Closed ctsa closed 1 year ago
Ok, thanks for the report. That was a really interesting one.
TBH, the code allowing match!=0 and ends-free is not ideal. This will improve greatly with the next major release (in speed and code robustness). Until then, I patched it.
Let me know.
Great, thank you! I confirm the issue appears to be fixed on my case.
Hi @smarco -
Thanks again for WFA2 and your prior help with a similar issue -- I have a new problem where I'm getting non-deterministic segfaults in client code. I traced this down to one issue in wfa2 which is likely to be the cause, a
Conditional jump or move depends on uninitialised value
here:https://github.com/smarco/WFA2-lib/blob/9fdf46efd95bcac751adae0a57b44c2e80e6eef7/wavefront/wavefront_extend_kernels.c#L143
I reduced the problem case to the following code:
After building this in linux debug mode against the library version currently on main (9fdf46e), the valgrind output is:
I tried to pursue this a bit further, and could at least figure out that the initialization problem seems to be specific to wavefront 2.
If you're able to reproduce and fix (or explain a workaround for) this case it would be very helpful. Thank you!