GeezerVison

The following extensions avoids the internal C API by defining the macro Py_ARGUMENT_CLINIC_AVOID_INTERNAL_CAPI:
Modules/_csv.c
Modules/_multiprocessing/posixshmem.c
Modules/_testcapi/exceptions.c
Modules/grpmodule.c
Modules/syslogmodule.c

If we convert these modules to the limited C API, see issue #85283, the Py_ARGUMENT_CLINIC_AVOID_INTERNAL_CAPI macro can be removed (from AC and from these modules).

Oh, global objects are outdated in the main branch, see PR #108714.

Hum. Windows fails to build Modules\_testinternalcapi.c:

Error: D:\a\cpython\cpython\Modules\clinic\_testinternalcapi.c.h(30,20): error C2099: initializer is not a constant [D:\a\cpython\cpython\PCbuild\_testinternalcapi.vcxproj]
Error: D:\a\cpython\cpython\Modules\clinic\_testinternalcapi.c.h(31,20): error C2099: initializer is not a constant [D:\a\cpython\cpython\PCbuild\_testinternalcapi.vcxproj]
Error: D:\a\cpython\cpython\Modules\clinic\_testinternalcapi.c.h(85,20): error C2099: initializer is not a constant [D:\a\cpython\cpython\PCbuild\_testinternalcapi.vcxproj]
Error: D:\a\cpython\cpython\Modules\clinic\_testinternalcapi.c.h(86,20): error C2099: initializer is not a constant [D:\a\cpython\cpython\PCbuild\_testinternalcapi.vcxproj]
Error: D:\a\cpython\cpython\Modules\clinic\_testinternalcapi.c.h(152,20): error C2099: initializer is not a constant [D:\a\cpython\cpython\PCbuild\_testinternalcapi.vcxproj]
Error: D:\a\cpython\cpython\Modules\clinic\_testinternalcapi.c.h(153,20): error C2099: initializer is not a constant [D:\a\cpython\cpython\PCbuild\_testinternalcapi.vcxproj]
Error: D:\a\cpython\cpython\Modules\clinic\_testinternalcapi.c.h(211,20): error C2099: initializer is not a constant [D:\a\cpython\cpython\PCbuild\_testinternalcapi.vcxproj]
Error: D:\a\cpython\cpython\Modules\clinic\_testinternalcapi.c.h(212,20): error C2099: initializer is not a constant [D:\a\cpython\cpython\PCbuild\_testinternalcapi.vcxproj]

Related code, I added error C2099: initializer is not a constant comments:

    static struct {
        PyGC_Head _this_is_not_used;
        PyObject_VAR_HEAD
        PyObject *ob_item[NUM_KEYWORDS];
    } _kwtuple = {
        .ob_base = PyVarObject_HEAD_INIT(&PyTuple_Type, NUM_KEYWORDS)  // error C2099: initializer is not a constant
        .ob_item = { &_Py_ID(doc), },  // error C2099: initializer is not a constant
    };

This approach doesn't work. I abandon this PR in favor of PR #108726 which is implemented differently.

Victor, next time can you use a draft PR for such experiments so 17 core devs aren't bombarded with review requests and internal monologue?

Victor, next time can you use a draft PR for such experiments so 17 core devs aren't bombarded with review requests and internal monologue?

If I create a PR as draft, no notification is sent?

Well, I expected this PR to be ready, but then I saw that tests failed on Windows. I wrote a new PR from scratch with a different approach to avoid this.

I'm not sure why so many developers are requested for reviews, the vast majority of modified files are generated files (*/clinic/*.h files generated by Argument Clinic).

If I create a PR as draft, no notification is sent?

yes, code owners are only requested for review when you take it out of draft mode

I'm not sure why so many developers are requested for reviews, the vast majority of modified files are generated files (*/clinic/*.h files generated by Argument Clinic).

I don't think the GitHub CODEOWNERS feature takes any notice of whether files are generated or not when figuring out whether people should be pinged for review -- it just looks at whether the modified files match any of the globs provided in https://github.com/python/cpython/blob/main/.github/CODEOWNERS