BUG: Support using libunwind for backtrack #22152
Conversation
Some system (e.g. musl) do not have "execinfo.h", and the backtracking is provided by libunwind. Fix: numpy#22084
| #include <execinfo.h> | ||
| #elif defined HAVE_LIBUNWIND_H | ||
| #include <libunwind.h> | ||
| #endif |
There was a problem hiding this comment.
Thanks, this makes sense to me now, although if neither is found, I guess the whole thing should be disabled (which is on line 61)
There was a problem hiding this comment.
Yes, HAVE_BACKTRACE should disabled those include if backtrack() not found. Unless new backtracking library showed up, which should throw an error to notice devs.
There was a problem hiding this comment.
The test for backtrace we are using is in numpy/core/feature_detection_misc.h:
#include <stddef.h>
int backtrace(void **, int);
int madvise(void *, size_t, int);It's not entirely clear to me how backtrace ends up usable when including stddef.h, but it does seem to work. And libunwind indeed has backtrace, as an alias to its own unw_backtrace: https://github.com/dropbox/libunwind/blob/16bf4e5e498c7fc528256843a4a724edc2753ffd/src/mi/backtrace.c#L58-L79
So this PR looks like an improvement. It solves a build issue, so let's just merge it as is. If we find another case of backtrace existing but not being provided by either execinfo.h or libunwind.h, let's deal with a bigger reshuffle of this code at that point.
Some system (e.g. musl) do not have "execinfo.h", and the backtracking
is provided by libunwind.
Fix: #22084