Re: [Interest] Crash when signal fires
> On Apr 22, 2020, at 11:19 AM, Thiago Macieira > wrote: > > On Tuesday, 21 April 2020 17:09:45 PDT Ben Haller via Interest wrote: >> Unfortunately, I’m on macOS 10.15.3 and Valgrind is not there yet (I think >> at present they have “preliminary” support for macOS 10.13). > > Then install a Linux VM and use Valgrind there. Valgrind is useful enough to > warrant a VM just for it. Ah, that is a good idea! I shall do so. >> I suspect this is a really dumb question, but what is “ASAN”? Is that >> another Valgrind-like tool? > > -fsanitize=address in your compilation and linking flags. > > With qmake, you can also get that by doing: > > CONFIG += sanitizer sanitize_address Aha, very interesting. I’ll try this as well. Thanks for the help. Cheers, -B. Benjamin C. Haller Messer Lab Cornell University ___ Interest mailing list Interest@qt-project.org https://lists.qt-project.org/listinfo/interest
Re: [Interest] Crash when signal fires
> On Apr 22, 2020, at 11:17 AM, Thiago Macieira > wrote: > > On Tuesday, 21 April 2020 16:44:19 PDT Ben Haller via Interest wrote: >>connect(qApp, ::focusChanged, [this]() { >> updateUIEnabling(); }); > > Your object probably outlived QApplication. Other way around, right? > Add a third argument of "this" to the connect() call. Yes, Giuseppe D’Angelo set me on the right path on this, and I’ve just gone through all the connect() calls in my app and made them all safe. Thanks! Cheers, -B. Benjamin C. Haller Messer Lab Cornell University ___ Interest mailing list Interest@qt-project.org https://lists.qt-project.org/listinfo/interest
Re: [Interest] Crash when signal fires
On Tuesday, 21 April 2020 17:09:45 PDT Ben Haller via Interest wrote: > Unfortunately, I’m on macOS 10.15.3 and Valgrind is not there yet (I think > at present they have “preliminary” support for macOS 10.13). Then install a Linux VM and use Valgrind there. Valgrind is useful enough to warrant a VM just for it. > I suspect this is a really dumb question, but what is “ASAN”? Is that > another Valgrind-like tool? -fsanitize=address in your compilation and linking flags. With qmake, you can also get that by doing: CONFIG += sanitizer sanitize_address -- Thiago Macieira - thiago.macieira (AT) intel.com Software Architect - Intel System Software Products ___ Interest mailing list Interest@qt-project.org https://lists.qt-project.org/listinfo/interest
Re: [Interest] Crash when signal fires
On Tuesday, 21 April 2020 16:44:19 PDT Ben Haller via Interest wrote: > connect(qApp, ::focusChanged, [this]() { > updateUIEnabling(); }); Your object probably outlived QApplication. Add a third argument of "this" to the connect() call. -- Thiago Macieira - thiago.macieira (AT) intel.com Software Architect - Intel System Software Products ___ Interest mailing list Interest@qt-project.org https://lists.qt-project.org/listinfo/interest
Re: [Interest] Crash when signal fires
> On Apr 21, 2020, at 7:51 PM, Giuseppe D'Angelo via Interest > wrote: > > On 4/22/20 1:44 AM, Ben Haller via Interest wrote: >> Hi folks. I’m seeing something weird that I don’t understand. I have a >> connection defined for the main window of my Qt Widgets app, like so: >> connect(qApp, ::focusChanged, [this]() { >> updateUIEnabling(); }); >> I’m seeing a 100% reproducible crash as a result of this. When the >> focusChanged signal fires, I end up in updateUIEnabling() as expected, but I >> immediately crash because something about the state of things is corrupted; >> the immediate symptom, according to the debugger, is that the ui pointer for >> my main window is equal to 0x1 (yuck). I try to access a button, >> ui->playOneStepButton, and it crashes. As I said, 100% reproducible. >> The weird thing is that this does*not* happen if I change the connect >> statement to this: >> connect(qApp, ::focusChanged, this, >> ::updateUIEnabling); >> Now, 100% of the time, there is no problem, no crash, and my app chugs along >> happily. No other code changes involved. I can literally flip this connect >> definition back and forth in my code and go from 100% crashing to 100% fine. > > A quick checklist: > > 0) I hope threads are not involved (you're not really allowed to do any GUI > work from any other thread). No threads. > 1) Never use the 3-arguments connect(). If you want to use the lambda, pass > "this" as 3rd parameter and lambda as fourth. Why: if "this" has been > destroyed, the first version won't disconnect automatically and crash. The > second will disconnect and not crash. > > More info: https://stackoverflow.com/a/27954502 Ah, I see! I was assuming it was my first window’s focusChanged