator, &dk2, password, salt, params, mode, io);
 
     var dk3: [111]u8 = undefined;
-    try kdf(allocator, &dk3, password, salt, params, mode);
+    try kdf(allocator, &dk3, password, salt, params, mode, io);
 }

lib/std/crypto/benchmark.zig

@@ -450,6 +450,7 @@ fn benchmarkPwhash(
     comptime ty: anytype,
     comptime params: *const anyopaque,
     comptime count: comptime_int,
+    io: std.Io,
 ) !f64 {
     const password = "testpass" ** 2;
     const opts = ty.HashOptions{
@@ -459,12 +460,20 @@ fn benchmarkPwhash(
     };
     var buf: [256]u8 = undefined;
 
+    const strHash = ty.strHash;
+    const strHashFnInfo = @typeInfo(@TypeOf(strHash)).@"fn";
+    const needs_io = strHashFnInfo.params.len == 4;
+
     var timer = try Timer.start();
     const start = timer.lap();
     {
         var i: usize = 0;
         while (i < count) : (i += 1) {
-            _ = try ty.strHash(password, opts, &buf);
+            if (needs_io) {
+                _ = try strHash(password, opts, &buf, io);
+            } else {
+                _ = try strHash(password, opts, &buf);
+            }
             mem.doNotOptimizeAway(&buf);
         }
     }
@@ -623,7 +632,7 @@ pub fn main() !void {
 
     inline for (pwhashes) |H| {
         if (filter == null or std.mem.indexOf(u8, H.name, filter.?) != null) {
-            const throughput = try benchmarkPwhash(arena_allocator, H.ty, H.params, mode(64));
+            const throughput = try benchmarkPwhash(arena_allocator, H.ty, H.params, mode(64), io);
             try stdout.print("{s:>17}: {d:10.3} s/ops\n", .{ H.name, throughput });
             try stdout.flush();
         }

lib/std/debug.zig

@@ -1604,6 +1604,13 @@ pub fn dumpStackPointerAddr(prefix: []const u8) void {
 
 test "manage resources correctly" {
     if (SelfInfo == void) return error.SkipZigTest;
+    if (builtin.zig_backend == .stage2_c) {
+        // The C backend emits an extremely large C source file, meaning it has a huge
+        // amount of debug information. Parsing this debug information makes this test
+        // take too long to be worth running.
+        return error.SkipZigTest;
+    }
+
     const S = struct {
         noinline fn showMyTrace() usize {
             return @returnAddress();

lib/std/hash/crc.zig

@@ -1,5 +1,6 @@
 //! This file is auto-generated by tools/update_crc_catalog.zig.
 
+const builtin = @import("builtin");
 const impl = @import("crc/impl.zig");
 
 pub const Crc = impl.Crc;
@@ -13,6 +14,17 @@ test {
     _ = @import("crc/test.zig");
 }
 
+pub const Crc32Iscsi = switch (builtin.cpu.hasAll(.x86, &.{ .@"64bit", .crc32 })) {
+    true => @import("crc/crc32c.zig").Wrapper,
+    else => Crc(u32, .{
+        .polynomial = 0x1edc6f41,
+        .initial = 0xffffffff,
+        .reflect_input = true,
+        .reflect_output = true,
+        .xor_output = 0xffffffff,
+    }),
+};
+
 pub const Crc3Gsm = Crc(u3, .{
     .polynomial = 0x3,
     .initial = 0x0,
@@ -797,14 +809,6 @@ pub const Crc32Cksum = Crc(u32, .{
     .xor_output = 0xffffffff,
 });
 
-pub const Crc32Iscsi = Crc(u32, .{
-    .polynomial = 0x1edc6f41,
-    .initial = 0xffffffff,
-    .reflect_input = true,
-    .reflect_output = true,
-    .xor_output = 0xffffffff,
-});
-
 pub const Crc32IsoHdlc = Crc(u32, .{
     .polynomial = 0x04c11db7,
     .initial = 0xffffffff,

