Memory leak in QHeaderView with hidden sections attached to QTreeView

jamesmpicone

Hi,

I've recently run into an issue with QTreeView/QHeaderView on Qt 4.6.2 that, although I'm loathe to blame the library, looks to me like a Qt bug. Sample code:

@
#include <QStandardItemModel>
#include <QTreeView>

class TreeviewTester : public QMainWindow
{
Q_OBJECT

public:
TreeviewTester::TreeviewTester(QWidget parent = 0, Qt::WFlags flags = 0)
: QMainWindow(parent, flags)
{
QTreeView view = new QTreeView(this);
setCentralWidget(view);

QList<QStandardItem*> items;
items << new QStandardItem("Test") << new QStandardItem("Test2") << new QStandardItem("Test3");

Model = new QStandardItemModel(this);
Model->appendRow(items);

view->setModel(Model);
view->setColumnHidden(1, true);

TimerId = startTimer(0);

}

TreeviewTester::~TreeviewTester()
{
killTimer(TimerId);
}

protected:

void timerEvent(QTimerEvent* event)
{
Model->sort(0);
}

private:
QStandardItemModel* Model;

int TimerId;
};
@

The important parts here are:

Create a QTreeView
Give it a model
Make sure the model has some rows
Hide a column on the header
Trigger the model's layout-changed signal processing (here done with the sort call).

This program leaks memory and eventually crashes, because the QList QHeaderViewPrivate::persistentHiddenSections is being appended to every time the model throws out the about-to-change-layout signal, and never cleared. It's supposed to be emptied out in the layout-changed signal handler, but QTreeView::setModel() disconnects the signal to its header for some reason, and never passes it down as far as I can tell.

Calling QTreeView::setHeader() after QTreeView::setModel() prevents the problem, because QTreeView::setHeader() doesn't disconnect the _q_layoutChanged() signal.

Looking at the 4.7 repository, it looks like it would do the same thing.

So am I misunderstanding something about the way these views should work, or is this an actual Qt bug?

[EDIT: fixed list formatting, Volker]

andre

Please report it as a bug in Jira. If such a simple testcase triggers the problem, then I think you're on to a real issue here.