From 25d232e4a417dd03ea2be69fb1ffad0d36f165f0 Mon Sep 17 00:00:00 2001
From: jdymitarai <o10040115@gmail.com>
Date: Thu, 8 Oct 2026 10:42:28 +0900
Subject: [PATCH 1/2] fix(mutex): hold lock during notify_all in Barrier::wait
 to prevent teardown race (fixes #1672)

---
 src/mutex.h             |  7 ++-----
 test/benchmark_gtest.cc | 46 +++++++++++++++++++++++++++++++++++++++++
 2 files changed, 48 insertions(+), 5 deletions(-)

diff --git a/src/mutex.h b/src/mutex.h
index b6b4b71ab0..a1e6f2acf5 100644
--- a/src/mutex.h
+++ b/src/mutex.h
@@ -117,11 +117,8 @@ class Barrier {
 
   // Called by each thread
   bool wait() EXCLUDES(lock_) {
-    bool last_thread = false;
-    {
-      MutexLock ml(lock_);
-      last_thread = createBarrier(ml);
-    }
+    MutexLock ml(lock_);
+    bool last_thread = createBarrier(ml);
     if (last_thread) phase_condition_.notify_all();
     return last_thread;
   }
diff --git a/test/benchmark_gtest.cc b/test/benchmark_gtest.cc
index bd6304ade2..bcc986cbad 100644
--- a/test/benchmark_gtest.cc
+++ b/test/benchmark_gtest.cc
@@ -1,8 +1,10 @@
 #include <map>
 #include <string>
+#include <thread>
 #include <vector>
 
 #include "../src/benchmark_register.h"
+#include "../src/mutex.h"
 #include "benchmark/benchmark.h"
 #include "gmock/gmock.h"
 #include "gtest/gtest.h"
@@ -250,6 +252,50 @@ TEST(TimingTest, PauseAndResumeInLoopReportSaneTime) {
   EXPECT_LT(runs[0].cpu_accumulated_time, 1.0);
 }
 
+TEST(BarrierTest, SingleThreadedWaitAndTeardown) {
+  for (int i = 0; i < 100; ++i) {
+    auto barrier = std::make_unique<Barrier>(1);
+    EXPECT_TRUE(barrier->wait());
+  }
+}
+
+TEST(BarrierTest, MultiThreadedWaitAndTeardown) {
+  constexpr int kNumThreads = 8;
+  for (int i = 0; i < 100; ++i) {
+    auto barrier = std::make_unique<Barrier>(kNumThreads);
+    std::vector<std::thread> threads;
+    threads.reserve(kNumThreads);
+    for (int t = 0; t < kNumThreads; ++t) {
+      threads.emplace_back([&barrier]() {
+        barrier->wait();
+      });
+    }
+    for (auto& thread : threads) {
+      thread.join();
+    }
+  }
+}
+
+TEST(BarrierTest, MultiThreadedRemoveThreadAndTeardown) {
+  constexpr int kNumThreads = 4;
+  for (int i = 0; i < 100; ++i) {
+    auto barrier = std::make_unique<Barrier>(kNumThreads);
+    std::vector<std::thread> threads;
+    threads.reserve(kNumThreads);
+    for (int t = 0; t < kNumThreads - 1; ++t) {
+      threads.emplace_back([&barrier]() {
+        barrier->wait();
+      });
+    }
+    threads.emplace_back([&barrier]() {
+      barrier->removeThread();
+    });
+    for (auto& thread : threads) {
+      thread.join();
+    }
+  }
+}
+
 }  // namespace
 }  // namespace internal
 }  // namespace benchmark

From 33141b3f0a27aa94522bcca756b021157d93afd5 Mon Sep 17 00:00:00 2001
From: jdymitarai <o10040115@gmail.com>
Date: Fri, 9 Oct 2026 22:09:53 +0900
Subject: [PATCH 2/2] style: fix clang-format formatting in BarrierTest

---
 test/benchmark_gtest.cc | 12 +++---------
 1 file changed, 3 insertions(+), 9 deletions(-)

diff --git a/test/benchmark_gtest.cc b/test/benchmark_gtest.cc
index bcc986cbad..d32ff4dd9e 100644
--- a/test/benchmark_gtest.cc
+++ b/test/benchmark_gtest.cc
@@ -266,9 +266,7 @@ TEST(BarrierTest, MultiThreadedWaitAndTeardown) {
     std::vector<std::thread> threads;
     threads.reserve(kNumThreads);
     for (int t = 0; t < kNumThreads; ++t) {
-      threads.emplace_back([&barrier]() {
-        barrier->wait();
-      });
+      threads.emplace_back([&barrier]() { barrier->wait(); });
     }
     for (auto& thread : threads) {
       thread.join();
@@ -283,13 +281,9 @@ TEST(BarrierTest, MultiThreadedRemoveThreadAndTeardown) {
     std::vector<std::thread> threads;
     threads.reserve(kNumThreads);
     for (int t = 0; t < kNumThreads - 1; ++t) {
-      threads.emplace_back([&barrier]() {
-        barrier->wait();
-      });
+      threads.emplace_back([&barrier]() { barrier->wait(); });
     }
-    threads.emplace_back([&barrier]() {
-      barrier->removeThread();
-    });
+    threads.emplace_back([&barrier]() { barrier->removeThread(); });
     for (auto& thread : threads) {
       thread.join();
     }