double free or corruption (out) when running test data

[wwtinwhu]

Hi, @Jerry-locker !

I followed the same environment and third party version as listed in the README. However, it died with a double free problem.

At first I thought it was because pcl did not turn off MARCH_NATIVE when compiling. After I turned off this option and recompiled, the memory error still occurred. Any suggestions? Thanks!

[zyj360507603]

+1

[tsuibeyond]

+1

[cxjbill]

+1

[lojmkjd]

有人解决了这个问题吗

[Geyuwen]

是内存泄漏的问题，gdb调试一下
修改代码：
#将/root/catkin_coco/src/Coco-LIC/src/odom/factor/analytic_diff/marginalization_factor.cpp修改如下：
// MarginalizationInfo::~MarginalizationInfo() {
// for (auto it = parameter_block_data.begin(); it != parameter_block_data.end();
// ++it) {
// delete it->second;
// }

// for (int i = 0; i < (int)factors.size(); i++) {
// delete factors[i]->cost_function;
// delete factors[i];
// }
// }

MarginalizationInfo::~MarginalizationInfo() {
// 1. 移除对 parameter_block_data 中数据的删除
// 因为这些数据很可能就是 Ceres Problem 正在管理的参数块
// for (auto it = parameter_block_data.begin(); it != parameter_block_data.end();
// ++it) {
// delete it->second; // <-- 移除这行，或者检查 it->second 是否是你自己 new 的副本，通常不是。
// }
// 如果 parameter_block_data 存储的是裸指针，且它们指向的是 Ceres 管理的内存，则不应删除。
// 如果 parameter_block_data 存储的是你自己的深拷贝数据，且你负责它们的生命周期，那这行可能是正确的。
// 但根据上下文，很可能它们指向的是优化参数，所以这行是错的。

// 2. 移除对 cost_function 的删除
// 因为 ceres::Problem 已经拥有并会删除这些 CostFunction。
for (int i = 0; i < (int)factors.size(); i++) {
// 确保 factors[i] 是 ResidualBlockInfo* 类型，且其内部的 cost_function_ 是裸指针
// ceres::CostFunction *cost_function = factors[i]->cost_function; // 取出指针
// delete cost_function; // <-- 移除这行！
// 确保 ResidualBlockInfo 的析构函数也不删除 CostFunction
delete factors[i]; // <-- 保留这行！因为 MarginalizationInfo 拥有 ResidualBlockInfo 对象
}
factors.clear(); // 清空 vector，确保没有野指针
parameter_block_data.clear(); // 清空 map
}
#将/root/catkin_coco/src/Coco-LIC/src/camera/rgb_map/pointcloud_rgbd.cpp修改如下：
Global_map::Global_map( int if_start_service ): m_should_terminate_thread(false)
{
m_mutex_pts_vec = std::make_shared< std::mutex >();
m_mutex_img_pose_for_projection = std::make_shared< std::mutex >();
m_mutex_recent_added_list = std::make_shared< std::mutex >();
m_mutex_rgb_pts_in_recent_hitted_boxes = std::make_shared< std::mutex >();
m_mutex_m_box_recent_hitted = std::make_shared< std::mutex >();
m_mutex_pts_last_visited = std::make_shared< std::mutex >();
// Allocate memory for pointclouds
if ( Common_tools::get_total_phy_RAM_size_in_GB() < 12 )
{
scope_color( ANSI_COLOR_RED_BOLD );
std::this_thread::sleep_for( std::chrono::seconds( 1 ) );
cout << "+++++++++++++++++++++++++++++++++++++++++++++++++++" << endl;
cout << "I have detected your physical memory smaller than 12GB (currently: " << Common_tools::get_total_phy_RAM_size_in_GB()
<< "GB). I recommend you to add more physical memory for improving the overall performance of R3LIVE." << endl;
cout << "+++++++++++++++++++++++++++++++++++++++++++++++++++" << endl;
std::this_thread::sleep_for( std::chrono::seconds( 5 ) );
m_rgb_pts_vec.reserve( 1e8 );
}
else
{
m_rgb_pts_vec.reserve( 1e9 );
}
// m_rgb_pts_in_recent_visited_voxels.reserve( 1e6 );
if ( if_start_service )
{
m_thread_service = std::make_shared< std::thread >( &Global_map::service_refresh_pts_for_projection, this );
}
}
// 在 pointcloud_rgbd.cpp 中
Global_map::~Global_map() {
// 确保线程在析构前完成其工作
if (m_thread_service) // 首先检查 shared_ptr 是否有效（不为 nullptr）
{
if (m_thread_service->joinable())
{
std::cout << "Global_map destructor: Signaling thread to terminate..." << std::endl; // 调试输出
m_should_terminate_thread.store(true); // 设置终止标志，原子地写入 true

