Exit code 134 is SIGABRT from a native library, not a Python traceback, so the software's own logs/ stay empty by design: that log is only created after the main window starts. Since even stdbuf ... | tee captured nothing, the abort happens before Python writes anything to stderr, so the Python-level handler must be enabled explicitly or the crash caught at the OS level.
Run these in order and stop at the first step that behaves differently:
Baseline interpreter check (rules out uv/Python itself):
cd ~/pyvideotrans
uv run python -X faulthandler -V
uv run python -X faulthandler -c "print('ok')"; echo "rc=$?"
If rc=134 already here, the abort is in the interpreter or its loader, not pyVideoTrans.
Force Python to dump the aborting stack (this is the key step — it prints a traceback on SIGABRT even with no buffered output):
uv run python -X faulthandler sp.py
echo "rc=$?"
Equivalent form if the above prints nothing: PYTHONFAULTHANDLER=1 uv run sp.py.
If still no output, capture it at the OS level:
ulimit -c unlimited
uv run python -X faulthandler sp.py; echo "rc=$?"
dmesg | tail -40
journalctl -xe | tail -40
Look for lines containing traps:, Aborted, segfault at, or a library path (e.g. libQt6Core, libtorch, libonnxruntime) — that names the aborting component.
Get a real backtrace with gdb:
sudo apt-get install -y gdb
gdb -batch -ex run -ex bt --args .venv/bin/python sp.py 2>&1 | tail -40
The top frames identify the library that calls abort().
Bisect the heavy native imports (very common cause of silent SIGABRT on minimal distros: a C++ extension needing a newer glibc or an unsupported CPU instruction set):
uv run python -X faulthandler -c "import PySide6; print('pyside6 ok')"
uv run python -X faulthandler -c "from PySide6.QtWidgets import QApplication; QApplication([]); print('qt ok')"
uv run python -X faulthandler -c "import torch; print('torch ok')"
uv run python -X faulthandler -c "import onnxruntime; print('onnx ok')"
ldd --version | head -1
Check the runtime environment variables Qt aborts on:
echo "DISPLAY=$DISPLAY WAYLAND_DISPLAY=$WAYLAND_DISPLAY XDG_RUNTIME_DIR=$XDG_RUNTIME_DIR"
df -h .
free -h
ls -ld /dev/shm
If DISPLAY is empty (headless or plain SSH without -X), Qt aborts immediately with no message in some builds.
Confirm uv sync actually finished cleanly — a half-installed environment also aborts silently:
uv sync
uv run python -c "import PySide6, torch, onnxruntime; print('all imports ok')"
Also verify Trisquel's base glibc supports the wheels uv installed; ldd --version output from step 5 is required for that check.
If no step produces any text at all, report: the output of steps 1, 4, 5, 6, plus ldd --version | head -1 and whether the machine is headless/remote. Without at least the faulthandler traceback or the gdb backtrace, the aborting library cannot be identified.
Please refer to the related documentation: