test_initconfig_get_api fails on platforms where malloc(0) returns NULL

[zuhu2195]

Bug report

Bug description:

test_initconfig_get_api (Programs/_testembed.c) executes the code

    assert(PyInitConfig_GetStrList(config, "xoptions", &length, &items) == 0);
    assert(length == 0);

PyInitConfig_GetStrList has malloc

*items = malloc(list->length * sizeof(char*));

if length is zero the behavior of malloc(0) is implementation defined, this seems to be problematic on platforms where malloc of zero returns NULL.

CPython versions tested on:

3.14

Operating systems tested on:

Other

[picnixz]

What's the issue of having items a nullptr then?

[picnixz]

Ah I see, the relevant lines were not included in your description:

    *items = malloc(list->length * sizeof(char*));
    if (*items == NULL) {
        config->status = _PyStatus_NO_MEMORY();
        return -1;
    }

[picnixz]

I think the issue is broader. I'm pretty sure there are other places where we assume malloc() returning NULL means an allocation failure without us actually checking for the size itself. Can we broaden the scope of this issue for such cases instead? that is, can you find the malloc() places where it's possible to pass a 0 as the size and where we would still treat a NULL as being an OOM issue?

[zuhu2195]

Yes, I should be able to do that on running a full regression on my system. meanwhile AI scan for such similar issues should also help.

[picnixz]

Yeah AI would be good for this specific task. But we should revies each finding carefully though.