        // **可选：给线程一个短暂的时间来检测到信号并开始退出过程**
        // std::this_thread::sleep_for(std::chrono::milliseconds(10)); 

        std::cout << "Global_map destructor: Joining m_thread_service..." << std::endl; // 调试输出
        m_thread_service->join(); // 等待线程完成执行
        std::cout << "Global_map destructor: m_thread_service joined." << std::endl; // 调试输出
    } else {
        std::cout << "Global_map destructor: m_thread_service not joinable." << std::endl; // 调试输出
    }
} else {
    std::cout << "Global_map destructor: m_thread_service is null (thread was not started or already reset)." << std::endl; // 调试输出
}

}

void Global_map::service_refresh_pts_for_projection()
{
eigen_q last_pose_q = eigen_q::Identity();
Common_tools::Timer timer;
std::shared_ptr< Image_frame > img_for_projection = std::make_shared< Image_frame >();
while (!m_should_terminate_thread.load())
{
std::this_thread::sleep_for(std::chrono::milliseconds(1));
if (m_should_terminate_thread.load()) {
break; // 如果标志已被设置，立即退出循环
}
m_mutex_img_pose_for_projection->lock();

    *img_for_projection = m_img_for_projection;
    m_mutex_img_pose_for_projection->unlock();
    if (img_for_projection->m_img_cols == 0 || img_for_projection->m_img_rows == 0)
    {
        continue;
    }

    if (img_for_projection->m_frame_idx == m_updated_frame_index)
    {
        continue;
    }
    timer.tic(" ");
    std::shared_ptr<std::vector<std::shared_ptr<RGB_pts>>> pts_rgb_vec_for_projection = std::make_shared<std::vector<std::shared_ptr<RGB_pts>>>();
    if (m_if_get_all_pts_in_boxes_using_mp)
    {
        std::vector<std::shared_ptr<RGB_pts>>  pts_in_recent_hitted_boxes;
        pts_in_recent_hitted_boxes.reserve(1e6);
        std::unordered_set< std::shared_ptr< RGB_Voxel> > boxes_recent_hitted;
        m_mutex_m_box_recent_hitted->lock();
        boxes_recent_hitted = m_voxels_recent_visited;
        m_mutex_m_box_recent_hitted->unlock();

        // get_all_pts_in_boxes(boxes_recent_hitted, pts_in_recent_hitted_boxes);
        m_mutex_rgb_pts_in_recent_hitted_boxes->lock();
        // m_rgb_pts_in_recent_visited_voxels = pts_in_recent_hitted_boxes;
        m_mutex_rgb_pts_in_recent_hitted_boxes->unlock();
    }
    Eigen::aligned_vector<Eigen::Vector3d> fake_points;
    Eigen::aligned_vector<Eigen::Vector2d> fake_pixs;
    selection_points_for_projection(false, fake_points, fake_pixs, img_for_projection, pts_rgb_vec_for_projection.get(), nullptr, 10.0, 1);
    m_mutex_pts_vec->lock();
    m_pts_rgb_vec_for_projection = pts_rgb_vec_for_projection;
    m_updated_frame_index = img_for_projection->m_frame_idx;
    // cout << ANSI_COLOR_MAGENTA_BOLD << "Refresh pts_for_projection size = " << m_pts_rgb_vec_for_projection->size()
    //      << " | " << m_rgb_pts_vec.size()
    //      << ", cost time = " << timer.toc() << ANSI_COLOR_RESET << endl;
    m_mutex_pts_vec->unlock();
    last_pose_q = img_for_projection->m_pose_w2c_q;
}
 // **可选：添加一条日志，确认线程已退出**
std::cout << "Global_map::service_refresh_pts_for_projection() thread exiting gracefully." << std::endl;