connection that was crashing, but it was probably the second (closed) window’s connection that was crashing, in fact, because the window was gone but the connection remained. > 2) ASAN, Valgrind and friends will help you locate the source of the crash if > it's due to memory corruption. Use them. Unfortunately, I’m on macOS 10.15.3 and Valgrind is not there yet (I think at present they have “preliminary” support for macOS 10.13). I suspect this is a really dumb question, but what is “ASAN”? Is that another Valgrind-like tool? Well, in any case, I think you have explained my bug. Thanks very much! Cheers, -B. Benjamin C. Haller Messer Lab Cornell University ___ Interest mailing list Interest@qt-project.org https://lists.qt-project.org/listinfo/interest
Re: [Interest] Crash when signal fires
On 4/22/20 1:44 AM, Ben Haller via Interest wrote: Hi folks. I’m seeing something weird that I don’t understand. I have a connection defined for the main window of my Qt Widgets app, like so: connect(qApp, ::focusChanged, [this]() { updateUIEnabling(); }); I’m seeing a 100% reproducible crash as a result of this. When the focusChanged signal fires, I end up in updateUIEnabling() as expected, but I immediately crash because something about the state of things is corrupted; the immediate symptom, according to the debugger, is that the ui pointer for my main window is equal to 0x1 (yuck). I try to access a button, ui->playOneStepButton, and it crashes. As I said, 100% reproducible. The weird thing is that this does*not* happen if I change the connect statement to this: connect(qApp, ::focusChanged, this, ::updateUIEnabling); Now, 100% of the time, there is no problem, no crash, and my app chugs along happily. No other code changes involved. I can literally flip this connect definition back and forth in my code and go from 100% crashing to 100% fine. A quick checklist: 0) I hope threads are not involved (you're not really allowed to do any GUI work from any other thread). 1) Never use the 3-arguments connect(). If you want to use the lambda, pass "this" as 3rd parameter and lambda as fourth. Why: if "this" has been destroyed, the first version won't disconnect automatically and crash. The second will disconnect and not crash. More info: https://stackoverflow.com/a/27954502 2) ASAN, Valgrind and friends will help you locate the source of the crash if it's due to memory corruption. Use them. HTH, -- Giuseppe D'Angelo | giuseppe.dang...@kdab.com | Senior Software Engineer KDAB (France) S.A.S., a KDAB Group company Tel. France +33 (0)4 90 84 08 53, http://www.kdab.com KDAB - The Qt, C++ and OpenGL Experts smime.p7s Description: S/MIME Cryptographic Signature ___ Interest mailing list Interest@qt-project.org https://lists.qt-project.org/listinfo/interest
[Interest] Crash when signal fires
[ I realized I ought to supply the crash backtrace; I have appended it below my signature. ] Hi folks. I’m seeing something weird that I don’t understand. I have a connection defined for the main window of my Qt Widgets app, like so: connect(qApp, ::focusChanged, [this]() { updateUIEnabling(); }); I’m seeing a 100% reproducible crash as a result of this. When the focusChanged signal fires, I end up in updateUIEnabling() as expected, but I immediately crash because something about the state of things is corrupted; the immediate symptom, according to the debugger, is that the ui pointer for my main window is equal to 0x1 (yuck). I try to access a button, ui->playOneStepButton, and it crashes. As I said, 100% reproducible. The weird thing is that this does *not* happen if I change the connect statement to this: connect(qApp, ::focusChanged, this, ::updateUIEnabling); Now, 100% of the time, there is no problem, no crash, and my app chugs along happily. No other code changes involved. I can literally flip this connect definition back and forth in my code and go from 100% crashing to 100% fine. I’m pretty new to Qt and maybe I’m just being clueless somehow; but I thought that the two definitions above would be essentially identical in their effect. Why does the first one crash? Note that it works fine in general; it seems to crash specifically when I open a *second* main window (which makes its own connection to focusChanged in the same way), and then close that second main window. Now the first main window’s connection is somehow unhappy, and the next time I trigger the focusChanged() signal, boom. It’s as if the connection for the second window somehow interferes with the connection for the first window, and closing the second window leads to a corruption in the connection state. Naturally, I am now