lib/std/hash/crc/crc32c.zig

@@ -0,0 +1,239 @@
+//! Implements CRC-32C (Castagnoli) using the SSE4.2 Intel CRC32 instruction.
+//!
+//! A couple useful links for understanding the approach taken here:
+//! - https://github.com/madler/brotli/blob/1d428d3a9baade233ebc3ac108293256bcb813d1/crc32c.c
+//! - https://github.com/madler/zlib/blob/5a82f71ed1dfc0bec044d9702463dbdf84ea3b71/crc32.c
+//! - http://www.ross.net/crc/download/crc_v3.txt
+
+// Reflected CRC-32C polynomial in binary form.
+const POLY = 0x82f63b78;
+
+const LONG = 8192;
+const SHORT = 256;
+const long_lookup_table = genTable(LONG);
+const short_lookup_table = genTable(SHORT);
+
+/// Generates the lookup table for efficiently combining CRCs over a block of a given length `length`.
+/// This works by building an operator that advances the CRC state as if `length` zero-bytes were appended.
+/// We pre-compute 4 tables of 256 entries each (one per byte offset).
+///
+///
+/// The idea behind this table is quite interesting. The CRC state is equivalent to the
+/// remainder of dividing the message polynomial (over GF(2)) by the CRC polynomial.
+///
+/// Advancing the CRC register by `k` zero bits is equivalent to multiplying the current
+/// CRC state by `x^k` modulo the CRC polynomial. This operation can be represented
+/// as a linear transformation in GF(2), i.e, a matrix.
+///
+/// We build up this matrix via repeated squaring:
+/// - odd represents the operator for 1 zero bit (i.e, multiplication by `x^1 mod POLY`)
+/// - even represents the operator for 2 zero bits (`x^2 mod POLY`)
+/// - squaring again gives `x^4 mod POLY`, and so on until we get to the right size.
+///
+/// By squaring the shifting `len`, we build the operator for `x^l mod POLY`.
+fn genTable(length: usize) [4][256]u32 {
+    @setEvalBranchQuota(250000);
+
+    var even: [32]u32 = undefined;
+    zeroes: {
+        var odd: [32]u32 = undefined;
+
+        // Initialize our `odd` array with the operator for a single zero bit:
+        // - odd[0] is the polynomial itself (acts on the MSB).
+        // - odd[1..32] represent shifting a single bit through 31 positions.
+        odd[0] = POLY;
+        var row: u32 = 1;
+        for (1..32) |n| {
+            odd[n] = row;
+            row <<= 1;
+        }
+
+        // even = odd squared: even represents `x^2 mod POLY`.
+        square(&even, &odd);
+        // odd = even squared: odd now represents `x^4 mod POLY`.
+        square(&odd, &even);
+
+        // Continue squaring to double the number of zeroes encoded each time:
+        //
+        // At each point in the process:
+        // - square(even, odd): even gets the operator for twice the current length.
+        // - square(odd, even): odd gets the operator for 4 times the original length.
+        var len = length;
+        while (true) {
+            square(&even, &odd);
+            len >>= 1;
+            if (len == 0) break :zeroes;
+            square(&odd, &even);
+            len >>= 1;
+            if (len == 0) break;
+        }
+
+        @memcpy(&even, &odd);
+    }
+
+    var zeroes: [4][256]u32 = undefined;
+    for (0..256) |n| {
+        zeroes[0][n] = times(&even, n);
+        zeroes[1][n] = times(&even, n << 8);
+        zeroes[2][n] = times(&even, n << 16);
+        zeroes[3][n] = times(&even, n << 24);
+    }
+    return zeroes;
+}
+
+/// Computes `mat * vec` over `GF(2)`, where `mat` is a 32x32 binary matrix and `vec`
+/// is a 32-bit vector. This somewhat "simulates" how bits propagate through the CRC register
+/// during shifting.
+///
+/// - In GF(2) (aka a field where the only values are 0 and 1, aka binary), multiplication is
+/// an `AND`, and addition is `XOR`.
+/// - This dot product determines how each bit in the input vector "contributes" to
+/// the final CRC state, by XORing (adding) rows of the matrix where `vec` has 1s.
+fn times(mat: *const [32]u32, vec: u32) u32 {
+    var sum: u32 = 0;
+    var v = vec;
+    var i: u32 = 0;
+    while (v != 0) {
+        if (v & 1 != 0) sum ^= mat[i];
+        v >>= 1;
+        i += 1;
+    }
+    return sum;
+}
+
+/// Computes the square of a matrix in GF(2), i.e `dst = dst x src`.
+///
+/// This produces the operator for doubling the number of zeroes:
+/// if `src` represents advancing the CRC by `k` zeroes, then `dest` will
+/// represent advancing by 2k zeroes.
+///
+/// Since polynomial multiplication mod POLY is linear, `mat(mat(x)) = mat^2(x)`
+/// gives the effect of two sequential applications of the operator.
+fn square(dst: *[32]u32, src: *const [32]u32) void {
+    for (dst, src) |*d, s| {
+        d.* = times(src, s);
+    }
+}
+
+fn shift(table: *const [4][256]u32, crc: u32) u32 {
+    return table[0][crc & 0xFF] ^ table[1][(crc >> 8) & 0xFF] ^ table[2][(crc >> 16) & 0xFF] ^ table[3][crc >> 24];
+}
+
+fn crc32(crc: u32, input: []const u8) u32 {
+    var crc0: u64 = ~crc;
+
+    // Compute the CRC for up to seven leading bytes to bring the
+    // `next` pointer to an eight-byte boundary.
+    var next = input;
+    while (next.len > 0 and @intFromPtr(next.ptr) & 7 != 0) {
+        asm volatile ("crc32b %[out], %[in]"
+            : [in] "+r" (crc0),
+            : [out] "rm" (next[0]),
+        );
+        next = next[1..];
+    }
+
+    // Compute the CRC on sets of LONG * 3 bytes, executing three independent
+    // CRC instructions, each on LONG bytes. This is an optimization for
+    // targets where the CRC instruction has a throughput of one CRC per
+    // cycle, but a latency of three cycles.
+    while (next.len >= LONG * 3) {
+        var crc1: u64 = 0;
+        var crc2: u64 = 0;
+
+        const start = next.len;
+        while (true) {
+            // Safe @alignCast(), since we've aligned the pointer to 8 bytes before this loop.
+            const long: [*]const u64 = @ptrCast(@alignCast(next));
+            asm volatile (
+                \\crc32q %[out0], %[in0]
+                \\crc32q %[out1], %[in1]
+                \\crc32q %[out2], %[in2]
+                : [in0] "+r" (crc0),
+                  [in1] "+r" (crc1),
+                  [in2] "+r" (crc2),
+                : [out0] "rm" (long[0 * LONG / 8]),
+                  [out1] "rm" (long[1 * LONG / 8]),
+                  [out2] "rm" (long[2 * LONG / 8]),
+            );
+            next = next[8..];
+            if (next.len <= start - LONG) break;
+        }
+
+        crc0 = shift(&long_lookup_table, @truncate(crc0)) ^ crc1;
+        crc0 = shift(&long_lookup_table, @truncate(crc0)) ^ crc2;
+        next = next[LONG * 2 ..];
+    }
+
+    // Same thing as above, but for smaller chunks of SHORT bytes.
+    while (next.len >= SHORT * 3) {
+        var crc1: u64 = 0;
+        var crc2: u64 = 0;
+
+        const start = next.len;
+        while (true) {
+            const long: [*]const u64 = @ptrCast(@alignCast(next));
+            asm volatile (
+                \\crc32q %[out0], %[in0]
+                \\crc32q %[out1], %[in1]
+                \\crc32q %[out2], %[in2]
+                : [in0] "+r" (crc0),
+                  [in1] "+r" (crc1),
+                  [in2] "+r" (crc2),
+                : [out0] "rm" (long[0 * SHORT / 8]),
+                  [out1] "rm" (long[1 * SHORT / 8]),
+                  [out2] "rm" (long[2 * SHORT / 8]),
+            );
+            next = next[8..];
+            if (next.len <= start - SHORT) break;
+        }
+
+        crc0 = shift(&short_lookup_table, @truncate(crc0)) ^ crc1;
+        crc0 = shift(&short_lookup_table, @truncate(crc0)) ^ crc2;
+        next = next[SHORT * 2 ..];
+    }
+
+    // Compute via 8-byte chunks, until we're left with less than 8 bytes.
+    while (next.len >= 8) {
+        const long: [*]const u64 = @ptrCast(@alignCast(next));
+        asm volatile ("crc32q %[out], %[in]"
+            : [in] "+r" (crc0),
+            : [out] "rm" (long[0]),
+        );
+        next = next[8..];
+    }
+
+    // Finish the last bytes with just single instructions.
+    while (next.len > 0) {
+        asm volatile ("crc32b %[out], %[in]"
+            : [in] "+r" (crc0),
+            : [out] "rm" (next[0]),
+        );
+        next = next[1..];
+    }
+
+    return @truncate(~crc0);
+}
+
+// Wrapper around the accelerated implementation to match the one in impl.zig.
+pub const Wrapper = struct {
+    crc: u32,
+
+    pub fn init() Wrapper {
+        return .{ .crc = 0 };
+    }
+
+    pub fn update(w: *Wrapper, bytes: []const u8) void {
+        w.crc = crc32(w.crc, bytes);
+    }
+
+    pub fn final(w: Wrapper) u32 {
+        return w.crc;
+    }
+
+    pub fn hash(bytes: []const u8) u32 {
+        var c = init();
+        c.update(bytes);
+        return c.final();
+    }
+};