}
#/root/catkin_coco/src/Coco-LIC/src/camera/rgb_map/pointcloud_rgbd.hpp中新增一行
struct Global_map
{
int m_map_major_version = R3LIVE_MAP_MAJOR_VERSION;
int m_map_minor_version = R3LIVE_MAP_MINOR_VERSION;
int m_if_get_all_pts_in_boxes_using_mp = 1;
std::vector< RGB_pt_ptr > m_rgb_pts_vec;
// std::vector< RGB_pt_ptr > m_rgb_pts_in_recent_visited_voxels;
std::shared_ptr< std::vector< RGB_pt_ptr> > m_pts_rgb_vec_for_projection = nullptr;
std::shared_ptr< std::mutex > m_mutex_pts_vec;
std::shared_ptr< std::mutex > m_mutex_recent_added_list;
std::shared_ptr< std::mutex > m_mutex_img_pose_for_projection;
std::shared_ptr< std::mutex > m_mutex_rgb_pts_in_recent_hitted_boxes;
std::shared_ptr< std::mutex > m_mutex_m_box_recent_hitted;
std::shared_ptr< std::mutex > m_mutex_pts_last_visited;
Image_frame m_img_for_projection;
double m_recent_visited_voxel_activated_time = 0.0;
bool m_in_appending_pts = 0;
int m_updated_frame_index = 0;
std::shared_ptr< std::thread > m_thread_service;
std::atomic m_should_terminate_thread;// <-- 新增这一行
int m_if_reload_init_voxel_and_hashed_pts = true;
//。。。后续代码

[Jerry-locker]

是内存泄漏的问题，gdb调试一下 修改代码： #将/root/catkin_coco/src/Coco-LIC/src/odom/factor/analytic_diff/marginalization_factor.cpp修改如下： // MarginalizationInfo::~MarginalizationInfo() { // for (auto it = parameter_block_data.begin(); it != parameter_block_data.end(); // ++it) { // delete it->second; // }

// for (int i = 0; i < (int)factors.size(); i++) { // delete factors[i]->cost_function; // delete factors[i]; // } // }

MarginalizationInfo::~MarginalizationInfo() { // 1. 移除对 parameter_block_data 中数据的删除 // 因为这些数据很可能就是 Ceres Problem 正在管理的参数块 // for (auto it = parameter_block_data.begin(); it != parameter_block_data.end(); // ++it) { // delete it->second; // <-- 移除这行，或者检查 it->second 是否是你自己 new 的副本，通常不是。 // } // 如果 parameter_block_data 存储的是裸指针，且它们指向的是 Ceres 管理的内存，则不应删除。 // 如果 parameter_block_data 存储的是你自己的深拷贝数据，且你负责它们的生命周期，那这行可能是正确的。 // 但根据上下文，很可能它们指向的是优化参数，所以这行是错的。

