Fix build and C++ tests for FreeBSD

.github/workflows/freebsd.yml

@@ -0,0 +1,33 @@
+name: FreeBSD
+
+on: [push, pull_request]
+
+permissions:
+  contents: read # to fetch code (actions/checkout)
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    name: A job to run test in FreeBSD
+    steps:
+    - uses: actions/checkout@v4
+      with:
+        submodules: 'true'
+    - name: Test in FreeBSD
+      id: test
+      uses: vmactions/freebsd-vm@v1
+      with:
+        usesh: true
+        prepare: |
+          pkg install -y cmake git ninja googletest
+
+        run: |
+          mkdir build
+          cd build
+          cmake .. -GNinja -DGOOGLE_TEST=ON
+          ninja -v
+          ./testxgboost

include/xgboost/collective/poll_utils.h

@@ -77,6 +77,8 @@ namespace utils {
 
 template <typename PollFD>
 int PollImpl(PollFD* pfd, int nfds, std::chrono::seconds timeout) noexcept(true) {
+  // For Windows and Linux, negative timeout means infinite timeout. For freebsd,
+  // INFTIM(-1) should be used instead.
 #if defined(_WIN32)
 
 #if IS_MINGW()
@@ -87,7 +89,7 @@ int PollImpl(PollFD* pfd, int nfds, std::chrono::seconds timeout) noexcept(true)
 #endif  // IS_MINGW()
 
 #else
-  return poll(pfd, nfds, std::chrono::milliseconds(timeout).count());
+  return poll(pfd, nfds, timeout.count() < 0 ? -1 : std::chrono::milliseconds(timeout).count());
 #endif  // IS_MINGW()
 }
 

src/c_api/coll_c_api.cc

@@ -75,8 +75,11 @@ using CollAPIThreadLocalStore = dmlc::ThreadLocalStore<CollAPIEntry>;
 
 void WaitImpl(TrackerHandleT *ptr, std::chrono::seconds timeout) {
   constexpr std::int64_t kDft{collective::DefaultTimeoutSec()};
-  std::chrono::seconds wait_for{collective::HasTimeout(timeout) ? std::min(kDft, timeout.count())
-                                                                : kDft};
+  std::int64_t timeout_clipped = kDft;
+  if (collective::HasTimeout(timeout)) {
+    timeout_clipped = std::min(kDft, static_cast<std::int64_t>(timeout.count()));
+  }
+  std::chrono::seconds wait_for{timeout_clipped};
 
   common::Timer timer;
   timer.Start();
@@ -171,7 +174,19 @@ XGB_DLL int XGTrackerFree(TrackerHandle handle) {
   common::Timer timer;
   timer.Start();
   // Make sure no one else is waiting on the tracker.
-  while (!ptr->first.unique()) {
+
+  // Quote from https://en.cppreference.com/w/cpp/memory/shared_ptr/use_count#Notes:
+  //
+  // In multithreaded environment, `use_count() == 1` does not imply that the object is
+  // safe to modify because accesses to the managed object by former shared owners may not
+  // have completed, and because new shared owners may be introduced concurrently.
+  //
+  // - We don't have the first case since we never access the raw pointer.
+  //
+  // - We don't hve the second case for most of the scenarios since tracker is an unique
+  //   object, if the free function is called before another function calls, it's likely
+  //   to be a bug in the user code. The use_count should only decrease in this function.
+  while (ptr->first.use_count() != 1) {
     auto ela = timer.Duration().count();
     if (collective::HasTimeout(ptr->first->Timeout()) && ela > ptr->first->Timeout().count()) {
       LOG(WARNING) << "Time out " << ptr->first->Timeout().count()

src/collective/socket.cc

@@ -22,10 +22,12 @@ namespace xgboost::collective {
 SockAddress MakeSockAddress(StringView host, in_port_t port) {
   struct addrinfo hints;
   std::memset(&hints, 0, sizeof(hints));
-  hints.ai_protocol = SOCK_STREAM;
+  hints.ai_socktype = SOCK_STREAM;
   struct addrinfo *res = nullptr;
   int sig = getaddrinfo(host.c_str(), nullptr, &hints, &res);
   if (sig != 0) {
+    LOG(FATAL) << "Failed to get addr info for: " << host
+      << ", error: " << gai_strerror(sig);
     return {};
   }
   if (res->ai_family == static_cast<std::int32_t>(SockDomain::kV4)) {

tests/cpp/collective/test_worker.h

@@ -105,7 +105,7 @@ inline Json MakeTrackerConfig(std::string host, std::int32_t n_workers,
   config["port"] = Integer{0};
   config["n_workers"] = Integer{n_workers};
   config["sortby"] = Integer{static_cast<std::int32_t>(Tracker::SortBy::kHost)};
-  config["timeout"] = timeout.count();
+  config["timeout"] = static_cast<std::int64_t>(timeout.count());
   return config;
 }
 

tests/cpp/common/test_random.cc

@@ -68,14 +68,20 @@ TEST(ColumnSampler, GPUTest) {
 // Test if different threads using the same seed produce the same result
 TEST(ColumnSampler, ThreadSynchronisation) {
   Context ctx;
-  const int64_t num_threads = 100;
+  // NOLINTBEGIN(clang-analyzer-deadcode.DeadStores)
+#if defined(__linux__)
+  std::int64_t const n_threads = std::thread::hardware_concurrency() * 128;
+#else
+  std::int64_t const n_threads = std::thread::hardware_concurrency();
+#endif
+  // NOLINTEND(clang-analyzer-deadcode.DeadStores)
   int n = 128;
   size_t iterations = 10;
   size_t levels = 5;
   std::vector<bst_feature_t> reference_result;
   std::vector<float> feature_weights;
   bool success = true; // Cannot use google test asserts in multithreaded region
-#pragma omp parallel num_threads(num_threads)
+#pragma omp parallel num_threads(n_threads)
   {
     for (auto j = 0ull; j < iterations; j++) {
       ColumnSampler cs(j);

tests/cpp/test_cache.cc

@@ -59,7 +59,11 @@ TEST(DMatrixCache, MultiThread) {
   std::size_t constexpr kRows = 2, kCols = 1, kCacheSize = 3;
   auto p_fmat = RandomDataGenerator(kRows, kCols, 0).GenerateDMatrix();
 
-  auto n = std::thread::hardware_concurrency() * 128u;
+#if defined(__linux__)
+  auto const n = std::thread::hardware_concurrency() * 128;
+#else
+  auto const n = std::thread::hardware_concurrency();
+#endif
   CHECK_NE(n, 0);
   std::vector<std::shared_ptr<CacheForTest>> results(n);
 

tests/cpp/test_learner.cc

@@ -267,8 +267,14 @@ TEST(Learner, MultiThreadedPredict) {
   learner->Configure();
 
   std::vector<std::thread> threads;
-  for (uint32_t thread_id = 0;
-       thread_id < 2 * std::thread::hardware_concurrency(); ++thread_id) {
+
+#if defined(__linux__)
+  auto n_threads = std::thread::hardware_concurrency() * 4u;
+#else
+  auto n_threads = std::thread::hardware_concurrency();
+#endif
+
+  for (decltype(n_threads) thread_id = 0; thread_id < n_threads; ++thread_id) {
     threads.emplace_back([learner, p_data] {
       size_t constexpr kIters = 10;
       auto &entry = learner->GetThreadLocal().prediction_entry;