Repository navigation
Description
Summary
Official OpenNOW Qt 1.0.4 Linux x64 AppImage intermittently aborts while closing on Fedora 44 / KDE Plasma Wayland / NVIDIA 610.57.04. In a controlled idle launch-and-close series, 6 normal exits were followed by 1 SIGABRT on attempt 7. The final application log message was:
2026-11-10T12:32:51.939 fatal default: QFontDatabase: Must construct a QGuiApplication before accessing QFontDatabase
This is a different crash signature from the earlier Wayland/NVIDIA SIGSEGV reported in #1329. This report does not claim that the Wayland Vulkan swapchain fix in #1361 is ineffective; it documents a new observed SIGABRT needing independent investigation.
Environment
- OS: Fedora Linux 44, x86_64
- Desktop/session: KDE Plasma, Wayland (
XDG_SESSION_TYPE=wayland) - Hardware: Lenovo Legion 5, hybrid GPU laptop, NVIDIA GeForce RTX 3060 Laptop GPU
- NVIDIA proprietary driver: 610.57.04
- Application: official stable OpenNOW Qt 1.0.4, Linux x64 AppImage
- AppImage SHA-256 (verified):
402b1ed0fc37c4bb25b2e3aa99007d9fad5ca61669ce88c7ca937555a61fff10 - Incident: 2026-11-10 12:32:51 CET
systemd-coredump: SIGABRT (signal 6), shell exit status 134
Observed reproduction / limits
- Start the official 1.0.4 AppImage from a terminal.
- Allow the UI to load. No stream was intentionally started for this series.
- Close the application normally using the window close button (X).
- Repeat. A 10-attempt series was planned, but the harness stopped on the first abnormal exit.
Observed results:
| Attempt | Exit code | Duration |
|---|---|---|
| 1 | 0 | 10 s |
| 2 | 0 | 7 s |
| 3 | 0 | 5 s |
| 4 | 0 | 14 s |
| 5 | 0 | 11 s |
| 6 | 0 | 14 s |
| 7 | 134 (SIGABRT) | 20 s |
Only one crash has been observed in this series; consistent reproduction is not yet established. No post-stream shutdown acceptance test has been performed yet.
Expected vs actual
Expected: Closing the application exits cleanly (status 0), and Qt Quick/QML teardown does not access GUI-only services after the QGuiApplication context is unavailable.
Actual: Attempt 7 was aborted by Qt, producing a systemd core dump and the fatal QFontDatabase message.
Symbolized backtrace (main thread, sanitized)
The initial core inspection missed symbols because the AppImage's ephemeral /tmp/.mount_... path had disappeared. The same SHA-256-verified AppImage was extracted locally, and its bundled Qt libraries were supplied to GDB. Relevant symbols from the resulting trace:
#0 __pthread_kill_implementation
#1 raise
#2 abort
#5 QMessageLogger::fatal(char const*, ...) const
#7 QFontDatabasePrivate::findFont(QFontDef const&, int, bool)
#8 QFontDatabasePrivate::load(QFontPrivate const*, int)
#9 QFontPrivate::engineForScript(int) const
#10 QFontMetricsF::height() const
#11 QQuickTextPrivate::lineHeightOffset() const
#12 QQuickText::contentHeight() const
#13 QQuickTextPrivate::updateSize()
#14 QQuickText::geometryChange(QRectF const&, QRectF const&)
#15 QQuickItem::setWidth(double)
#19 QPropertyBindingPrivate::evaluateRecursive(...)
#20 QQmlPropertyBindingJS::expressionChanged()
#21 QQmlNotifier::emitNotify(...)
#23 QQuickItem::parentChanged(QQuickItem*)
#24 QQuickItem::setParentItem(QQuickItem*)
#25 QQuickItem::~QQuickItem()
#27 QObjectPrivate::deleteChildren()
#28 QObject::~QObject()
#30 QObjectPrivate::deleteChildren()
#31 QObject::~QObject()
#32 runApplicationSession(int, char**, QString&)
#33 runApplication(int, char**)
The stack indicates that QML property-binding reevaluation occurred during QQuickItem/QObject destruction. It attempted text metric calculation via QFontDatabase, leading to Qt's fatal check. This identifies the call path, not the ultimate source-level defect or proof of precisely when the GUI application became unavailable.
Potentially relevant code (stable v1.0.4): ApplicationStartup.cpp constructs QGuiApplication and QQmlApplicationEngine, then after application.exec() explicitly deletes engine root objects before returning. A QML binding/child-destruction interaction during shutdown may merit investigation; this is a hypothesis, not a confirmed patch diagnosis.
Requested follow-up
- Review Qt Quick/QML object destruction, property bindings and text metrics during shutdown.
- Check whether any cleanup path permits QML text bindings to run after the relevant GUI context is no longer available.
- If a defect is confirmed, add a regression test for repeated idle close on Linux/Wayland.
- I can provide additional sanitized diagnostics or rerun targeted acceptance tests.
Privacy: No raw core dump, private session credentials, user paths, host identifiers or tokens are attached.
Activity
wojko6 commentedon Nov 12, 2026
Retest results — official stable OpenNOW 1.0.5 (2026-11-12)
Thank you for the QML engine teardown/lifetime fix in #1443. I retested the official OpenNOW Qt 1.0.5 Linux x64 AppImage on the same Lenovo Legion 5 / NVIDIA RTX 3060 Laptop GPU, using KDE Plasma on Wayland.
Important environment change: the original SIGABRT in this issue was observed on Fedora 44 with NVIDIA 610.57.04; this retest was performed after upgrading to Fedora 45. Therefore, this is a positive acceptance test on the current setup, not a controlled same-OS comparison or proof that the bug is universally eliminated.
Results
1c219105e630048b64b55d02c0584e4255b01b69cb82e8e70603928a6e9310e6.EXIT_CODE=0in each scripted iteration); 0 failures. Windows were closed normally with the X button, without starting a stream during these ten runs.For reference, on v1.0.4 the earlier controlled idle-close sequence encountered a SIGABRT on attempt 7/10, with the
QFontDatabasefatal message.Outcome: PASS for the 10-cycle idle shutdown regression on Fedora 45 / KDE Wayland / RTX 3060. No recurrence of the reported SIGABRT was observed in these tests. This is encouraging evidence for the fix, while the OS change and limited sample size remain important qualifications.
I suggest keeping the issue open until you have reviewed the results or confirmed whether any further reproduction is needed. I can run additional targeted checks and provide sanitized diagnostics if useful.