// 2. 移除对 cost_function 的删除 // 因为 ceres::Problem 已经拥有并会删除这些 CostFunction。 for (int i = 0; i < (int)factors.size(); i++) { // 确保 factors[i] 是 ResidualBlockInfo* 类型，且其内部的 cost_function_ 是裸指针 // ceres::CostFunction *cost_function = factors[i]->cost_function; // 取出指针 // delete cost_function; // <-- 移除这行！ // 确保 ResidualBlockInfo 的析构函数也不删除 CostFunction delete factors[i]; // <-- 保留这行！因为 MarginalizationInfo 拥有 ResidualBlockInfo 对象 } factors.clear(); // 清空 vector，确保没有野指针 parameter_block_data.clear(); // 清空 map } #将/root/catkin_coco/src/Coco-LIC/src/camera/rgb_map/pointcloud_rgbd.cpp修改如下： Global_map::Global_map( int if_start_service ): m_should_terminate_thread(false) { m_mutex_pts_vec = std::make_shared< std::mutex >(); m_mutex_img_pose_for_projection = std::make_shared< std::mutex >(); m_mutex_recent_added_list = std::make_shared< std::mutex >(); m_mutex_rgb_pts_in_recent_hitted_boxes = std::make_shared< std::mutex >(); m_mutex_m_box_recent_hitted = std::make_shared< std::mutex >(); m_mutex_pts_last_visited = std::make_shared< std::mutex >(); // Allocate memory for pointclouds if ( Common_tools::get_total_phy_RAM_size_in_GB() < 12 ) { scope_color( ANSI_COLOR_RED_BOLD ); std::this_thread::sleep_for( std::chrono::seconds( 1 ) ); cout << "+++++++++++++++++++++++++++++++++++++++++++++++++++" << endl; cout << "I have detected your physical memory smaller than 12GB (currently: " << Common_tools::get_total_phy_RAM_size_in_GB() << "GB). I recommend you to add more physical memory for improving the overall performance of R3LIVE." << endl; cout << "+++++++++++++++++++++++++++++++++++++++++++++++++++" << endl; std::this_thread::sleep_for( std::chrono::seconds( 5 ) ); m_rgb_pts_vec.reserve( 1e8 ); } else { m_rgb_pts_vec.reserve( 1e9 ); } // m_rgb_pts_in_recent_visited_voxels.reserve( 1e6 ); if ( if_start_service ) { m_thread_service = std::make_shared< std::thread >( &Global_map::service_refresh_pts_for_projection, this ); } } // 在 pointcloud_rgbd.cpp 中 Global_map::~Global_map() { // 确保线程在析构前完成其工作 if (m_thread_service) // 首先检查 shared_ptr 是否有效（不为 nullptr） { if (m_thread_service->joinable()) { std::cout << "Global_map destructor: Signaling thread to terminate..." << std::endl; // 调试输出 m_should_terminate_thread.store(true); // 设置终止标志，原子地写入 true

        // **可选：给线程一个短暂的时间来检测到信号并开始退出过程**
        // std::this_thread::sleep_for(std::chrono::milliseconds(10)); 

        std::cout << "Global_map destructor: Joining m_thread_service..." << std::endl; // 调试输出
        m_thread_service->join(); // 等待线程完成执行
        std::cout << "Global_map destructor: m_thread_service joined." << std::endl; // 调试输出
    } else {
        std::cout << "Global_map destructor: m_thread_service not joinable." << std::endl; // 调试输出
    }
} else {
    std::cout << "Global_map destructor: m_thread_service is null (thread was not started or already reset)." << std::endl; // 调试输出
}

}

void Global_map::service_refresh_pts_for_projection() { eigen_q last_pose_q = eigen_q::Identity(); Common_tools::Timer timer; std::shared_ptr< Image_frame > img_for_projection = std::make_shared< Image_frame >(); while (!m_should_terminate_thread.load()) { std::this_thread::sleep_for(std::chrono::milliseconds(1)); if (m_should_terminate_thread.load()) { break; // 如果标志已被设置，立即退出循环 } m_mutex_img_pose_for_projection->lock();

