Update to Zig 0.16

.github/workflows/release.yml

@@ -71,7 +71,7 @@ jobs:
       - name: Install Zig
         run: |
           ZIG_ARCH=$(uname -m | sed 's/arm64/aarch64/')
-          ZIG_VERSION="0.15.2"
+          ZIG_VERSION="0.16.0"
           curl -sSfL "https://ziglang.org/download/${ZIG_VERSION}/zig-${ZIG_ARCH}-macos-${ZIG_VERSION}.tar.xz" \
             | tar xJ -C /tmp
           echo "/tmp/zig-${ZIG_ARCH}-macos-${ZIG_VERSION}" >> $GITHUB_PATH

.github/workflows/test.yml

@@ -33,7 +33,7 @@ jobs:
       - name: Install Zig
         if: ${{ !matrix.fallback }}
         run: |
-          ZIG_VERSION="0.15.2"
+          ZIG_VERSION="0.16.0"
           ZIG_ARCH=$(uname -m | sed 's/arm64/aarch64/')
           # Zig uses "macos" in download slugs, not "darwin" as uname reports.
           ZIG_OS=$(uname -s | tr '[:upper:]' '[:lower:]' | sed 's/darwin/macos/')

.mise.toml

@@ -1,3 +1,3 @@
 [tools]
-ruby = ["4.0.2", "3.4.8"]
-zig = "0.15.2"
+ruby = ["4.0.3", "3.4.9"]
+zig = "0.16.0"

CHANGELOG.md

@@ -1,3 +1,9 @@
+## [Unreleased]
+
+- Update to Zig 0.16.0; update libxev and zig.rb to the latest versions.
+- Performance improvements. README updated with latest numbers.
+- Improve benchmarking: default to 3 measured runs plus 1 warmup run. Now reporting `±CV%` and the delta column tags rows `(noisy)` when the change is within jitter. Add the ability to benchmark released vs. unreleased carbon_fiber.
+
 ## [0.1.2] - 2026-04-29
 
 - Fix the `x86_64-linux` precompiled gem (erroneously shipped an aarch64 

Gemfile.lock

@@ -35,7 +35,7 @@ GEM
     prism (1.9.0)
     racc (1.8.1)
     rainbow (3.1.1)
-    rake (13.3.1)
+    rake (13.4.2)
     rake-compiler-dock (1.11.1)
     regexp_parser (2.12.0)
     rspec (3.13.2)
@@ -124,7 +124,7 @@ CHECKSUMS
   prism (1.9.0) sha256=7b530c6a9f92c24300014919c9dcbc055bf4cdf51ec30aed099b06cd6674ef85
   racc (1.8.1) sha256=4a7f6929691dbec8b5209a0b373bc2614882b55fc5d2e447a21aaa691303d62f
   rainbow (3.1.1) sha256=039491aa3a89f42efa1d6dec2fc4e62ede96eb6acd95e52f1ad581182b79bc6a
-  rake (13.3.1) sha256=8c9e89d09f66a26a01264e7e3480ec0607f0c497a861ef16063604b1b08eb19c
+  rake (13.4.2) sha256=cb825b2bd5f1f8e91ca37bddb4b9aaf345551b4731da62949be002fa89283701
   rake-compiler-dock (1.11.1) sha256=9234ded6c3311f4ba6db77f2833f0aa33c073b4d33c896a2156e2551af828e0a
   regexp_parser (2.12.0) sha256=35a916a1d63190ab5c9009457136ae5f3c0c7512d60291d0d1378ba18ce08ebb
   rspec (3.13.2) sha256=206284a08ad798e61f86d7ca3e376718d52c0bc944626b2349266f239f820587

README.md

@@ -50,18 +50,18 @@ end
 
 ## Performance
 
-AWS EC2 c7a.2xlarge, 8 dedicated vCPUs, Ubuntu 24.04 LTS, kernel 6.17, Ruby 4.0.2 + YJIT, io_uring. 5-run median.
+AWS EC2 c7a.2xlarge, 8 dedicated vCPUs, Amazon Linux 2023, kernel 6.18.20, Ruby 4.0.2 + YJIT, io_uring. 5-run median.
 
