QVector（或std :: vector）保留（或调整大小）不会释放内存

Question

Say I have a QVector and I call QVector::reserve for 420000 elements. Right after that I call QVector::reserve again for 42 elements. What I noticed is that after doing this, the process still has memory allocated for those 420000 elements. Why doesn't memory get freed when I shrink the QVector size to 42?

Take the following code:

#include <QCoreApplication>
#include <QVector>
#include <QDebug>

typedef QVector<QPointF> PointVec;
typedef QList<PointVec*> PointVecList;

int main(int argc, char *argv[])
{
    QCoreApplication a(argc, argv);

    qDebug() << "Processing...";

    PointVecList list;

    for (int i = 0; i < 4200; i++) {
        PointVec *v = new PointVec();

        v->reserve(420000);
        v->resize(420000);

        v->reserve(42);
        v->resize(42);

        list.append(v);
    }

    qDebug() << "End of processing...";

    return a.exec();
}

This code crashes throwing std::bad_aloc exception( Unhandled exception at 0x7729C41F in QtConsoleApplication1.exe: Microsoft C++ exception: std::bad_alloc at memory location 0x0034F5E8. ) on a 32 bit system(as the system doesn't support much RAM for one process).

However if I comment lines v->reserve(420000); and v->resize(420000); everything works fine. This is how I noticed that if I call v->reserve(42); and v->resize(42); (after reserving/resizing the vector to 420000 elements beforehand) exceeding memory doesn't get freed.

What I also noticed is that std::vector behaves exactly the same.

Is it possible to force Qt to also free the memory? If not, is it possible to free it myself, manually, somehow?

Answer 1

Unless there is actual memory pressure it makes sense for the underlying library to not do the work of actually releasing the allocated memory back to the OS but instead keep it around and use it to satisfy new allocations (without incurring a context switch to get it from the kernel) that you are likely to make soon anyway.

Same goes for the OS. Even if the application actually releases the memory there's no point in doing the work of actually reclaiming it until it is actually needed elsewhere (with luck it never will be until the process exits and it can just do a single bulk reclaim of everything).

So, the reason is efficiency. There's a cost involved in actually freeing the memory and there's no reason to pay the cost until you really have to - which, in many cases will be never.