    *img_for_projection = m_img_for_projection;
    m_mutex_img_pose_for_projection->unlock();
    if (img_for_projection->m_img_cols == 0 || img_for_projection->m_img_rows == 0)
    {
        continue;
    }

    if (img_for_projection->m_frame_idx == m_updated_frame_index)
    {
        continue;
    }
    timer.tic(" ");
    std::shared_ptr<std::vector<std::shared_ptr<RGB_pts>>> pts_rgb_vec_for_projection = std::make_shared<std::vector<std::shared_ptr<RGB_pts>>>();
    if (m_if_get_all_pts_in_boxes_using_mp)
    {
        std::vector<std::shared_ptr<RGB_pts>>  pts_in_recent_hitted_boxes;
        pts_in_recent_hitted_boxes.reserve(1e6);
        std::unordered_set< std::shared_ptr< RGB_Voxel> > boxes_recent_hitted;
        m_mutex_m_box_recent_hitted->lock();
        boxes_recent_hitted = m_voxels_recent_visited;
        m_mutex_m_box_recent_hitted->unlock();

        // get_all_pts_in_boxes(boxes_recent_hitted, pts_in_recent_hitted_boxes);
        m_mutex_rgb_pts_in_recent_hitted_boxes->lock();
        // m_rgb_pts_in_recent_visited_voxels = pts_in_recent_hitted_boxes;
        m_mutex_rgb_pts_in_recent_hitted_boxes->unlock();
    }
    Eigen::aligned_vector<Eigen::Vector3d> fake_points;
    Eigen::aligned_vector<Eigen::Vector2d> fake_pixs;
    selection_points_for_projection(false, fake_points, fake_pixs, img_for_projection, pts_rgb_vec_for_projection.get(), nullptr, 10.0, 1);
    m_mutex_pts_vec->lock();
    m_pts_rgb_vec_for_projection = pts_rgb_vec_for_projection;
    m_updated_frame_index = img_for_projection->m_frame_idx;
    // cout << ANSI_COLOR_MAGENTA_BOLD << "Refresh pts_for_projection size = " << m_pts_rgb_vec_for_projection->size()
    //      << " | " << m_rgb_pts_vec.size()
    //      << ", cost time = " << timer.toc() << ANSI_COLOR_RESET << endl;
    m_mutex_pts_vec->unlock();
    last_pose_q = img_for_projection->m_pose_w2c_q;
}
 // **可选：添加一条日志，确认线程已退出**
std::cout << "Global_map::service_refresh_pts_for_projection() thread exiting gracefully." << std::endl;

} #/root/catkin_coco/src/Coco-LIC/src/camera/rgb_map/pointcloud_rgbd.hpp中新增一行 struct Global_map { int m_map_major_version = R3LIVE_MAP_MAJOR_VERSION; int m_map_minor_version = R3LIVE_MAP_MINOR_VERSION; int m_if_get_all_pts_in_boxes_using_mp = 1; std::vector< RGB_pt_ptr > m_rgb_pts_vec; // std::vector< RGB_pt_ptr > m_rgb_pts_in_recent_visited_voxels; std::shared_ptr< std::vector< RGB_pt_ptr> > m_pts_rgb_vec_for_projection = nullptr; std::shared_ptr< std::mutex > m_mutex_pts_vec; std::shared_ptr< std::mutex > m_mutex_recent_added_list; std::shared_ptr< std::mutex > m_mutex_img_pose_for_projection; std::shared_ptr< std::mutex > m_mutex_rgb_pts_in_recent_hitted_boxes; std::shared_ptr< std::mutex > m_mutex_m_box_recent_hitted; std::shared_ptr< std::mutex > m_mutex_pts_last_visited; Image_frame m_img_for_projection; double m_recent_visited_voxel_activated_time = 0.0; bool m_in_appending_pts = 0; int m_updated_frame_index = 0; std::shared_ptr< std::thread > m_thread_service; std::atomic m_should_terminate_thread;// <-- 新增这一行 int m_if_reload_init_voxel_and_hashed_pts = true; //。。。后续代码

Sorry for the late response! Thanks to Geyuwen for the explanation.🍭 I will check and work on a fix for this issue as soon as possible.