 Some benchmarks:
 
 | Workload | Carbon Fiber | Async | Itsi | Carbon Fiber vs. Async |
 |---|---|---|---|---|
-| `http_server` | **48.175k req/s** | 37.409k req/s | 29.708k req/s | +29% |
-| `http_client_api` | **15.528k req/s** | 13.373k req/s | timeout | +16% |
-| `http_client_download` | **6.747k dl/s** | 5.920k dl/s | timeout | +14% |
-| `tcp_echo` | **50.392k ops/s** | 38.907k ops/s | 30.660k ops/s | +29% |
-| `cascading_timeout` | **4.659k ops/s** | 4.488k ops/s | error | +4% |
-| `connection_pool` | **4.989k co/s** | 4.912k co/s | 4.968k co/s | +2% |
+| `http_server` | **49.380k req/s** | 30.823k req/s | 30.864k req/s | +60% |
+| `http_client_api` | **19.500k req/s** | 16.721k req/s | timeout | +17% |
+| `http_client_download` | **8.426k dl/s** | 7.062k dl/s | timeout | +19% |
+| `tcp_echo` | **52.973k ops/s** | 32.330k ops/s | 32.046k ops/s | +64% |
+| `cascading_timeout` | **4.668k ops/s** | 4.414k ops/s | error | +6% |
+| `connection_pool` | **4.967k co/s** | 4.612k co/s | 4.954k co/s | +8% |
 
 Wins on most workloads against Async, Itsi, fiber_scheduler, io-event, and libev. [See detailed benchmarks →](#benchmarks)
 
@@ -256,22 +256,22 @@ If the native extension can't be loaded (on Windows, for example), a pure-Ruby f
 
 ## Benchmarks
 
-AWS EC2 c7a.2xlarge, 8 dedicated vCPUs, Ubuntu 24.04 LTS, kernel 6.17, Ruby 4.0.2 + YJIT, io_uring. 5-run median.
+AWS EC2 c7a.2xlarge, 8 dedicated vCPUs, Amazon Linux 2023, kernel 6.18.20, Ruby 4.0.2 + YJIT, io_uring. 5-run median.
 
 ### Ruby Fiber Schedulers (leading ones): Carbon Fiber vs. Async vs. Itsi
 
 Measuring pure Ruby Fiber Scheduler performance (`Fiber.set_scheduler`).
 
 | Workload | Unit | Carbon Fiber | Async | Itsi | Carbon Fiber vs. Async |
 |---|---|---|---|---|---|
-| `http_client_api` | req/s | **15,528** | 13,373 | timeout | +16% |
-| `http_client_download` | dl/s | **6,747** | 5,920 | timeout | +14% |
-| `http_server` | req/s | **48,175** | 37,409 | 29,708 | +29% |
-| `tcp_echo` | ops/s | **50,392** | 38,907 | 30,660 | +29% |
-| `connection_pool` | co/s | **4,989** | 4,912 | 4,968 | +2% |
-| `fan_out_gather` | cyc/s | 2,024 | 2,046 | **2,104** | −1% |
-| `db_query_mix` | qry/s | 1,660 | 1,652 | **1,662** | +0.5% |
-| `cascading_timeout` | ops/s | **4,659** | 4,488 | error | +4% |
+| `http_client_api` | req/s | **19,500** | 16,721 | timeout | +17% |
+| `http_client_download` | dl/s | **8,426** | 7,062 | timeout | +19% |
+| `http_server` | req/s | **49,380** | 30,823 | 30,864 | +60% |
+| `tcp_echo` | ops/s | **52,973** | 32,330 | 32,046 | +64% |
+| `connection_pool` | co/s | **4,967** | 4,612 | 4,954 | +8% |
+| `fan_out_gather` | cyc/s | 2,022 | 1,923 | **2,094** | +5% |
+| `db_query_mix` | qry/s | 1,660 | 1,623 | **1,662** | +2% |
+| `cascading_timeout` | ops/s | **4,668** | 4,414 | error | +6% |
 
 Enabling YJIT turned out to be very beneficial for Async as well—numbers here are with `--yjit` on both sides.
 
