gh-132314: suppress maybe-uninitialized compiler warning

[StanFromIreland]

(Bonus: 3 space indentation :-)

Found in this buildbot log.

[vstinner]

Found in this buildbot log.

I don't see uninitialized newargs in warnings.

Python/ceval.c: In function '_PyEvalFramePushAndInit_Ex':
Python/ceval.c:1899:38: warning: 'stack_array' may be used uninitialized [-Wmaybe-uninitialized]
 1899 |     _PyInterpreterFrame *new_frame = _PyEvalFramePushAndInit(
      |                                      ^~~~~~~~~~~~~~~~~~~~~~~~
 1900 |         tstate, func, locals,
      |         ~~~~~~~~~~~~~~~~~~~~~         
 1901 |         newargs, nargs, kwnames, previous
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |     );
      |     ~                                 
Python/ceval.c:1822:1: note: by argument 4 of type 'const _PyStackRef *' to '_PyEvalFramePushAndInit' declared here
 1822 | _PyEvalFramePushAndInit(PyThreadState *tstate, _PyStackRef func,
      | ^~~~~~~~~~~~~~~~~~~~~~~
Python/ceval.c:1868:17: note: 'stack_array' declared here
 1868 |     _PyStackRef stack_array[8];
      |                 ^~~~~~~~~~~

[StanFromIreland]

I gathered that the error is because the compiler does not know if stack_array would ever be used to initialize newargs, and therefore assumed it might remain uninitialized by the time it's passed to _PyEvalFramePushAndInit. Maybe I am wrong?

You can call the buildbot !buildbot Raspbian (Or !buildbot ARM64 Raspbian though that may call all ARM64 buildbots too?) to verify.

[chris-eibl]

There is already this issue #132314 and PR #132376 for this false-positive warning.

[StanFromIreland]

I clicked the Update button accidentally somehow.

[StanFromIreland]

The pr proposed 2 different approaches, what do you think about this one?

[chris-eibl]

I do not think that your approach here can silence the maybe uninitialized warning on stack_array. Which is a false-positive and I still think it should be silenced using a diagnostic pragma, which PR #132376 is already taking care about.

[StanFromIreland]

@chris-eibl The other PR seems to have been forgotten, I propose we just ignore as normal then.

[chris-eibl]

Yeah, after all, your approach here deals just with clang, which is the only compiler that issues the warning so far, so most probably no need to go with something like _Py_COMP_DIAG_IGNORE_MAYBE_UNINITIALIZED.

[picnixz]

Is it right that this is attached to gh-132314? if so, please add that issue number.

[chris-eibl]

deals just with clang

Ups, I was too fast: in the issue and in your referred log

Found in this buildbot log.

gcc is raising this warning ...

[StanFromIreland]

@chris-eibl We have to move it, it did not work, see log.

[vstinner]

I'm unsure how to deal with this warning which seems to be only triggered on a specific buildbot with GCC 12 (gcc (Debian 12.2.0-14) 12.2.0). One solution would be to initialize stack_array like _PyStackRef stack_array[8] = {0};. But it can have a negative impact on performance. I'm not fully satisfied with this solution using compiler pragma since it's specific to the a compiler and it's a little bit "ugly" to me.

Another solution is to ignore this warning (don't pay attention to it) until the buildbot is upgraded since a new GCC version which doesn't emit the warning.

Yet another solution would be to build Python in debug mode on this buildbot, I expect that gcc -Og doesn't emit the warning.