[lojmkjd]

@Geyuwen 谢谢您的解决方案，但是在我仔细对照您的修改后，还是出现了如上的报错；若您有时间，可以告诉我解决办法或是参考的方向吗？十分感谢您！

[lojmkjd]

[@Geyuwen](https://github.com/Geyuwen) 谢谢您的解决方案，但是在我仔细对照您的修改后，还是出现了如上的报错；若您有时间，可以告诉我解决办法或是参考的方向吗？十分感谢您！

我今天使用ASan进行了调试，希望能为您提供更多的信息：
`

==70284==ERROR: AddressSanitizer: attempting free on address which was not malloc()-ed: 0x606000337840 in thread T0
#0 0x7365520cf40f in __interceptor_free ../../../../src/libsanitizer/asan/asan_malloc_linux.cc:122
#1 0x5b00ef7c198c in void Eigen::internal::permute_symm_to_fullsymm<2, Eigen::SparseMatrix<double, 0, int>, 0>(Eigen::SparseMatrix<double, 0, int> const&, Eigen::SparseMatrix<Eigen::SparseMatrix<double, 0, int>::Scalar, 0, Eigen::SparseMatrix<double, 0, int>::StorageIndex>&, Eigen::SparseMatrix<double, 0, int>::StorageIndex const*) [clone .constprop.0] (/catkin_coco/devel/lib/cocolic/odometry_node+0x88a98c)
#2 0x5b00ef7cd452 in Eigen::SimplicialCholeskyBase<Eigen::SimplicialLDLT<Eigen::SparseMatrix<double, 0, int>, 2, Eigen::NaturalOrdering > >::ordering(Eigen::SparseMatrix<double, 0, int> const&, Eigen::SparseMatrix<double, 0, int> const*&, Eigen::SparseMatrix<double, 0, int>&) (/catkin_coco/devel/lib/cocolic/odometry_node+0x896452)
#3 0x5b00ef7cd7c2 in ceres::internal::EigenSparseCholeskyTemplate<Eigen::SimplicialLDLT<Eigen::SparseMatrix<double, 0, int>, 2, Eigen::NaturalOrdering > >::Factorize(ceres::internal::CompressedRowSparseMatrix*, std::__cxx11::basic_string<char, std::char_traits, std::allocator >) (/catkin_coco/devel/lib/cocolic/odometry_node+0x8967c2)
#4 0x5b00ef69e624 in ceres::internal::SparseCholesky::FactorAndSolve(ceres::internal::CompressedRowSparseMatrix, double const*, double*, std::__cxx11::basic_string<char, std::char_traits, std::allocator >) (/catkin_coco/devel/lib/cocolic/odometry_node+0x767624)
#5 0x5b00ef69f34f in ceres::internal::SparseNormalCholeskySolver::SolveImpl(ceres::internal::BlockSparseMatrix, double const*, ceres::internal::LinearSolver::PerSolveOptions const&, double*) (/catkin_coco/devel/lib/cocolic/odometry_node+0x76834f)
#6 0x5b00ef673866 in ceres::internal::TypedLinearSolverceres::internal::BlockSparseMatrix::Solve(ceres::internal::LinearOperator*, double const*, ceres::internal::LinearSolver::PerSolveOptions const&, double*) (/catkin_coco/devel/lib/cocolic/odometry_node+0x73c866)
#7 0x5b00ef66020f in ceres::internal::LevenbergMarquardtStrategy::ComputeStep(ceres::internal::TrustRegionStrategy::PerSolveOptions const&, ceres::internal::SparseMatrix*, double const*, double*) (/catkin_coco/devel/lib/cocolic/odometry_node+0x72920f)
#8 0x5b00ef63a97d in ceres::internal::TrustRegionMinimizer::ComputeTrustRegionStep() (/catkin_coco/devel/lib/cocolic/odometry_node+0x70397d)
#9 0x5b00ef63ee09 in ceres::internal::TrustRegionMinimizer::Minimize(ceres::internal::Minimizer::Options const&, double*, ceres::Solver::Summary*) (/catkin_coco/devel/lib/cocolic/odometry_node+0x707e09)
#10 0x5b00ef61291d in ceres::Solver::Solve(ceres::Solver::Options const&, ceres::Problem*, ceres::Solver::Summary*) (/catkin_coco/devel/lib/cocolic/odometry_node+0x6db91d)
#11 0x5b00ef61310d in ceres::Solve(ceres::Solver::Options const&, ceres::Problem*, ceres::Solver::Summary*) (/catkin_coco/devel/lib/cocolic/odometry_node+0x6dc10d)
#12 0x5b00ef2e34df in cocolic::TrajectoryEstimator::Solve(int, bool, int) /catkin_coco/src/Coco-LIC/src/odom/trajectory_estimator.cpp:506
#13 0x5b00ef19a27e in cocolic::TrajectoryManager::InitTrajWithPropagation() /catkin_coco/src/Coco-LIC/src/odom/trajectory_manager.cpp:309
#14 0x5b00ef19b697 in cocolic::TrajectoryManager::PredictTrajectory(long, long, long, int, bool) /catkin_coco/src/Coco-LIC/src/odom/trajectory_manager.cpp:271
#15 0x5b00ef489608 in cocolic::OdometryManager::ProcessLICData() /catkin_coco/src/Coco-LIC/src/odom/odometry_manager.cpp:291
#16 0x5b00ef493024 in cocolic::OdometryManager::SolveLICO() /catkin_coco/src/Coco-LIC/src/odom/odometry_manager.cpp:256
#17 0x5b00ef493024 in cocolic::OdometryManager::RunBag() /catkin_coco/src/Coco-LIC/src/odom/odometry_manager.cpp:223
#18 0x5b00ef04c645 in main /catkin_coco/src/Coco-LIC/src/odometry_node.cpp:44
#19 0x7365462b0082 in __libc_start_main ../csu/libc-start.c:308
#20 0x5b00ef055b9d in _start (/catkin_coco/devel/lib/cocolic/odometry_node+0x11eb9d)

