Fix clang warnings. (#9447)

- static function in header. (which is marked as unused due to translation unit
visibility).
- Implicit copy operator is deprecated.
- Unused lambda capture.
- Moving a temporary variable prevents copy elision.

include/xgboost/base.h

@@ -271,10 +271,11 @@ class GradientPairInt64 {
   GradientPairInt64() = default;
 
   // Copy constructor if of same value type, marked as default to be trivially_copyable
-  GradientPairInt64(const GradientPairInt64 &g) = default;
+  GradientPairInt64(GradientPairInt64 const &g) = default;
+  GradientPairInt64 &operator=(GradientPairInt64 const &g) = default;
 
-  XGBOOST_DEVICE T GetQuantisedGrad() const { return grad_; }
+  XGBOOST_DEVICE [[nodiscard]] T GetQuantisedGrad() const { return grad_; }
-  XGBOOST_DEVICE T GetQuantisedHess() const { return hess_; }
+  XGBOOST_DEVICE [[nodiscard]] T GetQuantisedHess() const { return hess_; }
 
   XGBOOST_DEVICE GradientPairInt64 &operator+=(const GradientPairInt64 &rhs) {
     grad_ += rhs.grad_;
@@ -323,17 +324,6 @@ using omp_ulong = dmlc::omp_ulong;  // NOLINT
 using bst_omp_uint = dmlc::omp_uint;  // NOLINT
 /*! \brief Type used for representing version number in binary form.*/
 using XGBoostVersionT = int32_t;
-
-/*!
- * \brief define compatible keywords in g++
- *  Used to support g++-4.6 and g++4.7
- */
-#if DMLC_USE_CXX11 && defined(__GNUC__) && !defined(__clang_version__)
-#if __GNUC__ == 4 && __GNUC_MINOR__ < 8
-#define override
-#define final
-#endif  // __GNUC__ == 4 && __GNUC_MINOR__ < 8
-#endif  // DMLC_USE_CXX11 && defined(__GNUC__) && !defined(__clang_version__)
 }  // namespace xgboost
 
 #endif  // XGBOOST_BASE_H_

src/common/math.h

@@ -134,12 +134,6 @@ inline float LogSum(Iterator begin, Iterator end) {
   return mx + std::log(sum);
 }
 
-// comparator functions for sorting pairs in descending order
-inline static bool CmpFirst(const std::pair<float, unsigned> &a,
-                            const std::pair<float, unsigned> &b) {
-  return a.first > b.first;
-}
-
 // Redefined here to workaround a VC bug that doesn't support overloading for integer
 // types.
 template <typename T>

src/data/iterative_dmatrix.cu

@@ -114,7 +114,7 @@ void IterativeDMatrix::InitFromCUDA(Context const* ctx, BatchParam const& p,
   this->info_.num_row_ = accumulated_rows;
   this->info_.num_nonzero_ = nnz;
 
-  auto init_page = [this, &proxy, &cuts, row_stride, accumulated_rows, get_device]() {
+  auto init_page = [this, &cuts, row_stride, accumulated_rows, get_device]() {
     if (!ellpack_) {
       // Should be put inside the while loop to protect against empty batch.  In
       // that case device id is invalid.

src/metric/rank_metric.cc

@@ -68,7 +68,8 @@ struct EvalAMS : public MetricNoCache {
     const auto &h_preds = preds.ConstHostVector();
     common::ParallelFor(ndata, ctx_->Threads(),
                         [&](bst_omp_uint i) { rec[i] = std::make_pair(h_preds[i], i); });
-    common::Sort(ctx_, rec.begin(), rec.end(), common::CmpFirst);
+    common::Sort(ctx_, rec.begin(), rec.end(),
+                 [](auto const& l, auto const& r) { return l.first > r.first; });
     auto ntop = static_cast<unsigned>(ratio_ * ndata);
     if (ntop == 0) ntop = ndata;
     const double br = 10.0;

src/predictor/gpu_predictor.cu

@@ -344,7 +344,7 @@ class DeviceModel {
     dh::safe_cuda(cudaSetDevice(gpu_id));
 
     // Copy decision trees to device
-    tree_segments = std::move(HostDeviceVector<size_t>({}, gpu_id));
+    tree_segments = HostDeviceVector<size_t>({}, gpu_id);
     auto& h_tree_segments = tree_segments.HostVector();
     h_tree_segments.reserve((tree_end - tree_begin) + 1);
     size_t sum = 0;
@@ -354,10 +354,8 @@ class DeviceModel {
       h_tree_segments.push_back(sum);
     }
 
-    nodes = std::move(HostDeviceVector<RegTree::Node>(h_tree_segments.back(), RegTree::Node(),
+    nodes = HostDeviceVector<RegTree::Node>(h_tree_segments.back(), RegTree::Node(), gpu_id);
-                                                      gpu_id));
+    stats = HostDeviceVector<RTreeNodeStat>(h_tree_segments.back(), RTreeNodeStat(), gpu_id);
-    stats = std::move(HostDeviceVector<RTreeNodeStat>(h_tree_segments.back(),
-                                                      RTreeNodeStat(), gpu_id));
     auto d_nodes = nodes.DevicePointer();
     auto d_stats = stats.DevicePointer();
     for (auto tree_idx = tree_begin; tree_idx < tree_end; tree_idx++) {
@@ -371,7 +369,7 @@ class DeviceModel {
           sizeof(RTreeNodeStat) * src_stats.size(), cudaMemcpyDefault));
     }
 
-    tree_group = std::move(HostDeviceVector<int>(model.tree_info.size(), 0, gpu_id));
+    tree_group = HostDeviceVector<int>(model.tree_info.size(), 0, gpu_id);
     auto& h_tree_group = tree_group.HostVector();
     std::memcpy(h_tree_group.data(), model.tree_info.data(), sizeof(int) * model.tree_info.size());
 
@@ -435,7 +433,7 @@ struct ShapSplitCondition {
   bool is_missing_branch;
 
   // Does this instance flow down this path?
-  XGBOOST_DEVICE bool EvaluateSplit(float x) const {
+  [[nodiscard]] XGBOOST_DEVICE bool EvaluateSplit(float x) const {
     // is nan
     if (isnan(x)) {
       return is_missing_branch;