lib/std/hash/crc/impl.zig

@@ -23,12 +23,7 @@ pub fn Crc(comptime W: type, comptime algorithm: Algorithm(W)) type {
         const I = if (@bitSizeOf(W) < 8) u8 else W;
         const lookup_table = blk: {
             @setEvalBranchQuota(2500);
-
+            const poly = reflect(algorithm.polynomial);
-            const poly = if (algorithm.reflect_input)
-                @bitReverse(@as(I, algorithm.polynomial)) >> (@bitSizeOf(I) - @bitSizeOf(W))
-            else
-                @as(I, algorithm.polynomial) << (@bitSizeOf(I) - @bitSizeOf(W));
-
             var table: [256]I = undefined;
             for (&table, 0..) |*e, i| {
                 var crc: I = i;
@@ -52,15 +47,13 @@ pub fn Crc(comptime W: type, comptime algorithm: Algorithm(W)) type {
         crc: I,
 
         pub fn init() Self {
-            const initial = if (algorithm.reflect_input)
+            const initial = reflect(algorithm.initial);
-                @bitReverse(@as(I, algorithm.initial)) >> (@bitSizeOf(I) - @bitSizeOf(W))
+            return .{ .crc = initial };
-            else
-                @as(I, algorithm.initial) << (@bitSizeOf(I) - @bitSizeOf(W));
-            return Self{ .crc = initial };
         }
 
         inline fn tableEntry(index: I) I {
-            return lookup_table[@as(u8, @intCast(index & 0xFF))];
+            const short: u8 = @truncate(index);
+            return lookup_table[short];
         }
 
         pub fn update(self: *Self, bytes: []const u8) void {
@@ -90,7 +83,7 @@ pub fn Crc(comptime W: type, comptime algorithm: Algorithm(W)) type {
             if (!algorithm.reflect_output) {
                 c >>= @bitSizeOf(I) - @bitSizeOf(W);
             }
-            return @as(W, @intCast(c ^ algorithm.xor_output));
+            return @intCast(c ^ algorithm.xor_output);
         }
 
         pub fn hash(bytes: []const u8) W {
@@ -98,5 +91,13 @@ pub fn Crc(comptime W: type, comptime algorithm: Algorithm(W)) type {
             c.update(bytes);
             return c.final();
         }
+
+        fn reflect(x: I) I {
+            const offset = @bitSizeOf(I) - @bitSizeOf(W);
+            if (algorithm.reflect_input)
+                return @bitReverse(x) >> offset
+            else
+                return x << offset;
+        }
     };
 }

lib/std/hash/crc/test.zig

@@ -26,6 +26,17 @@ test "crc32 koopman regression" {
     try testing.expectEqual(crc32.hash("abc"), 0xba2322ac);
 }
 
+test "CRC-32/ISCSI" {
+    const Crc32Iscsi = crc.Crc32Iscsi;
+
+    try testing.expectEqual(@as(u32, 0xe3069283), Crc32Iscsi.hash("123456789"));
+
+    var c = Crc32Iscsi.init();
+    c.update("1234");
+    c.update("56789");
+    try testing.expectEqual(@as(u32, 0xe3069283), c.final());
+}
+
 test "CRC-3/GSM" {
     const Crc3Gsm = crc.Crc3Gsm;
 
@@ -1104,17 +1115,6 @@ test "CRC-32/CKSUM" {
     try testing.expectEqual(@as(u32, 0x765e7680), c.final());
 }
 
-test "CRC-32/ISCSI" {
-    const Crc32Iscsi = crc.Crc32Iscsi;
-
-    try testing.expectEqual(@as(u32, 0xe3069283), Crc32Iscsi.hash("123456789"));
-
-    var c = Crc32Iscsi.init();
-    c.update("1234");
-    c.update("56789");
-    try testing.expectEqual(@as(u32, 0xe3069283), c.final());
-}
-
 test "CRC-32/ISO-HDLC" {
     const Crc32IsoHdlc = crc.Crc32IsoHdlc;
 