0x606000337840 is located 32 bytes inside of 56-byte region [0x606000337820,0x606000337858)
allocated by thread T0 here:
#0 0x7365520cf808 in __interceptor_malloc ../../../../src/libsanitizer/asan/asan_malloc_linux.cc:144
#1 0x5b00ef1d03e8 in Eigen::internal::handmade_aligned_malloc(unsigned long) /usr/include/eigen3/Eigen/src/Core/util/Memory.h:88
#2 0x5b00ef1d03e8 in Eigen::internal::aligned_malloc(unsigned long) /usr/include/eigen3/Eigen/src/Core/util/Memory.h:164
#3 0x5b00ef7c1a1c in void Eigen::internal::permute_symm_to_fullsymm<2, Eigen::SparseMatrix<double, 0, int>, 0>(Eigen::SparseMatrix<double, 0, int> const&, Eigen::SparseMatrix<Eigen::SparseMatrix<double, 0, int>::Scalar, 0, Eigen::SparseMatrix<double, 0, int>::StorageIndex>&, Eigen::SparseMatrix<double, 0, int>::StorageIndex const*) [clone .constprop.0] (/catkin_coco/devel/lib/cocolic/odometry_node+0x88aa1c)

SUMMARY: AddressSanitizer: bad-free ../../../../src/libsanitizer/asan/asan_malloc_linux.cc:122 in __interceptor_free
==70284==ABORTING
`

[weiyi808]

Hi Jerry-locker,

I modified the thread exit mechanism and projection service in Global_map as follows:

Used std::atomic for thread flag and ensured join() in destructor.
Added full exception handling and sleep control in service_refresh_pts_for_projection().
Could you review if these changes might introduce:

1. Any race conditions I missed?
2. Performance impact on real-time processing?
3. Conflicts with other parts of the system?

