Add a way of printing a C backtrace to `faulthandler`

[jakkdl]

Feature or enhancement

Proposal:

The faulthandler module allows registering a SIGSEGV handler to print a Python stacktrace if the program encounters a segfault. However, if developing C/C++ extension modules that may not be particularly useful on its own, and you also want the the C stacktrace.
The suggested API would be a kwarg to faulthandler.enable().

Implementation could use https://github.com/timmaxw/cfaulthandler as a starting point, timmaxw/cfaulthandler@561dbdd in particular.

The availability/usability of the feature would likely depend on platform and/or compile flags.

Has this already been discussed elsewhere?

I have already discussed this feature proposal on Discourse

Links to previous discussion of this feature:

https://discuss.python.org/t/print-c-stacktrace-with-faulthandler/56834
where @gpshead approved of opening a feature request

Linked PRs

• gh-127604: Add C stack dumps to faulthandler #128159
• gh-127604: Don't rely on dprintf() for faulthandler C stacks #132800
• gh-127604: Fix refleak in faulthandler_dump_c_stack_py() #132840
• gh-127604: Allow faulthandler to dumpC stack on MacOS #132841
• gh-127604: Replace dprintf() with _Py_write_noraise() #132854
• gh-127604: Only define dump_pointer() if CAN_C_BACKTRACE #132897
• gh-127604: ensure -ldl is passed to the linker when dladdr1 is found #133040
• gh-127604: ensure -ldl is passed only once to the linker #133071
• gh-127604: Optimize -ldl usage on platforms that use dlopen for libFFI. #133081

[ZeroIntensity]

Would you like to submit the starting PR?

[jakkdl]

Would you like to submit the starting PR?

I might give it a try, but if anybody else wants to you're super welcome. I'm not especially well versed with faulthandler myself.

[ZeroIntensity]

I went ahead and tried this myself: #128159

[gpshead]

For future reference as I expect people may want something better in the future beyond what we're able to provide via this issue: https://github.com/abseil/abseil-cpp/tree/master/absl/debugging has code that does a nice job of emitting symbolized C/C++/Rust stack traces on failure. Also: a lot more complicated.

[pablogsal]

We do a similar thing in https://github.com/bloomberg/pystack where we have a custom DWARF parser. Also: a lot more complcated as well :)

[serhiy-storchaka]

dprintf() is not available on OpenIndiana.

Python/traceback.c: In function '_Py_backtrace_symbols_fd':
Python/traceback.c:1213:13: error: implicit declaration of function 'dprintf'; did you mean 'wprintf'? [-Wimplicit-function-declaration]
 1213 |             dprintf(fd, "  Binary file '<unknown>' [%p]\n", array[i]);
      |             ^~~~~~~
      |             wprintf

[ZeroIntensity]

Will put up a fix later today.

[kulikjak]

Hi, not sure whether this is a platform difference or some other issue, but on Oracle Solaris, I see all test_dump_c_stack/test_dump_c_stack_file fail with:

======================================================================
FAIL: test_dump_c_stack (test.test_faulthandler.FaultHandlerTests.test_dump_c_stack) (line='  Binary file "/builds/python/build/amd64/libpython3.14.so.1.0", at _PyEval_EvalFrameDefault+0x96a1 [7fe012df2091]')
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/builds/python/cpython-main/Lib/test/test_faulthandler.py", line 971, in check_c_stack
    self.assertRegex(line, C_STACK_REGEX[1])
    ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError: Regex didn't match: '(  Binary file ".+"(, at .*(\\+|-)0x[0-9a-f]+)? \\[0x[0-9a-f]+\\])|(<.+>)' not found in '  Binary file "/builds/python/build/amd64/libpython3.14.so.1.0", at _PyEval_EvalFrameDefault+0x96a1 [7fe012df2091]'

presumably because the address in square brackets is not prefixed with 0x.

When I change ADDRESS_EXPR = "0x[0-9a-f]+" to "(0x)?[0-9a-f]+", the issue goes away.

[ZeroIntensity]

I'd like to prevent conflicts, so feel free to make a PR once #132800 gets merged.