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;
+    const have_posix_memalign = switch (builtin.os.tag) {
-    pub const malloc_size = if (@TypeOf(c.malloc_size) != void)
+        .dragonfly,
-        c.malloc_size
+        .netbsd,
-    else if (@TypeOf(c.malloc_usable_size) != void)
+        .freebsd,
-        c.malloc_usable_size
+        .illumos,
-    else if (@TypeOf(c._msize) != void)
+        .openbsd,
-        c._msize
+        .linux,
-    else {};
+        .driverkit,
-
+        .ios,
-    pub const supports_posix_memalign = switch (builtin.os.tag) {
+        .maccatalyst,
-        .dragonfly, .netbsd, .freebsd, .illumos, .openbsd, .linux, .driverkit, .ios, .maccatalyst, .macos, .tvos, .visionos, .watchos, .serenity => true,
+        .macos,
+        .tvos,
+        .visionos,
+        .watchos,
+        .serenity,
+        => true,
         else => false,
     };
 
-    fn getHeader(ptr: [*]u8) *[*]u8 {
+    fn allocStrat(need_align: Alignment) union(enum) {
-        return @ptrCast(@alignCast(ptr - @sizeOf(usize)));
+        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 {
+    /// If `allocStrat(a) == .manual_align`, an allocation looks like this:
-        const alignment_bytes = alignment.toByteUnits();
+    ///
-        if (supports_posix_memalign) {
+    /// unaligned_ptr   hdr_ptr  aligned_ptr
-            // The posix_memalign only accepts alignment values that are a
+    /// v               v        v
-            // multiple of the pointer size
+    /// +---------------+--------+--------------+
-            const effective_alignment = @max(alignment_bytes, @sizeOf(usize));
+    /// |    padding    | header | usable bytes |
-
+    /// +---------------+--------+--------------+
-            var aligned_ptr: ?*anyopaque = undefined;
+    ///
-            if (c.posix_memalign(&aligned_ptr, effective_alignment, len) != 0)
+    /// * `unaligned_ptr` is the raw return value of `malloc`.
-                return null;
+    /// * `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
-            return @ptrCast(aligned_ptr);
+    ///   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.
-
+    ///
-        // Thin wrapper around regular malloc, overallocate to account for
+    /// This function accepts `aligned_ptr` and offsets it backwards to return `hdr_ptr`.
-        // alignment padding and store the original malloc()'ed pointer before
+    fn manualAlignHeader(aligned_ptr: [*]u8) *[*]u8 {
-        // the aligned address.
+        return @ptrCast(@alignCast(aligned_ptr - @sizeOf(usize)));
-        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;
     }
 
     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) {
+        assert(new_len > 0);
-            return true;
+        if (new_len <= memory.len) {
+            return true; // in-place shrink always works
         }
-        if (CAllocator.supports_malloc_size) {
+        const mallocSize = func: {
-            const full_len = alignedAllocSize(buf.ptr);
+            if (@TypeOf(c.malloc_size) != void) break :func c.malloc_size;
-            if (new_len <= full_len) {
+            if (@TypeOf(c.malloc_usable_size) != void) break :func c.malloc_usable_size;
-                return true;
+            if (@TypeOf(c._msize) != void) break :func c._msize;
-            }
+            return false; // we don't know how much space is actually available
-        }
+        };
-        return false;
+        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
+        assert(new_len > 0);
-        // respect the original alignment.
+        // Prefer resizing in-place if possible, since `realloc` could be expensive even if legal.
-        return if (resize(context, memory, alignment, new_len, return_address)) memory.ptr else null;
+        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
+/// Asserts that allocations have alignments which `malloc` can satisfy. This means that
-/// `malloc_usable_size` if available. For an allocator that directly calls
+/// the requested alignment is no greater than `@min(@alignOf(std.c.max_align_t), size)`.
-/// `malloc`/`free`, see `raw_c_allocator`.
+///
-pub const c_allocator: Allocator = .{
+/// This allocator is rarely appropriate to use. In general, prefer `c_allocator`, which
-    .ptr = undefined,
+/// does not have any special requirements of its input, but is still highly efficient for
-    .vtable = &CAllocator.vtable,
+/// allocation requests which obey `malloc` alignment rules.
-};
-
-/// 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`.
 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)));
+    // `std.c.max_align_t` isn't the whole story, because if `len` is smaller than
-    // Note that this pointer cannot be aligncasted to max_align_t because if
+    // every C type with alignment `max_align_t`, the allocation can be less-aligned.
-    // len is < max_align_t then the alignment can be smaller. For example, if
+    // The implementation need only guarantee that any type of length `len` would be
-    // max_align_t is 16, but the user requests 8 bytes, there is no built-in
+    // suitably aligned.
-    // 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
+    // For instance, if `len == 8` and `alignment == .@"16"`, then `malloc` may not
-    // is allowed to return a 1-byte aligned pointer.
+    // 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/os/linux.zig

@@ -1748,9 +1748,7 @@ pub fn settimeofday(tv: *const timeval, tz: *const timezone) usize {
 }
 
 pub fn nanosleep(req: *const timespec, rem: ?*timespec) usize {
-    if (native_arch == .riscv32) {
+    return syscall2(.nanosleep, @intFromPtr(req), @intFromPtr(rem));
-        @compileError("No nanosleep syscall on this architecture.");
-    } else return syscall2(.nanosleep, @intFromPtr(req), @intFromPtr(rem));
 }
 
 pub fn pause() usize {
@@ -3773,6 +3771,7 @@ pub const SIG = if (is_mips) enum(u32) {
     PROF = 29,
     XCPU = 30,
     XFZ = 31,
+    _,
 } else if (is_sparc) enum(u32) {
     pub const BLOCK = 1;
     pub const UNBLOCK = 2;
@@ -3818,6 +3817,7 @@ pub const SIG = if (is_mips) enum(u32) {
     LOST = 29,
     USR1 = 30,
     USR2 = 31,
+    _,
 } else enum(u32) {
     pub const BLOCK = 0;
     pub const UNBLOCK = 1;
@@ -3861,6 +3861,7 @@ pub const SIG = if (is_mips) enum(u32) {
     IO = 29,
     PWR = 30,
     SYS = 31,
+    _,
 };
 
 pub const kernel_rwf = u32;

lib/std/posix.zig

@@ -1360,6 +1360,7 @@ pub fn writev(fd: fd_t, iov: []const iovec_const) WriteError!usize {
             .PIPE => return error.BrokenPipe,
             .CONNRESET => return error.ConnectionResetByPeer,
             .BUSY => return error.DeviceBusy,
+            .CANCELED => return error.Canceled,
             else => |err| return unexpectedErrno(err),
         }
     }

lib/std/process.zig

@@ -690,6 +690,9 @@ pub const ArgIteratorWasi = struct {
 
     /// Call to free the internal buffer of the iterator.
     pub fn deinit(self: *ArgIteratorWasi) void {
+        // Nothing is allocated when there are no args
+        if (self.args.len == 0) return;
+
         const last_item = self.args[self.args.len - 1];
         const last_byte_addr = @intFromPtr(last_item.ptr) + last_item.len + 1; // null terminated
         const first_item_ptr = self.args[0].ptr;