@@ -281,12 +281,12 @@ Swapped the io-event selector for Carbon Fiber's native backend. Same Async code
 
 | Workload | Unit | Stock Async | Carbon Fiber | Delta |
 |---|---|---|---|---|
-| `http_client_api` | req/s | 13,375 | **14,331** | +7.1% |
-| `http_client_download` | dl/s | 3,893 | **3,956** | +1.6% |
-| `task_churn` | task/s | **87,883** | 85,027 | −3.3% |
-| `condition_signal` | sig/s | 337,282 | **361,089** | +7.1% |
-| `cascading_timeout` | ops/s | 4,497 | **4,511** | +0.3% |
-| `tcp_throughput` | ops/s | 42,292 | **51,930** | +22.8% |
+| `http_client_api` | req/s | 17,119 | **17,323** | +1.2% |
+| `http_client_download` | dl/s | 2,263 | **2,474** | +9.3% |
+| `task_churn` | task/s | 119,477 | **120,985** | +1.3% |
+| `condition_signal` | sig/s | 44,887 | **46,921** | +4.5% |
+| `cascading_timeout` | ops/s | 4,414 | **4,472** | +1.3% |
+| `tcp_throughput` | ops/s | 32,606 | **48,214** | +47.9% |
 
 ### Examples of how to run benchmarks
 

Rakefile

@@ -20,14 +20,9 @@ task :sync_version do
   end
 end
 
-# Zig 0.15's Mach-O linker has issues with Xcode 26.4
-#  - https://codeberg.org/ziglang/zig/pulls/31673
-#  - https://codeberg.org/ziglang/zig/issues/31658
-# Workaround: DEVELOPER_DIR=/dev/null makes Zig use its bundled libSystem.tbd.
 desc "Compile the Zig native extension (also syncs version)"
 task compile: :sync_version do
-  env = RUBY_PLATFORM.include?("darwin") ? {"DEVELOPER_DIR" => "/dev/null"} : {}
-  sh env, "zig", "build", "-Doptimize=ReleaseFast"
+  sh "zig", "build", "-Doptimize=ReleaseFast"
   fix_macos_install_names if RUBY_PLATFORM.include?("darwin")
 end
 
@@ -57,12 +52,18 @@ def fix_macos_install_names
       sh "install_name_tool", "-delete_rpath", rpath, bundle if rpath.start_with?("/")
     end
 
+    # Add a portable rpath so the bundle can locate libruby on any host.
+    unless rpaths.include?("@executable_path/../lib")
+      sh "install_name_tool", "-add_rpath", "@executable_path/../lib", bundle
+    end
+
     sh "codesign", "--sign", "-", "--force", bundle
 
     final_libs = `otool -L #{bundle.shellescape}`
     leaked = final_libs.lines.find do |l|
       next false unless l.start_with?("\t/")
       next false if l.match?(%r{^\t/usr/lib/}) || l.match?(%r{^\t/System/})
+
       true
     end
     raise "Non-portable dylib in #{bundle}: #{leaked.strip}\n\n#{final_libs}" if leaked
@@ -73,26 +74,26 @@ def fix_macos_install_names
   end
 end
 
-ZIG_VERSION = "0.15.2"
+ZIG_VERSION = "0.16.0"
 
 # Ruby versions to compile for. Full version must exactly match the directory
 # name inside the RCD image: docker run --rm <image> ls /usr/local/rake-compiler/ruby/x86_64-linux-gnu/
 RUBY_CROSS_VERSIONS = [
   {full: "3.4.8", api: "3.4.0"},
   {full: "4.0.0", api: "4.0.0"}
-]
+].freeze
 
