莫名其妙的free()：C++中delete[]的大小无效

Question

在这两种情况下，我都使用完全相同的二维数组（它是类的一个字段）。第二个lsh_distances 不允许我删除[] 析构函数中的内部一维数组，并给出free(): invalid size Aborted (core dumped) 的错误。如果我不使用 for 循环删除 [] 内部数组，我当然会有内存泄漏（使用 valgrind 检查）。

为什么会发生这种情况？这没有道理。我不会在其他任何地方删除lsh_distances，我实际上可以打印析构函数中的值。

帮助

这里是代码。

long long int NearestNeighboursSearch::calculate_exact_distances() {
    unsigned int q_size = query_dataset.points.size();
    unsigned int ind_size = index_dataset.points.size();
    exact_distances = new long double*[q_size];
    for (int i = 0 ; i < q_size ; i++) {
        exact_distances[i] = new long double[ind_size];
    }
    chrono::steady_clock::time_point begin = std::chrono::steady_clock::now();
    for (int i = 0 ; i < q_size ; i++) {
        for (int j = 0 ; j < ind_size ; j++) {
            exact_distances[i][j] = used_distance(*query_dataset.points.at(i), *index_dataset.points.at(j));
        }
    }
    chrono::steady_clock::time_point end = std::chrono::steady_clock::now();
    return chrono::duration_cast<chrono::milliseconds>(end - begin).count();
}

long long int NearestNeighboursSearch::calculate_lsh_distances() {
    unsigned int q_size = query_dataset.points.size();
    unsigned int ind_size = index_dataset.points.size();
    lsh_distances = new long double*[q_size];
    for (int i = 0 ; i < q_size ; i++) {
        lsh_distances[i] = new long double[ind_size];
        for (int j = 0 ; j < ind_size ; j++) {
            lsh_distances[i][j] = -1.0;   // not calculated
        }
    }
    // calculate and fill only distances between points that hash into the same bucket for the L hash tables
    chrono::steady_clock::time_point begin = std::chrono::steady_clock::now();
    for (int i = 0 ; i < q_size ; i++) {
        Point *p1 = query_dataset.points.at(i);
        Bucket **buckets = index_dataset.get_buckets_for_point(p1);   // L buckets
        for (int j = 0 ; j < index_dataset.get_hashtables_count(); j++) {
            for (int k = 0 ; k < buckets[j]->points.size() ; k++) {           // might be empty
                const Point *p2 = buckets[j]->points[k];
                // update appropriate distance at [i, p2->pos]
                lsh_distances[i][p2->pos] = used_distance(*p1, *p2);
            }
        }
        delete[] buckets;
    }
    chrono::steady_clock::time_point end = std::chrono::steady_clock::now();
    return chrono::duration_cast<chrono::milliseconds>(end - begin).count();
}

NearestNeighboursSearch::~NearestNeighboursSearch() {
    if (exact_distances != NULL) {
        for (int i = 0; i < query_dataset.points.size(); i++) {
            delete[] exact_distances[i];
        }
        delete[] exact_distances;
    }
    if (lsh_distances != NULL) {
        // TODO: Somehow this causes an error (it makes no sense to me)
        for (int i = 0; i < query_dataset.points.size(); i++) {
            delete[] lsh_distances[i];
        }
        delete[] lsh_distances;
    }
}

编辑：我按照建议使用 valgrind 进行了运行。我得到了这个：

==376== Invalid write of size 4
==376==    at 0x11508A: NearestNeighboursSearch::calculate_lsh_distances() (neighbour_search.cpp:66)
==376==    by 0x10B2CA: main (lsh_main.cpp:132)
==376==  Address 0x4e385c0 is 16 bytes before a block of size 160 alloc'd
==376==    at 0x483C583: operator new[](unsigned long) (in /usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so)
==376==    by 0x114F17: NearestNeighboursSearch::calculate_lsh_distances() (neighbour_search.cpp:52)
==376==    by 0x10B2CA: main (lsh_main.cpp:132)
==376==
==376== Invalid write of size 2
==376==    at 0x11508A: NearestNeighboursSearch::calculate_lsh_distances() (neighbour_search.cpp:66)
==376==    by 0x10B2CA: main (lsh_main.cpp:132)
==376==  Address 0x4e385c8 is 8 bytes before a block of size 160 alloc'd
==376==    at 0x483C583: operator new[](unsigned long) (in /usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so)
==376==    by 0x114F17: NearestNeighboursSearch::calculate_lsh_distances() (neighbour_search.cpp:52)
==376==    by 0x10B2CA: main (lsh_main.cpp:132)
==376==

我将对此进行调查...看来 p2->pos 不知何故为

Answer 1

好的。我明白了为什么要感谢评论和 valgrind。 p2->pos 是 -1 在其他地方的一个粗心错误之后，我可能重写了一些重要的东西。逻辑已恢复。