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;

lib/std/zig/Ast.zig

@@ -1195,6 +1195,9 @@ pub fn lastToken(tree: Ast, node: Node.Index) TokenIndex {
                 if (extra.section_node.unwrap()) |section_node| {
                     end_offset += 1; // for the rparen
                     n = section_node;
+                } else if (extra.addrspace_node.unwrap()) |addrspace_node| {
+                    end_offset += 1; // for the rparen
+                    n = addrspace_node;
                 } else if (extra.align_node.unwrap()) |align_node| {
                     end_offset += 1; // for the rparen
                     n = align_node;

lib/std/zig/parser_test.zig

@@ -6084,6 +6084,16 @@ test "zig fmt: do not canonicalize invalid cast builtins" {
     );
 }
 
+test "zig fmt: extern addrspace in struct" {
+    try testCanonical(
+        \\const namespace = struct {
+        \\    extern const num: u8 addrspace(.generic);
+        \\};
+        \\// comment
+        \\
+    );
+}
+
 test "recovery: top level" {
     try testError(
         \\test "" {inline}

src/Sema.zig

@@ -175,9 +175,11 @@ const ComptimeAlloc = struct {
 
 /// `src` may be `null` if `is_const` will be set.
 fn newComptimeAlloc(sema: *Sema, block: *Block, src: LazySrcLoc, ty: Type, alignment: Alignment) !ComptimeAllocIndex {
+    const pt = sema.pt;
+    const init_val = try sema.typeHasOnePossibleValue(ty) orelse try pt.undefValue(ty);
     const idx = sema.comptime_allocs.items.len;
     try sema.comptime_allocs.append(sema.gpa, .{
-        .val = .{ .interned = try sema.pt.intern(.{ .undef = ty.toIntern() }) },
+        .val = .{ .interned = init_val.toIntern() },
         .is_const = false,
         .src = src,
         .alignment = alignment,

src/Type.zig

@@ -3445,13 +3445,22 @@ pub fn optEuBaseType(ty: Type, zcu: *const Zcu) Type {
 
 pub fn toUnsigned(ty: Type, pt: Zcu.PerThread) !Type {
     const zcu = pt.zcu;
-    return switch (ty.zigTypeTag(zcu)) {
+    return switch (ty.toIntern()) {
-        .int => pt.intType(.unsigned, ty.intInfo(zcu).bits),
+        // zig fmt: off
-        .vector => try pt.vectorType(.{
+        .usize_type,       .isize_type      => .usize,
-            .len = ty.vectorLen(zcu),
+        .c_ushort_type,    .c_short_type    => .c_ushort,
-            .child = (try ty.childType(zcu).toUnsigned(pt)).toIntern(),
+        .c_uint_type,      .c_int_type      => .c_uint,
-        }),
+        .c_ulong_type,     .c_long_type     => .c_ulong,
-        else => unreachable,
+        .c_ulonglong_type, .c_longlong_type => .c_ulonglong,
+        // zig fmt: on
+        else => switch (ty.zigTypeTag(zcu)) {
+            .int => pt.intType(.unsigned, ty.intInfo(zcu).bits),
+            .vector => try pt.vectorType(.{
+                .len = ty.vectorLen(zcu),
+                .child = (try ty.childType(zcu).toUnsigned(pt)).toIntern(),
+            }),
+            else => unreachable,
+        },
     };
 }
 

src/codegen/c.zig

@@ -5478,14 +5478,16 @@ fn airAsm(f: *Function, inst: Air.Inst.Index) !CValue {
             // for the string, we still use the next u32 for the null terminator.
             extra_i += (constraint.len + name.len + (2 + 3)) / 4;
 
-            if (constraint.len < 2 or constraint[0] != '=' or
+            // +constraint
-                (constraint[1] == '{' and constraint[constraint.len - 1] != '}'))
+            // =constraint
-            {
+            if (constraint.len > 1 and
-                return f.fail("CBE: constraint not supported: '{s}'", .{constraint});
+                (constraint[0] == '=' or constraint[0] == '+') and
-            }
+                constraint[1] != '{') continue;
 
-            const is_reg = constraint[1] == '{';
+            // ={reg}
-            if (is_reg) {
+            if (std.mem.startsWith(u8, constraint, "={") and
+                std.mem.endsWith(u8, constraint, "}"))
+            {
                 const output_ty = if (output == .none) inst_ty else f.typeOf(output).childType(zcu);
                 try w.writeAll("register ");
                 const output_local = try f.allocLocalValue(.{
@@ -5503,7 +5505,7 @@ fn airAsm(f: *Function, inst: Air.Inst.Index) !CValue {
                 }
                 try w.writeByte(';');
                 try f.object.newline();
-            }
+            } else return f.fail("CBE: constraint not supported: '{s}'", .{constraint});
         }
         for (inputs) |input| {
             const extra_bytes = mem.sliceAsBytes(f.air.extra.items[extra_i..]);

