gh-106307: Fix PyMapping_GetOptionalItemString()

[serhiy-storchaka]

The resulting pointer was not set to NULL if the creation of a temporary string object was failed.

The tests were also missed due to oversight.

[vstinner]

LGTM. I just propose adding a comment on the new macro.

Does Valgrind or similar tool detect the usage of uninitialized memory?

[vstinner]

LGTM.

The _testcapi now intializes the variable used in the test:

static PyObject *
mapping_getoptionalitemstring(PyObject *self, PyObject *args)
{
    PyObject *obj, *value = UNINITIALIZED_PTR;

[vstinner]

@erlend-aasland: Idea for the devguide: We should suggest to always initialized the &value parameter to NULL in case of errors. I just avoid any risk of using an uninitialized variable by misusing the C API. What do you think?

[erlend-aasland]

@erlend-aasland: Idea for the devguide: We should suggest to always initialized the &value parameter to NULL in case of errors. I just avoid any risk of using an uninitialized variable by misusing the C API. What do you think?

I have a PR for this, and you already approved it twice 😃 python/devguide#1128

[serhiy-storchaka]

There are exceptions, like _PyEval_SliceIndex(), where the output parameter should not be set in some cases.

[erlend-aasland]

Exceptions are fine; guidelines are for the general case, not the special case.

[vstinner]

I have a PR for this, and you already approved it twice 😃 python/devguide#1128

Alright, I should approve it 3 or 4 times 👍👍👍👍