gh-133548: fix handling of empty and 1-item tuples for sys.exit

[picnixz]

cc @markshannon@iritkatriel

• Issue: sys.exit unpacks its argument if it is a 0- or 1-element tuple #133548

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @picnixz for commit ae37e95 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F135789%2Fmerge

If you want to schedule another build, you need to add the 🔨 test-with-refleak-buildbots label again.

[iritkatriel]

Isn't this a problem with how PyErr_SetObject creates the exception? Shouldn't it be fixed there?

[picnixz]

I think it's a bit unfortunate because that's how _PyErr_CreateException handles tuples:

staticPyObject*_PyErr_CreateException(PyObject*exception_type, PyObject*value)
{
PyObject*exc;
if (value==NULL||value==Py_None) {
exc=_PyObject_CallNoArgs(exception_type);
}
elseif (PyTuple_Check(value)) {
exc=PyObject_Call(exception_type, value, NULL);
}
else {
exc=PyObject_CallOneArg(exception_type, value);
}
if (exc!=NULL&& !PyExceptionInstance_Check(exc)) {
PyErr_Format(PyExc_TypeError,
"calling %R should have returned an instance of ""BaseException, not %s",
exception_type, Py_TYPE(exc)->tp_name);
Py_CLEAR(exc);
}
returnexc;
}

Unless you want me to change _PyErr_CreateException to prevent the fast paths, it could be annoying for downstream users.

[vstinner]

Changing PyErr_SetObject() is a much larger change, it should be carefully prepared since it would affect many C extensions.

[iritkatriel]

It's not a "fast path" if removing it changes behaviour. We need to understand whether _PyErr_CreateException has a bug or not.

[picnixz]

I think it was designed as such for conveniency. It was added in 3a84097 but AFAICT, that was just refactoring of c0dc92a (1997) which already had this fast path.

[github-actions]

This PR is stale because it has been open for 30 days with no activity.

[vstinner]

I like this approach but I have remarks on the implementation.

[vstinner]

There is no need to mention SystemExit, the PR only changes sys.exit().

[vstinner]

exc can be NULL. You can use:

PyObject*exc=PyObject_CallOneArg(PyExc_SystemExit, status);
if (exc!=NULL) {
PyErr_SetObject(PyExc_SystemExit, exc);
Py_DECREF(exc);
}

I don't think that we need a special path if PyTuple_Check(). Use the same code path for any status.

[vstinner]

Changing PyErr_SetObject() is a much larger change, it should be carefully prepared since it would affect many C extensions.