You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
test_cuda_gdb_shows_nvrtc_debug_source_lines fails on L4T Orin (CUDA debugger back-end unavailable) #2711
Split out from #2693 to track only the cuda-gdb failure. The three test_graph_memory_resource.py::test_graph_alloc_with_output[*] failures in that
run stay with the parent issue / #2527.
Symptom
Automatic testing against CTK 13.3 (board local_l4t, Sun Aug 23 12:48:06 AM IST 2026,
first run that showed 4 failures instead of 3):
FAILED cuda_core/tests/test_program.py::test_cuda_gdb_shows_nvrtc_debug_source_lines - AssertionError: [Thread debugging using libthread_db enabled]
The abbreviated assertion message is just the first line of the captured cuda-gdb
output, which is why the report is not self-explanatory.
Analysis
The test was added in #2679. On this board cuda-gdb is on PATH, so the existing shutil.which("cuda-gdb") guard passes, but the debugger back-end is missing:
[Inferior 1 (process ...) exited normally]
Could not find CUDA Debugger back-end (libcudadebugger.so.1). Debugging is not possible
Without the back-end, set cuda break_on_launch application never stops the
inferior, so list never shows a device frame and the assertion on the NVRTC
temp .cu file name fails.
This is an environment/guard gap, not a regression in the NVRTC debug source
materialization from #2679: the three test_nvrtc_debug_* tests pass in the same run.
Fix
#2707 widens the guard so that Debugging is not possible is recognized and the
test reports xfail instead of failing.
Activity