Description

mcepl

opened on Jun 17, 2026

Issue body actions

Bug report

Bug description:

As a followup to #151040 I am creating this separate ticket for i586. The traceback is exactly the same:

[ 3208s]   test_manual_unwind_finds_expected_frames (test.test_c_stack_unwind.ManualStackUnwindTests.test_manual_unwind_finds_expected_frames) (env={'PYTHON_JIT': '0'}) ... FAIL
[ 3208s] 
[ 3208s] ======================================================================
[ 3208s] FAIL: test_manual_unwind_finds_expected_frames (test.test_c_stack_unwind.ManualStackUnwindTests.test_manual_unwind_finds_expected_frames) (env={'PYTHON_JIT': '0'})
[ 3208s] ----------------------------------------------------------------------
[ 3208s] Traceback (most recent call last):
[ 3208s]   File "/home/abuild/rpmbuild/BUILD/python315-3.15.0_b2-build/Python-3.15.0b2/Lib/test/test_c_stack_unwind.py", line 285, in test_manual_unwind_finds_expected_frames
[ 3208s]     self.assertGreaterEqual(
[ 3208s]     ~~~~~~~~~~~~~~~~~~~~~~~^
[ 3208s]         python_frames,
[ 3208s]         ^^^^^^^^^^^^^^
[ 3208s]         STACK_DEPTH,
[ 3208s]         ^^^^^^^^^^^^
[ 3208s]         f"expected to find Python frames on {self.machine} with env {env}",
[ 3208s]         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[ 3208s]     )
[ 3208s]     ^
[ 3208s] AssertionError: 1 not greater than or equal to 10 : expected to find Python frames on i686 with env {'PYTHON_JIT': '0'}
[ 3208s] 
[ 3208s] ----------------------------------------------------------------------
[ 3208s] Ran 1 test in 0.276s
[ 3208s] 
[ 3208s] FAILED (failures=1)

Build log shows all used packages and steps required to be taken. The repository with the package source is https://src.opensuse.org/python-interpreters/python315

@vstinner on the other ticket commented:

That's a different architecture. If you care, you should open a separated issue for i586. You pass explicitly -fomit-frame-pointer which goes against PEP 831 (approved in Python 3.15).

Yes, we have -fomit-frame-pointer specifically for i586 architecture.

CPython versions tested on:

3.15

Operating systems tested on:

Linux

Activity

1. mcepl

   added

   type-bugAn unexpected behavior, bug, or errorAn unexpected behavior, bug, or error

   on Jun 17, 2026
2. vstinner commented on Jun 17, 2026

   

   vstinner

   on Jun 17, 2026

   Member

   More actions

   The test checks:

       if "no-omit-frame-pointer" in cflags:
           # For example, configure adds -fno-omit-frame-pointer by default on
           # supported GCC-compatible builds.
           return True
       if "omit-frame-pointer" in cflags:
           return False

   It checks for -fno-omit-frame-pointer first, even if -fomit-frame-pointer is defined after. Instead, we should check for flags in order and remembers the last option. See for example:

   def python_is_optimized():
       """Find if Python was built with optimizations."""
       cflags = sysconfig.get_config_var('PY_CFLAGS') or ''
       final_opt = ""
       for opt in cflags.split():
           if opt.startswith('-O'):
               final_opt = opt
       if sysconfig.get_config_var("CC") == "gcc":
           non_opts = ('', '-O0', '-Og')
       else:
           non_opts = ('', '-O0')
       return final_opt not in non_opts
3. vstinner commented on Jun 17, 2026

   

   vstinner

   on Jun 17, 2026

   Member

   More actions

   Yes, we have -fomit-frame-pointer specifically for i586 architecture.

   You don't want to implement PEP 831? Is there a reason to not implement it?
4. mcepl commented on Jun 17, 2026

   

   mcepl

   on Jun 17, 2026

   ContributorAuthor

   More actions

   Our GCC people persuaded me that we omit frame pointers only on archs like i586 for a reason, so in the end I added to ./configure command for 3.15 only

   %ifarch &ix86
       --without-frame-pointers \
   %endif

   I don’t think we care that much about the maximum performance and profiling on those architectures. We’ll see how it ends up working.

   Perhaps what this ticket can stand for in the end is that in this situation a user would get some better error/warning message, but &ix86 architectures are probably so inconsequential that it doesn’t matter that much.
5. picnixz

   added

   testsTests in the Lib/test dirTests in the Lib/test dir

   on Jun 17, 2026
6. vstinner commented on Jun 18, 2026

   

   vstinner

   on Jun 18, 2026

   Member

   More actions

   Oh. I forgot that _Py_WITH_FRAME_POINTERS macro was added recently to pyconfig.h. It's set to 1 by default.

   Can you test setting _Py_WITH_FRAME_POINTERS to 0 in pyconfig.h to see if it does fix test_c_stack_unwind?

Issue actions