src/codegen/x86_64/CodeGen.zig

@@ -177437,6 +177437,10 @@ fn airAsm(self: *CodeGen, inst: Air.Inst.Index) !void {
         fixed_mnem_size: {
             const fixed_mnem_size: Memory.Size = switch (mnem_tag) {
                 .clflush => .byte,
+                .crc32 => {
+                    mnem_size.op_has_size.unset(1);
+                    break :fixed_mnem_size;
+                },
                 .fldcw, .fnstcw, .fstcw, .fnstsw, .fstsw => .word,
                 .fldenv, .fnstenv, .fstenv => .none,
                 .frstor, .fsave, .fnsave, .fxrstor, .fxrstor64, .fxsave, .fxsave64 => .none,

src/link/Elf/Atom.zig

@@ -1499,22 +1499,18 @@ const aarch64 = struct {
             .ABS64 => {
                 try atom.scanReloc(symbol, rel, dynAbsRelocAction(symbol, elf_file), elf_file);
             },
-
             .ADR_PREL_PG_HI21 => {
                 try atom.scanReloc(symbol, rel, pcRelocAction(symbol, elf_file), elf_file);
             },
-
             .ADR_GOT_PAGE => {
                 // TODO: relax if possible
                 symbol.flags.needs_got = true;
             },
-
             .LD64_GOT_LO12_NC,
             .LD64_GOTPAGE_LO15,
             => {
                 symbol.flags.needs_got = true;
             },
-
             .CALL26,
             .JUMP26,
             => {
@@ -1522,25 +1518,21 @@ const aarch64 = struct {
                     symbol.flags.needs_plt = true;
                 }
             },
-
             .TLSLE_ADD_TPREL_HI12,
             .TLSLE_ADD_TPREL_LO12_NC,
             => {
                 if (is_dyn_lib) try atom.reportPicError(symbol, rel, elf_file);
             },
-
             .TLSIE_ADR_GOTTPREL_PAGE21,
             .TLSIE_LD64_GOTTPREL_LO12_NC,
             => {
                 symbol.flags.needs_gottp = true;
             },
-
             .TLSGD_ADR_PAGE21,
             .TLSGD_ADD_LO12_NC,
             => {
                 symbol.flags.needs_tlsgd = true;
             },
-
             .TLSDESC_ADR_PAGE21,
             .TLSDESC_LD64_LO12,
             .TLSDESC_ADD_LO12,
@@ -1551,18 +1543,17 @@ const aarch64 = struct {
                     symbol.flags.needs_tlsdesc = true;
                 }
             },
-
             .ADD_ABS_LO12_NC,
             .ADR_PREL_LO21,
-            .LDST8_ABS_LO12_NC,
+            .CONDBR19,
+            .LDST128_ABS_LO12_NC,
             .LDST16_ABS_LO12_NC,
             .LDST32_ABS_LO12_NC,
             .LDST64_ABS_LO12_NC,
-            .LDST128_ABS_LO12_NC,
+            .LDST8_ABS_LO12_NC,
             .PREL32,
             .PREL64,
             => {},
-
             else => try atom.reportUnhandledRelocError(rel, elf_file),
         }
     }
@@ -1599,7 +1590,6 @@ const aarch64 = struct {
                     r_offset,
                 );
             },
-
             .CALL26,
             .JUMP26,
             => {
@@ -1611,27 +1601,26 @@ const aarch64 = struct {
                 };
                 util.writeBranchImm(disp, code);
             },
-
+            .CONDBR19 => {
+                const value = math.cast(i19, S + A - P) orelse return error.Overflow;
+                util.writeCondBrImm(value, code);
+            },
             .PREL32 => {
                 const value = math.cast(i32, S + A - P) orelse return error.Overflow;
                 mem.writeInt(u32, code, @bitCast(value), .little);
             },
-
             .PREL64 => {
                 const value = S + A - P;
                 mem.writeInt(u64, code_buffer[r_offset..][0..8], @bitCast(value), .little);
             },
-
             .ADR_PREL_LO21 => {
                 const value = math.cast(i21, S + A - P) orelse return error.Overflow;
                 util.writeAdrInst(value, code);
             },
-
             .ADR_PREL_PG_HI21 => {
                 // TODO: check for relaxation of ADRP+ADD
                 util.writeAdrInst(try util.calcNumberOfPages(P, S + A), code);
             },
-
             .ADR_GOT_PAGE => if (target.flags.has_got) {
                 util.writeAdrInst(try util.calcNumberOfPages(P, G + GOT + A), code);
             } else {
@@ -1644,18 +1633,15 @@ const aarch64 = struct {
                     r_offset,
                 });
             },
-
             .LD64_GOT_LO12_NC => {
                 assert(target.flags.has_got);
                 const taddr = @as(u64, @intCast(G + GOT + A));
                 util.writeLoadStoreRegInst(@divExact(@as(u12, @truncate(taddr)), 8), code);
             },
-
             .ADD_ABS_LO12_NC => {
                 const taddr = @as(u64, @intCast(S + A));
                 util.writeAddImmInst(@truncate(taddr), code);
             },
-
             .LDST8_ABS_LO12_NC,
             .LDST16_ABS_LO12_NC,
             .LDST32_ABS_LO12_NC,
@@ -1674,44 +1660,37 @@ const aarch64 = struct {
                 };
                 util.writeLoadStoreRegInst(off, code);
             },
-
             .TLSLE_ADD_TPREL_HI12 => {
                 const value = math.cast(i12, (S + A - TP) >> 12) orelse
                     return error.Overflow;
                 util.writeAddImmInst(@bitCast(value), code);
             },
-
             .TLSLE_ADD_TPREL_LO12_NC => {
                 const value: i12 = @truncate(S + A - TP);
                 util.writeAddImmInst(@bitCast(value), code);
             },
-
             .TLSIE_ADR_GOTTPREL_PAGE21 => {
                 const S_ = target.gotTpAddress(elf_file);
                 relocs_log.debug("      [{x} => {x}]", .{ P, S_ + A });
                 util.writeAdrInst(try util.calcNumberOfPages(P, S_ + A), code);
             },
-
             .TLSIE_LD64_GOTTPREL_LO12_NC => {
                 const S_ = target.gotTpAddress(elf_file);
                 relocs_log.debug("      [{x} => {x}]", .{ P, S_ + A });
                 const off: u12 = try math.divExact(u12, @truncate(@as(u64, @bitCast(S_ + A))), 8);
                 util.writeLoadStoreRegInst(off, code);
             },
-
             .TLSGD_ADR_PAGE21 => {
                 const S_ = target.tlsGdAddress(elf_file);
                 relocs_log.debug("      [{x} => {x}]", .{ P, S_ + A });
                 util.writeAdrInst(try util.calcNumberOfPages(P, S_ + A), code);
             },
-
             .TLSGD_ADD_LO12_NC => {
                 const S_ = target.tlsGdAddress(elf_file);
                 relocs_log.debug("      [{x} => {x}]", .{ P, S_ + A });
                 const off: u12 = @truncate(@as(u64, @bitCast(S_ + A)));
                 util.writeAddImmInst(off, code);
             },
-
             .TLSDESC_ADR_PAGE21 => {
                 if (target.flags.has_tlsdesc) {
                     const S_ = target.tlsDescAddress(elf_file);
@@ -1722,7 +1701,6 @@ const aarch64 = struct {
                     util.encoding.Instruction.nop().write(code);
                 }
             },
-
             .TLSDESC_LD64_LO12 => {
                 if (target.flags.has_tlsdesc) {
                     const S_ = target.tlsDescAddress(elf_file);
@@ -1734,7 +1712,6 @@ const aarch64 = struct {
                     util.encoding.Instruction.nop().write(code);
                 }
             },
-
             .TLSDESC_ADD_LO12 => {
                 if (target.flags.has_tlsdesc) {
                     const S_ = target.tlsDescAddress(elf_file);
@@ -1747,13 +1724,11 @@ const aarch64 = struct {
                     util.encoding.Instruction.movz(.x0, value, .{ .lsl = .@"16" }).write(code);
                 }
             },
-
             .TLSDESC_CALL => if (!target.flags.has_tlsdesc) {
                 relocs_log.debug("      relaxing br => movk(x0, {x})", .{S + A - TP});
                 const value: u16 = @bitCast(@as(i16, @truncate(S + A - TP)));
                 util.encoding.Instruction.movk(.x0, value, .{}).write(code);
             },
-
             else => try atom.reportUnhandledRelocError(rel, elf_file),
         }
     }

