https://bugs.kde.org/show_bug.cgi?id=418316
--- Comment #6 from Oded Arbel <o...@geek.co.il> --- Created attachment 126603 --> https://bugs.kde.org/attachment.cgi?id=126603&action=edit Defensive programming patch This looks like some kind of "use after free": 1. PlasmaQuick::Dialog registers PlasmaQuick::DialogPrivate::updateInputShape() as the slot for QWindow::visibleChanged(bool). 2. QWindow d'tor calls QWindowPrivate::setVisible(false) (through QWindowPrivate::destroy()), which will in turn signal QWindow::visibleChanged(bool). 3. PlasmaQuick::DialogPrivate::updateInputShape() gets called after PlasmaQuick::Dialog had already run and released the DialogPrivate reference (and apparently zeroed it out?) and calls back into PlasmaQuick::Dialog::isVisible() 4. When isVisible() tries to access DialogPrivate back, the pointer is already zero and we get SIGSEGVED The attached defensive patch works around the problem for me (a bit - I did get one crash after it, but I didn't get a stack trace - so it may not be the same). This is obviously not the correct approach. Maybe unregister from the signals during d'tor? -- You are receiving this mail because: You are watching all bug changes.