nervous about using the first type of connection anywhere in my code, since I don’t understand what the issue is. Can anyone shed light on this? Cheers, -B. Benjamin C. Haller Messer Lab Cornell University Thread 0 Crashed:: Dispatch queue: com.apple.main-thread 0 com.sticksoftware.QtSLiM0x0001003c84b1 QtSLiMWindow::updateUIEnabling() + 17 (QtSLiMWindow.cpp:1589) 1 com.sticksoftware.QtSLiM0x0001003d224e QtPrivate::FunctorCall, QtPrivate::List<>, void, QtSLiMWindow::init()::$_3>::call(QtSLiMWindow::init()::$_3&, void**) + 14 (qobjectdefs_impl.h:146) 2 org.qt-project.QtCore 0x00010228ca95 void doActivate(QObject*, int, void**) + 1157 (memory:2597) 3 org.qt-project.QtWidgets0x0001013d67de QApplicationPrivate::setFocusWidget(QWidget*, Qt::FocusReason) + 510 4 org.qt-project.QtWidgets0x00010140f6fd QWidget::setFocus(Qt::FocusReason) + 637 (qwidget.cpp:6356) 5 org.qt-project.QtWidgets0x0001013d7dd8 QApplication::setActiveWindow(QWidget*) + 1160 6 org.qt-project.QtWidgets0x0001013d8333 QApplicationPrivate::notifyActiveWindowChange(QWindow*) + 51 (qapplication.cpp:2101) 7 org.qt-project.QtGui0x000101a35623 QGuiApplicationPrivate::processActivatedEvent(QWindowSystemInterfacePrivate::ActivatedWindowEvent*) + 675 (qguiapplication.cpp:2438) 8 org.qt-project.QtGui0x000101a12cc3 bool QWindowSystemInterfacePrivate::handleWindowSystemEvent(QWindowSystemInterfacePrivate::WindowSystemEvent*) + 115 (qwindowsysteminterface.cpp:106) 9 libqcocoa.dylib 0x0001063511ee QCocoaWindow::windowDidBecomeKey() + 238 (qcocoawindow.mm:1194) 10 org.qt-project.QtCore 0x0001022648d6 QMetaMethod::invoke(QObject*, Qt::ConnectionType, QGenericReturnArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument, QGenericArgument) const + 1270 11 libqcocoa.dylib 0x000106354830 invocation function for block in qRegisterNotificationCallbacks() + 1936 (qcocoawindow.mm:131) 12 com.apple.Foundation0x7fff3860cfab -[__NSObserver _doit:] + 296 13 com.apple.CoreFoundation0x7fff35f8c35f __CFNOTIFICATIONCENTER_IS_CALLING_OUT_TO_AN_OBSERVER__ + 12 14 com.apple.CoreFoundation0x7fff35f8c2f3 ___CFXRegistrationPost1_block_invoke + 63 15 com.apple.CoreFoundation0x7fff35f8c268 _CFXRegistrationPost1 + 372 16 com.apple.CoreFoundation0x7fff35f8bebe ___CFXNotificationPost_block_invoke + 97 17 com.apple.CoreFoundation0x7fff35f5b7e2 -[_CFXNotificationRegistrar find:object:observer:enumerator:] + 1575 18 com.apple.CoreFoundation0x7fff35f5ac82 _CFXNotificationPost + 1351 19 com.apple.Foundation0x7fff385e0a22 -[NSNotificationCenter postNotificationName:object:userInfo:] + 59 20 com.apple.AppKit
[Interest] Crash when signal fires
Hi folks. I’m seeing something weird that I don’t understand. I have a connection defined for the main window of my Qt Widgets app, like so: connect(qApp, ::focusChanged, [this]() { updateUIEnabling(); }); I’m seeing a 100% reproducible crash as a result of this. When the focusChanged signal fires, I end up in updateUIEnabling() as expected, but I immediately crash because something about the state of things is corrupted; the immediate symptom, according to the debugger, is that the ui pointer for my main window is equal to 0x1 (yuck). I try to access a button, ui->playOneStepButton, and it crashes. As I said, 100% reproducible. The weird thing is that this does *not* happen if I change the connect statement to this: connect(qApp, ::focusChanged, this, ::updateUIEnabling); Now, 100% of the time, there is no problem, no crash, and my app chugs along happily. No other code changes involved. I can literally flip this connect definition back and forth in my code and go from 100% crashing to 100% fine. I’m pretty new to Qt and maybe I’m just being clueless somehow; but I thought that the two definitions above would be essentially identical in their effect. Why does the first one crash? Note that it works fine in general; it seems to crash specifically when I open a *second* main window (which makes its own connection to focusChanged in the same way), and then close that second main window. Now the first main window’s connection is somehow unhappy, and the next time I trigger the focusChanged() signal, boom. It’s as if the connection for the second window somehow interferes with the connection for the first window, and closing the second window leads to a corruption in the connection state. Naturally, I am now nervous about using the first type of connection anywhere in my code, since I don’t understand what the issue is. Can anyone shed light on this? Cheers, -B. Benjamin C. Haller Messer Lab Cornell University ___ Interest mailing list Interest@qt-project.org https://lists.qt-project.org/listinfo/interest