src/link/aarch64.zig

@@ -29,6 +29,12 @@ pub fn writeBranchImm(disp: i28, code: *[4]u8) void {
     inst.write(code);
 }
 
+pub fn writeCondBrImm(disp: i19, code: *[4]u8) void {
+    var inst: encoding.Instruction = .read(code);
+    inst.branch_exception_generating_system.conditional_branch_immediate.group.imm19 = @intCast(@shrExact(disp, 2));
+    inst.write(code);
+}
+
 const assert = std.debug.assert;
 const builtin = @import("builtin");
 const math = std.math;

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) {
+        if (builtin.link_libc) break :gpa .{ std.heap.c_allocator, false };
-            // 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 };
-        }
         break :gpa switch (builtin.mode) {
             .Debug, .ReleaseSafe => .{ debug_allocator.allocator(), true },
             .ReleaseFast, .ReleaseSmall => .{ std.heap.smp_allocator, false },

src/print_zir.zig

@@ -595,7 +595,7 @@ const Writer = struct {
             },
 
             .reify_slice_arg_ty => {
-                const reify_slice_arg_info: Zir.Inst.ReifySliceArgInfo = @enumFromInt(extended.operand);
+                const reify_slice_arg_info: Zir.Inst.ReifySliceArgInfo = @enumFromInt(extended.small);
                 const extra = self.code.extraData(Zir.Inst.UnNode, extended.operand).data;
                 try stream.print("{t}, ", .{reify_slice_arg_info});
                 try self.writeInstRef(stream, extra.operand);

test/behavior/abs.zig

@@ -1,5 +1,6 @@
 const builtin = @import("builtin");
 const std = @import("std");
+const assert = std.debug.assert;
 const expect = std.testing.expect;
 
 test "@abs integers" {
@@ -48,6 +49,33 @@ fn testAbsIntegers() !void {
     }
 }
 
+test "@abs signed C ABI integers" {
+    if (builtin.zig_backend == .stage2_aarch64) return error.SkipZigTest; // TODO
+    if (builtin.zig_backend == .stage2_arm) return error.SkipZigTest; // TODO
+    if (builtin.zig_backend == .stage2_sparc64) return error.SkipZigTest; // TODO
+    if (builtin.zig_backend == .stage2_riscv64) return error.SkipZigTest; // TODO
+
+    const S = struct {
+        fn doTheTest() !void {
+            try testOne(isize, usize);
+            try testOne(c_short, c_ushort);
+            try testOne(c_int, c_uint);
+            try testOne(c_long, c_ulong);
+            if (!builtin.cpu.arch.isSpirV()) try testOne(c_longlong, c_ulonglong);
+        }
+        fn testOne(comptime Signed: type, comptime Unsigned: type) !void {
+            var negative_one: Signed = undefined;
+            negative_one = -1;
+            const one = @abs(negative_one);
+            comptime assert(@TypeOf(one) == Unsigned);
+            try expect(one == 1);
+        }
+    };
+
+    try S.doTheTest();
+    try comptime S.doTheTest();
+}
+
 test "@abs unsigned integers" {
     if (builtin.zig_backend == .stage2_arm) return error.SkipZigTest; // TODO
     if (builtin.zig_backend == .stage2_sparc64) return error.SkipZigTest; // TODO
@@ -87,6 +115,32 @@ fn testAbsUnsignedIntegers() !void {
     }
 }
 
