Hope this is the right place to report the following issue.
Context: I'm using Advisor with a Python/C code (Python calls C functions to speed up expensive code sections).
Issue: I get non-deterministic deadlocks (the program stops executing) unless I disable threading. No problems w/o Advisor.
More insights: Initially, I (empirically) observed that no deadlock happening when setting OMP_NUM_THREADS=1. Then, I observed (again empirically) no deadlock w/o importing NumPy. My NumPy uses OpenBLAS underneath. So I tried disabling both OpenBLAS threading (OPENBLAS_NUM_THREADS=1) and Numaexpr threading (NUMEXPR_NUM_THREADS=1), leaving OMP_NUM_THREADS unset. Result: no deadlock (hence the thread title).
Unfortunately, I couldn't come up with a minimal reproducible example (or better, I do have one which is only five lines, but it requires importing my own Python modules). Before trying harder, I wonder whether there's an explanation off top of your head, or whether I'm doing something stupid (entirely possible)?
my command line is no special:
advixe-cl -data-limit=500 -collect survey -project-dir myprojdir -search-dir src:r=/tmp -run-pass-thru=--no-altstack -- python myprogram.py