GeezerVison

I would prefer to open a separated issue on bugs.python.org for this work.

@vstinner The static library will contain '.c' files that reference internal implementations FOR the runtime it's designed for.
The goal of the static library is that end users link against it when building their extensions WITHOUT having access to internals directly.

As it currently stands, the current Py_TYPE still requires direct access to internals for end users while this static library remove that requirement.

example would be:
Py_TYPE macro references a method called "PyObject_GetType", but the definition of the method is not in the header, instead, it's in the static library which will be linked into the final binary.

This allows for implementation independent access while still allowing the C compiler to aggressively inline out the small functions.

@vstinner Before I proceed any further, I want you to review "slib_pythoncore" first.

The goal of the static library is that end users link against it when building their extensions WITHOUT having access to internals directly.

This is the purpose of the limited C API which provides a stable ABI. It's already implemented since Python 3.2. I don't see the point of adding a second flavor of the "libpython" library.

As it stands, functions like _Py_TYPE, _Py_SIZE, _Py_INCREF, _Py_DECREF rely on "PyObject" being a complete type. Opaque objects on the other hand are classified as incomplete types.

Incomplete types can't be used for anything but pointers (*) and references (&), these functions will fail when PyObject goes opaque in the limited API, infact, sizeof() also fails on incomplete types.

The purpose of the static library is not to be a different flavour of Python, but to act as a bridge into Python's internals while respecting the limitations of incomplete types, while also eliminating function call overhead.

As an added bonus, this static library can be "reskinned" for runtimes that take a different approach to PyObject (if any) and descendant built-in types while end users don't have to worry about what the runtime's schematic is.

Added a basic implementation of the flag, but I do not know where all usages of "tp_basicsize" is, I'm also debating on whether or not I need to make GET_TYPE_TOTALSIZE into an inlinable function.

Py_TPFLAGS_USES_OPAQUE_OBJECT looks useful, but other changes like "slib" looks useless to me.

A Python core developer has requested some changes be made to your pull request before we can consider merging it. If you could please address their requests along with any other requests in other reviews from core developers that would be appreciated.

Once you have made the requested changes, please leave a comment on this pull request containing the phrase I have made the requested changes; please review again. I will then notify any core developers who have left a review that you're ready for them to take another look at this pull request.

Consider the following:

struct myobject;

void use_myobject(struct myobect *obj); // Valid, incomplete types can be used as pointers.

size_t get_object_value(struct my object *obj)
{
    return obj->m_value; // ERROR: my_object is an incomplete type!
}

Above demonstrates why functions like _Py_Type will fail in limited api.

'Py_TPFLAGS_OMIT_OBJECT_SIZE' was intended as the stepping stone for making types like PyTypeObject opaque as well, this makes knowing the true offset off the base time a runtime dependant scenario, one with minimal impact to performance, probably less than having _Py_TYPE, etc, exported from python binary.

@vstinner One way or another, PY_TYPE and co need to have their implementations PURGED from the Python header files for a truly opaque PyObject.

Another thing that needs to be purged is "ob_base" usage of objects like PyTupleObject, PyTypeObject, PyListObject to name a few.

I'm willing to replace slib with a precompiled header trio, but that's a can of worms on its own (at least on MSVC).

I'm not budging in regards to "tp_obj_offset" and "tp_obj_size", see above mentioned purging.