 # [gem_platform_name, rcd_platform (image suffix), zig_target_triple]
 # rcd_platform also names the dir inside the RCD container *most* of the time;
 # scripts/rcd_build.sh discovers the actual path because the x86_64-musl image
 # uses x86_64-unknown-linux-musl as its top dir.
 LINUX_PLATFORMS = [
-  ["x86_64-linux", "x86_64-linux-gnu", "x86_64-linux-gnu"],
-  ["aarch64-linux", "aarch64-linux-gnu", "aarch64-linux-gnu"],
-  ["x86_64-linux-musl", "x86_64-linux-musl", "x86_64-linux-musl"],
-  ["aarch64-linux-musl", "aarch64-linux-musl", "aarch64-linux-musl"]
-]
-DARWIN_PLATFORMS = %w[arm64-darwin]
+  %w[x86_64-linux x86_64-linux-gnu x86_64-linux-gnu],
+  %w[aarch64-linux aarch64-linux-gnu aarch64-linux-gnu],
+  %w[x86_64-linux-musl x86_64-linux-musl x86_64-linux-musl],
+  %w[aarch64-linux-musl aarch64-linux-musl aarch64-linux-musl]
+].freeze
+DARWIN_PLATFORMS = %w[arm64-darwin].freeze
 ALL_PLATFORMS = LINUX_PLATFORMS.map(&:first) + DARWIN_PLATFORMS
 
 # Cross-compile tasks
@@ -103,11 +104,11 @@ namespace :cross do
       task gem_platform do
         require "rake_compiler_dock"
 
-        build_cmds = RUBY_CROSS_VERSIONS.map { |r|
+        build_cmds = RUBY_CROSS_VERSIONS.map do |r|
           "RUBY_FULL_VERSION=#{r[:full]} RUBY_API_VERSION=#{r[:api]} " \
           "RCD_PLATFORM=#{rcd_platform} TARGET_TRIPLE=#{zig_triple} " \
           "ZIG_VERSION=#{ZIG_VERSION} bash scripts/rcd_build.sh"
-        }.join(" && ")
+        end.join(" && ")
 
         RakeCompilerDock.sh(build_cmds, platform: rcd_platform)
       end
@@ -132,7 +133,7 @@ def build_platform_gem(platform)
 
   spec = Gem::Specification.load("carbon_fiber.gemspec")
   spec.platform = Gem::Platform.new(platform)
-  spec.extensions = []  # pre-built; no compilation on install
+  spec.extensions = [] # pre-built; no compilation on install
 
   ext = platform.include?("darwin") ? "bundle" : "so"
   files = Dir["lib/**/*.rb"].reject { |f| File.directory?(f) } +
@@ -210,4 +211,4 @@ task :lint do
 end
 
 desc "Compile native extension and run specs"
-task default: [:compile, :spec]
+task default: %i[compile spec]

benchmarks/Dockerfile

@@ -1,6 +1,6 @@
 FROM ruby:4.0.2-bookworm
 
-ARG ZIG_VERSION=0.15.2
+ARG ZIG_VERSION=0.16.0
 
 ENV PATH=/opt/zig:$PATH
 

benchmarks/README.md

@@ -46,7 +46,7 @@ Orchestrator. Spawns each (target, workload) combination as an isolated subproce
 Usage: bench [options]
     -t, --targets LIST     Comma-separated target IDs (default: all)
     -w, --workloads LIST   Comma-separated workload IDs (default: all)
-    -r, --runs N           Number of runs per combo, takes median (default: 1)
+    -r, --runs N           Number of runs per combo, takes median (default: 3)
         --timeout N        Per-workload timeout in seconds (default: 30)
         --setup            Install target gems and exit
 ```
@@ -62,6 +62,8 @@ Output goes to stdout (the table). Progress and setup messages go to stderr.
 
 When exactly 2 targets are specified, a delta column shows the percentage change from the first target to the second.
 
+When `runs >= 2`, each cell is annotated with the coefficient of variation (e.g. `±2.5%`), and the delta column tags entries as `(noisy)` when the magnitude of the change is within either target's CV. Use this to tell genuine regressions/wins from run-to-run jitter.
+
 ### `benchmarks/bench_one`
 
 Worker subprocess. Not meant to be called directly. Loads one target scheduler, runs one workload, prints a single floating-point metric value to stdout.
@@ -163,7 +165,7 @@ Orchestrator for async benchmarks. Same CLI interface as `bench`.
 Usage: async_bench [options]
     -t, --targets LIST     Comma-separated target IDs: stock, carbon (default: all)
     -w, --workloads LIST   Comma-separated workload IDs (default: all)
-    -r, --runs N           Number of runs per combo, takes median (default: 1)
+    -r, --runs N           Number of runs per combo, takes median (default: 3)
         --timeout N        Per-workload timeout in seconds (default: 30)
 ```