[Simulation.Core] WorkerThread: fix memory leak/resource retention

Sofa/framework/Simulation/Core/test/CMakeLists.txt

@@ -7,6 +7,7 @@ set(SOURCE_FILES
     MappingGraph_test.cpp
     ParallelForEach_test.cpp
     RequiredPlugin_test.cpp
+    TaskDestructorLeak_test.cpp
     SceneCheckRegistry_test.cpp
     Simulation_test.cpp
     ParallelSparseMatrixProduct_test.cpp

Sofa/framework/Simulation/Core/test/TaskDestructorLeak_test.cpp

@@ -0,0 +1,128 @@
+/******************************************************************************
+*                 SOFA, Simulation Open-Framework Architecture                *
+*                    (c) 2006 INRIA, USTL, UJF, CNRS, MGH                     *
+*                                                                             *
+* This program is free software; you can redistribute it and/or modify it     *
+* under the terms of the GNU Lesser General Public License as published by    *
+* the Free Software Foundation; either version 2.1 of the License, or (at     *
+* your option) any later version.                                             *
+*                                                                             *
+* This program is distributed in the hope that it will be useful, but WITHOUT *
+* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or       *
+* FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License *
+* for more details.                                                           *
+*                                                                             *
+* You should have received a copy of the GNU Lesser General Public License    *
+* along with this program. If not, see <http://www.gnu.org/licenses/>.        *
+*******************************************************************************
+* Authors: The SOFA Team and external contributors (see Authors.txt)          *
+*                                                                             *
+* Contact information: contact@sofa-framework.org                             *
+******************************************************************************/
+#include <gtest/gtest.h>
+#include <sofa/simulation/task/CpuTask.h>
+#include <sofa/simulation/task/CpuTaskStatus.h>
+#include <sofa/simulation/task/DefaultTaskScheduler.h>
+#include <sofa/simulation/task/MainTaskSchedulerFactory.h>
+
+#include <atomic>
+#include <memory>
+
+namespace sofa
+{
+
+// Reproduction for a leak in WorkerThread::runTask. After running a task that
+// returns MemoryAlloc::Dynamic, the framework calls Task::operator delete to
+// release the memory but does NOT call the task's destructor. From
+// WorkerThread::runTask:
+//
+//   if (task->run() & Task::MemoryAlloc::Dynamic) {
+//       // pooled memory: call destructor and free
+//       //task->~Task();                                    <-- commented out
+//       task->operator delete(task, sizeof(*task));
+//   }
+//
+// As a result, any task with non-trivially-destructible members (std::function,
+// std::shared_ptr, std::vector, ...) leaks those members' resources every time
+// it runs. The std::function-based addTask(Status&, lambda) overload always
+// triggers this because CallableTask wraps the lambda in a std::function.
+//
+// We demonstrate the leak with a custom CpuTask holding a std::shared_ptr.
+// If the destructor ran, every task copy of the shared_ptr would release a
+// reference and the original's use_count would return to 1 after the burst.
+// On the buggy code, use_count stays at 1 + (number of dispatched tasks).
+
+namespace
+{
+
+class SharedPtrHoldingTask : public simulation::CpuTask
+{
+public:
+    SharedPtrHoldingTask(simulation::CpuTask::Status* status,
+                         std::shared_ptr<int> resource,
+                         std::atomic<int>* counter)
+        : simulation::CpuTask(status)
+        , m_resource(std::move(resource))
+        , m_counter(counter)
+    {}
+
+    sofa::simulation::Task::MemoryAlloc run() final
+    {
+        // Touch the resource so the compiler can't elide the member.
+        if (m_resource)
+        {
+            m_counter->fetch_add(*m_resource, std::memory_order_relaxed);
+        }
+        return sofa::simulation::Task::MemoryAlloc::Dynamic;
+    }
+
+private:
+    std::shared_ptr<int> m_resource;
+    std::atomic<int>* m_counter;
+};
+
+} // namespace
+
+// Dispatch many tasks, each holding a copy of the same shared_ptr. After
+// workUntilDone, all task instances must have been destroyed; the only
+// remaining holder is the test's local `resource`, so use_count must be 1.
+//
+// On the buggy code, the destructors are skipped and use_count equals
+// 1 + kNumTasks.
+TEST(TaskDestructorLeak, SharedPtrTasksReleaseTheirReference)
+{
+    constexpr int kNumTasks = 64;
+
+    auto* scheduler = simulation::MainTaskSchedulerFactory::createInRegistry(
+        simulation::DefaultTaskScheduler::name());
+    ASSERT_NE(scheduler, nullptr);
+
+    scheduler->init(0);
+    if (scheduler->getThreadCount() < 2)
+    {
+        GTEST_SKIP() << "scheduler has fewer than 2 threads; skipping";
+    }
+
+    auto resource = std::make_shared<int>(1);
+    std::atomic<int> counter { 0 };
+
+    {
+        simulation::CpuTaskStatus status;
+        for (int i = 0; i < kNumTasks; ++i)
+        {
+            // Each task holds its own shared_ptr copy.
+            scheduler->addTask(new SharedPtrHoldingTask(&status, resource, &counter));
+        }
+        scheduler->workUntilDone(&status);
+    }
+
+    EXPECT_EQ(counter.load(std::memory_order_relaxed), kNumTasks);
+    // After all tasks have been disposed of, only `resource` should hold the
+    // shared_ptr. If the framework skipped the destructor, every task's copy
+    // is leaked and use_count == 1 + kNumTasks.
+    EXPECT_EQ(resource.use_count(), 1);
+
+    scheduler->stop();
+}
+
+} // namespace sofa