Run knucleotide, gvl_release_acquire and railsbench on Windows

Windows has no fork, no /dev/zero and no shebang execution of bin/rails. knucleotide computes its workers sequentially there.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Author: hsbt

benchmarks/gvl_release_acquire/benchmark.rb

@@ -1,8 +1,18 @@
 require_relative "../../harness/loader"
 
+# Windows has no /dev/zero, so read the zeros from a file there.
+ZERO_PATH = if File.exist?("/dev/zero")
+  "/dev/zero"
+else
+  require "tmpdir"
+  File.join(Dir.tmpdir, "ruby-bench-zero").tap do |path|
+    File.binwrite(path, "\0" * 1_000_000) unless File.size?(path) == 1_000_000
+  end
+end.freeze
+
 run_benchmark(5) do |num_rs, ractor_args|
-  output = File.open("/dev/null", "wb")
-  input = File.open("/dev/zero", "rb")
+  output = File.open(File::NULL, "wb")
+  input = File.open(ZERO_PATH, "rb")
   100_000.times do
     output.write(input.read(10))
   end

benchmarks/knucleotide/benchmark.rb

@@ -37,6 +37,13 @@ def find_seq(seq, s)
 
 class Worker
   def initialize(&block)
+    # Platforms without fork (Windows) compute the result sequentially in this process,
+    # so their timings are not comparable with the parallel ones.
+    unless Process.respond_to?(:fork)
+      @result = yield
+      return
+    end
+
     @r, @w = IO.pipe
     @p = Process.fork do
       @r.close
@@ -47,6 +54,8 @@ def initialize(&block)
   end
 
   def result
+    return @result unless @p
+
     ret = @r.read
     @r.close
     Process.wait(@p)

benchmarks/railsbench/benchmark.rb

@@ -8,7 +8,7 @@
 # and this app's db/seeds.rb will delete and repopulate
 # the database, so rows shouldn't accumulate.
 Dir.chdir __dir__
-use_gemfile extra_setup_cmd: "bin/rails db:migrate db:seed"
+use_gemfile extra_setup_cmd: "ruby bin/rails db:migrate db:seed" # Windows cannot run bin/rails by its shebang
 
 require_relative 'config/environment'