I tested under MacOS.  With Qt4, the new code systematically segfaults when loading a document that opens a warning window (e.g. some newer entries in the bibliography) and then closing.
With Qt5 this still occurs sporadically, but in a non-deterministic manner.  With Qt5, I also get the following messages when closing Qt5 (after loading the document as above):

QBasicTimer::start: QBasicTimer can only be used with threads started with QThread
QBasicTimer::start: QBasicTimer can only be used with threads started with QThread

I #ifdeffed the Mac case.

Max: any idea why we have this problem under MacOS?

I now have several simple patterns that crash TeXmacs with certainty or very high probability, even with debug symbols and debug switches.

The crash seems to always occur In QTMWindow.cpp on a line :
string name= ( !is_nil(tmwid) ? concrete(tmwid)->get_nickname () : "QTMWindow")
i.e. most likely when trying to figure the window's name (but failing along the way because a pointer became invalid?)

As Jeroen and I reported, the crash mostly occurs in QTMWindow::moveEvent, definitely after the window has been destroyed, which likely explains the crash.

I note that at the time of the crash, on the console I get a qt warning message as reported in https://bugreports.qt.io/browse/QTBUG-56893 and https://bugreports.qt.io/browse/QTBUG-87141, signaling an event was sent to a destroyed object (an occurence that cannot be avoided and which should be begnin according to the qt folks). This qt warning is also present when using the Appimage (that does not crash) which seems to confirm the Qt folks opinion.

The bug is still present in a qt6 build. So, fixing it is not just a matter of waiting we get to qt6 ...

Maybe the code should explictly forbid the window to receive further events, at the time it is closed?


Side remark : While investigating this, I realized that the notify_window_destroy call in QTMWindow::closeEvent is probably too early because the user may later cancel the window closing in case the buffer was not saved. Yet deleting the window name does not seem to be a serious issue in the AppImage...

 

I just tried this with the Qt4 version from the website (from the rpm for Fedora) and the Qt5 AppImage version by Philippe. I can't reproduce it in either. It may be something about my particular version of Qt.

I get the following on the terminal:

------------------------------------------------------------------------------
TeXmacs] With linked TrueType support
Throwing segmentation fault
-------------------------------------------------
Error message:
  segmentation fault

System information:
  TeXmacs version  : 2.1
  Built by         : jeroen
  Building date    : 2022-01-27T09:48:56
  Operating system : linux-gnu
  Vendor           : pc
  Processor        : x86_64
  Crash date       : Thu 27 Jan 09:50:42 GMT 2022

Editor status:
  Root path          : [ 9 ]
  Current path       : [ 9, 0, 0 ]
  Shifted path       : [ 9, 0, 0 ]
  Physical selection : [ 9, 0, 0 ] -- [ 9, 0, 0 ]

Backtrace of C++ stack:
  texmacs.bin() [0x5e6206]
  texmacs.bin() [0x4a5137]
  texmacs.bin() [0x475650]
  /lib64/libc.so.6 : () + 0x42750
  texmacs.bin() [0x7c70e9]
  /lib64/libQt5Widgets.so.5 : QWidget::event(QEvent*) + 0xc8c
  /lib64/libQt5Widgets.so.5 : QApplicationPrivate::notify_helper(QObject*, QEvent*) + 0x83
  texmacs.bin() [0x4819d2]
  /lib64/libQt5Core.so.5 : QCoreApplication::notifyInternal2(QObject*, QEvent*) + 0x128
  /lib64/libQt5Widgets.so.5 : () + 0x20bc6a
  /lib64/libQt5Widgets.so.5 : () + 0x20da24
  /lib64/libQt5Widgets.so.5 : QApplicationPrivate::notify_helper(QObject*, QEvent*) + 0x83
  texmacs.bin() [0x4819d2]
  /lib64/libQt5Core.so.5 : QCoreApplication::notifyInternal2(QObject*, QEvent*) + 0x128
  /lib64/libQt5Gui.so.5 : QGuiApplicationPrivate::processGeometryChangeEvent(QWindowSystemInterfacePrivate::GeometryChangeEvent*) + 0x201
  /lib64/libQt5Gui.so.5 : QWindowSystemInterface::sendWindowSystemEvents(QFlags<QEventLoop::ProcessEventsFlag>) + 0xac
  /lib64/libQt5XcbQpa.so.5 : () + 0x6f11e
  /lib64/libglib-2.0.so.0 : g_main_context_dispatch() + 0x1a0
  /lib64/libglib-2.0.so.0 : () + 0xaa208
  /lib64/libglib-2.0.so.0 : g_main_context_iteration() + 0x33
  /lib64/libQt5Core.so.5 : QEventDispatcherGlib::processEvents(QFlags<QEventLoop::ProcessEventsFlag>) + 0x68
  /lib64/libQt5Core.so.5 : QEventLoop::exec(QFlags<QEventLoop::ProcessEventsFlag>) + 0x132
  /lib64/libQt5Core.so.5 : QCoreApplication::exec() + 0x94
  texmacs.bin() [0x7d1af8]
  texmacs.bin() [0x7d1b55]
  texmacs.bin() [0x479104]
  texmacs.bin() [0x4c8f88]
  /lib64/libguile.so.17 : () + 0x71653
  /lib64/libguile.so.17 : () + 0x3f5be
  /lib64/libguile.so.17 : scm_c_catch() + 0x161
  /lib64/libguile.so.17 : scm_i_with_continuation_barrier() + 0xb5
  /lib64/libguile.so.17 : scm_c_with_continuation_barrier() + 0x44
  /lib64/libguile.so.17 : scm_i_with_guile_and_parent() + 0x3a
  /lib64/libguile.so.17 : scm_boot_guile() + 0x39
  texmacs.bin() [0x4c8fdd]
  texmacs.bin() [0x47b5a6]
  /lib64/libc.so.6 : () + 0x2d560
  /lib64/libc.so.6 : __libc_start_main() + 0x7c
  texmacs.bin() [0x475575]

-------------------------------------------------
terminate called after throwing an instance of 'string'
Aborted (core dumped)

According to gdb the segmentation fault happens on line 57 of `QTMWindow.cpp` in side `void QTMWindow::moveEvent (QMoveEvent * event)`. This is with Qt 5.15.2.

The problem does not, in fact, seem to be related to unsaved changes. An apparently reliable way of triggering it is to open about 10 to 15 new blank documents in separate windows and then closing one of them.