+test "@abs unsigned C ABI integers" {
+    if (builtin.zig_backend == .stage2_aarch64) return error.SkipZigTest; // TODO
+    if (builtin.zig_backend == .stage2_arm) return error.SkipZigTest; // TODO
+    if (builtin.zig_backend == .stage2_sparc64) return error.SkipZigTest; // TODO
+
+    const S = struct {
+        fn doTheTest() !void {
+            try testOne(usize);
+            try testOne(c_ushort);
+            try testOne(c_uint);
+            try testOne(c_ulong);
+            if (!builtin.cpu.arch.isSpirV()) try testOne(c_ulonglong);
+        }
+        fn testOne(comptime Unsigned: type) !void {
+            var one: Unsigned = undefined;
+            one = 1;
+            const still_one = @abs(one);
+            comptime assert(@TypeOf(still_one) == Unsigned);
+            try expect(still_one == 1);
+        }
+    };
+
+    try S.doTheTest();
+    try comptime S.doTheTest();
+}
+
 test "@abs big int <= 128 bits" {
     if (builtin.zig_backend == .stage2_arm) return error.SkipZigTest; // TODO
     if (builtin.zig_backend == .stage2_sparc64) return error.SkipZigTest; // TODO

test/behavior/type.zig

@@ -427,3 +427,12 @@ test "undefined type value" {
     };
     comptime assert(@TypeOf(S.undef_type) == type);
 }
+
+test "reify struct with zero fields through const arrays" {
+    const names: [0][]const u8 = .{};
+    const types: [0]type = .{};
+    const attrs: [0]std.builtin.Type.StructField.Attributes = .{};
+    const S = @Struct(.auto, null, &names, &types, &attrs);
+    comptime assert(@typeInfo(S) == .@"struct");
+    comptime assert(@typeInfo(S).@"struct".fields.len == 0);
+}

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
+    // These emulated targets need a lot more RAM for unknown reasons.
-    const max_rss = if (target.os.tag == .wasi)
+    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

tools/crc/catalog.txt

@@ -97,7 +97,8 @@ width=32  poly=0xa833982b  init=0xffffffff  refin=true  refout=true  xorout=0xff
 width=32  poly=0x04c11db7  init=0xffffffff  refin=false  refout=false  xorout=0xffffffff  check=0xfc891918  residue=0xc704dd7b  name="CRC-32/BZIP2"
 width=32  poly=0x8001801b  init=0x00000000  refin=true  refout=true  xorout=0x00000000  check=0x6ec2edc4  residue=0x00000000  name="CRC-32/CD-ROM-EDC"
 width=32  poly=0x04c11db7  init=0x00000000  refin=false  refout=false  xorout=0xffffffff  check=0x765e7680  residue=0xc704dd7b  name="CRC-32/CKSUM"
-width=32  poly=0x1edc6f41  init=0xffffffff  refin=true  refout=true  xorout=0xffffffff  check=0xe3069283  residue=0xb798b438  name="CRC-32/ISCSI"
+# CRC-32C implementation is defined manually, since it has an accelerated variant.
+# width=32  poly=0x1edc6f41  init=0xffffffff  refin=true  refout=true  xorout=0xffffffff  check=0xe3069283  residue=0xb798b438  name="CRC-32/ISCSI"
 width=32  poly=0x04c11db7  init=0xffffffff  refin=true  refout=true  xorout=0xffffffff  check=0xcbf43926  residue=0xdebb20e3  name="CRC-32/ISO-HDLC"
 width=32  poly=0x04c11db7  init=0xffffffff  refin=true  refout=true  xorout=0x00000000  check=0x340bc6d9  residue=0x00000000  name="CRC-32/JAMCRC"
 width=32  poly=0x741b8cd7  init=0xffffffff  refin=true  refout=true  xorout=0xffffffff  check=0x2d3dd0ae  residue=0x00000000  name="CRC-32/KOOPMAN"

tools/update_crc_catalog.zig

@@ -36,6 +36,7 @@ pub fn main() anyerror!void {
     try code_writer.writeAll(
         \\//! This file is auto-generated by tools/update_crc_catalog.zig.
         \\
+        \\const builtin = @import("builtin");
         \\const impl = @import("crc/impl.zig");
         \\
         \\pub const Crc = impl.Crc;
@@ -49,6 +50,17 @@ pub fn main() anyerror!void {
         \\    _ = @import("crc/test.zig");
         \\}
         \\
+        \\pub const Crc32Iscsi = switch (builtin.cpu.has(.x86, .crc32)) {
+        \\    true => @import("crc/crc32c.zig").Wrapper,
+        \\    else => Crc(u32, .{
+        \\        .polynomial = 0x1edc6f41,
+        \\        .initial = 0xffffffff,
+        \\        .reflect_input = true,
+        \\        .reflect_output = true,
+        \\        .xor_output = 0xffffffff,
+        \\    }),
+        \\};
+        \\
     );
 
     var zig_test_file = try crc_target_dir.createFile("test.zig", .{});
@@ -80,12 +92,23 @@ pub fn main() anyerror!void {
         \\}
         \\
         \\test "crc32 koopman regression" {
-        \\    const crc32 = crc.Koopman;
+        \\    const crc32 = crc.Crc32Koopman;
         \\    try testing.expectEqual(crc32.hash(""), 0x00000000);
         \\    try testing.expectEqual(crc32.hash("a"), 0x0da2aa8a);
         \\    try testing.expectEqual(crc32.hash("abc"), 0xba2322ac);
         \\}
         \\
+        \\test "CRC-32/ISCSI" {
+        \\    const Crc32Iscsi = crc.Crc32Iscsi;
+        \\
+        \\    try testing.expectEqual(@as(u32, 0xe3069283), Crc32Iscsi.hash("123456789"));
+        \\
+        \\    var c = Crc32Iscsi.init();
+        \\    c.update("1234");
+        \\    c.update("56789");
+        \\    try testing.expectEqual(@as(u32, 0xe3069283), c.final());
+        \\}
+        \\
     );
 
     var reader: std.Io.Reader = .fixed(catalog_txt);

tools/update_glibc.zig

@@ -1,6 +1,6 @@
 //! This script updates the .c, .h, .s, and .S files that make up the start
 //! files such as crt1.o. Not to be confused with
-//! https://github.com/ziglang/glibc-abi-tool/ which updates the `abilists`
+//! https://codeberg.org/ziglang/libc-abi-tools which updates the `abilists`
 //! file.
 //!
 //! Example usage: