Bug report

There have been changes to how the macros for casting PyObject pointers per #91959 (comment) by @vstinner any new compiler warnings are a regression.

There were two types of errors in greenlet, one related to const correctness which has been fixed via #92138) and one related to casting wrapper classes (that has not been fixed).

Per @serge-sans-paille in #92138 (comment)

That's an interesting situation: greenlets are written in C++, and the object that raises issues is of type greenlet::refs::OwnedErrPiece that has a conversion operator to PyObject* (namely: https://github.com/python-greenlet/greenlet/blob/be41e1a24925326b72a02ef5cb6d1ed9643eb062/src/greenlet/greenlet_refs.hpp#L924 ) Using C-style cast or static_cast directly uses that operator so that's fine. But a reinterpret_cast won't work.

and suggests a patch to CPython.

python-greenlet/greenlet#302 is a fix on the greenlet side.

The key line of the error is

src/greenlet/greenlet_refs.hpp: In member function ‘void greenlet::refs::PyErrPieces::normalize()’:
/home/tcaswell/.pybuild/bleeding/include/python3.12/pyport.h:30:25: error: invalid cast from type ‘greenlet::refs::OwnedErrPiece’ to type ‘const PyObject*’ {aka ‘const _object*’}
   30 |        const_cast<type>(reinterpret_cast<const type>(expr))
      |                         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This is also affecting python/pyperformance#198

A bunch of links: #92800 (comment)

Your environment

• CPython versions tested on: main branch
• Operating system and architecture: linux and x86_64