It can be consistently reproduced when running the following command (passes with QT_QPA_DISABLE_ENHANCED_MOUSE=1):

./tst_qtreeview selection statusTip

but not when running the statusTip alone ./tst_qtreeview statusTip. Git bisect points to https://codereview.qt-project.org/#/c/193415/ . The code paths between the passing test and the failing test diverge in void QWidgetWindow::handleMouseEvent(QMouseEvent *event), at:

    if (event->type() == QEvent::MouseButtonPress)
        qt_button_down = widget;

    QWidget *receiver = QApplicationPrivate::pickMouseReceiver(m_widget, event->windowPos().toPoint(), &mapped, event->type(), event->buttons(),
                                                               qt_button_down, widget);

My initial guess is that https://codereview.qt-project.org/#/c/193415/ exposes an issue which was somehow not apparent before. Test run by qtestlib are often flaky because they are affected by the test functions that were run earlier. It seems to be the case here as well.