On Wed, Jul 01, 2026 at 11:44:02AM +0100, Colin Watson wrote:
On Wed, Jun 24, 2026 at 05:24:59PM +0100, Adam D. Barratt wrote:
python-memray FTBFS on armhf with a segfault. This has occurred on both
arm-conova-01 and arm-ubc-02.
Relevant-looking log extract:
tests/integration/test_object_tracking.py::test_object_tracking_in_threads
PASSED [ 10%]
tests/integration/test_object_tracking.py::test_object_stack_trace_with_native_traces
Fatal Python error: Segmentation fault
This stubbornly refuses to reproduce on a porterbox, so I'm
reluctantly just going to skip this test on armhf as there seems to be
plenty of precedent for that sort of thing in this case.
Ah, it would help if I ran the build with --arch=armhf, whoops.
This turns out to be another manifestation of
https://bugs.debian.org/984706, which also points to a workaround: set
UNW_ARM_UNWIND_METHOD=4. I'll do that. (Users of this package on armhf
might need that too in some cases; but I'm reluctant to patch the actual
Python code to set it itself as I don't fully understand all the
implications, and I think it's unlikely that all that many people are
doing serious Python memory profiling on armhf anyway.)
--
Colin Watson (he/him) [[email protected]]