Thanks for your guidance!

std::atomic m_stop_thread_flag = false;
Global_map::~Global_map(){
m_stop_thread_flag = true;
if (m_thread_service && m_thread_service->joinable()) {
m_thread_service->join(); // 确保线程安全退出
}};

void Global_map::service_refresh_pts_for_projection()
{
try {
eigen_q last_pose_q = eigen_q::Identity();
Common_tools::Timer timer;
std::shared_ptr<Image_frame> img_for_projection = std::make_shared<Image_frame>();
while (!m_stop_thread_flag)
{ std::this_thread::sleep_for(std::chrono::milliseconds(10));
m_mutex_img_pose_for_projection->lock();
*img_for_projection = m_img_for_projection;
m_mutex_img_pose_for_projection->unlock();
if (img_for_projection->m_img_cols == 0 || img_for_projection->m_img_rows == 0)
continue;
if (img_for_projection->m_frame_idx == m_updated_frame_index)
continue;
timer.tic(" ");
std::shared_ptr<std::vector<std::shared_ptr<RGB_pts>>> pts_rgb_vec_for_projection =
std::make_shared<std::vector<std::shared_ptr<RGB_pts>>>();

        if (m_if_get_all_pts_in_boxes_using_mp)
        {
            std::vector<std::shared_ptr<RGB_pts>> pts_in_recent_hitted_boxes;
            pts_in_recent_hitted_boxes.reserve(1e6);

            std::unordered_set<std::shared_ptr<RGB_Voxel>> boxes_recent_hitted;
            m_mutex_m_box_recent_hitted->lock();
            boxes_recent_hitted = m_voxels_recent_visited;
            m_mutex_m_box_recent_hitted->unlock();

            m_mutex_rgb_pts_in_recent_hitted_boxes->lock();
            // m_rgb_pts_in_recent_visited_voxels = pts_in_recent_hitted_boxes;
            m_mutex_rgb_pts_in_recent_hitted_boxes->unlock();
        }

        Eigen::aligned_vector<Eigen::Vector3d> fake_points;
        Eigen::aligned_vector<Eigen::Vector2d> fake_pixs;
        selection_points_for_projection(false, fake_points, fake_pixs, img_for_projection, pts_rgb_vec_for_projection.get(), nullptr, 10.0, 1);

        m_mutex_pts_vec->lock();
        m_pts_rgb_vec_for_projection = pts_rgb_vec_for_projection;
        m_updated_frame_index = img_for_projection->m_frame_idx;
        m_mutex_pts_vec->unlock();

        last_pose_q = img_for_projection->m_pose_w2c_q;
    }

} catch (const std::exception& e) {
    std::cerr << "[Global_map::service_refresh_pts_for_projection] Exception: " << e.what() << std::endl;
} catch (...) {
    std::cerr << "[Global_map::service_refresh_pts_for_projection] Unknown exception occurred." << std::endl;
}}

[robotvision2BJTU]

便 Start Coco-LIC Odometry 便
便 Odometry Mode: LiDAR-Inertial-Camera Odometry 便

 The number of multiple LiDARs is 1.

 LoadBag /project/DATA/Fast-LIVO/hku1.bag start at 0 with duration 127.696.

 Static initialization succeeds.

 Trajectory start time: 1577836979704858112 ns.
double free or corruption (out)
[odometry_node_irobot_Super_Server_24318_8258771888953405027-2] process has died [pid 24343, exit code -6, cmd /project/catkin_ws/devel/lib/cocolic/odometry_node __name:=odometry_node_irobot_Super_Server_24318_8258771888953405027 __log:=/root/.ros/log/7de10c42-74e1-11f0-903e-ac1f6bfc8c2c/odometry_node_irobot_Super_Server_24318_8258771888953405027-2.log].
log file: /root/.ros/log/7de10c42-74e1-11f0-903e-ac1f6bfc8c2c/odometry_node_irobot_Super_Server_24318_8258771888953405027-2*.log
================================================================================REQUIRED process [rviz_odom-3] has died!
process has finished cleanly
log file: /root/.ros/log/7de10c42-74e1-11f0-903e-ac1f6bfc8c2c/rviz_odom-3*.log
Initiating shutdown!

[Victordmss]

+1