gh-91321: Add _testcppext C++ extension

[vstinner]

Build a C++ test extension to check that the Python C API is
compatible with C++ and does not emit C++ compiler warnings.

Add test_cppext.py and setup_testcppext.py to Lib/test/.

https://bugs.python.org/issue39355

[vstinner]

Quick & dirty hack of the Python C API (Include/ header files) to make it compatible with C++.

By the way, using NULL in a macro doesn't emit a warning even with clang -Wzero-as-null-pointer-constant. It's a clang feature! Extract of a LLVM clang comment:

// If it is a macro from system header, and if the macro name is not "NULL",
// do not warn

[iritkatriel]

https://bugs.python.org/issue39355 is closed. What is the status of this PR?

[vstinner]

This PR is now for https://bugs.python.org/issue47165

I'm working on fixing C++ compiler warnings: https://bugs.python.org/issue47164

[vstinner]

I updated this PR for #91320

[vstinner]

Oops, sorry, this PR is for #91321

[vstinner]

I created #91959 for the "CAST" commit.

[vstinner]

Oh, the build fails with MSVC:

Error: Modules/_testcppext.cpp(229): error C7555: use of designated initializers requires at least '/std:c++20'
test test_cppext failed
Error: Modules/_testcppext.cpp(244): error C7555: use of designated initializers requires at least '/std:c++20'
Error: Modules/_testcppext.cpp(364): error C7555: use of designated initializers requires at least '/std:c++20'

I initialize structures with .attr = value syntax:

static PyType_Spec Xxo_Type_spec = {
    .name = "_testcppext.Xxo",
    .basicsize = sizeof(XxoObject),
    .itemsize = 0,
    .flags = Py_TPFLAGS_DEFAULT | Py_TPFLAGS_HAVE_GC,
    .slots = Xxo_Type_slots,
};

[vstinner]

Tests / Windows (x64) (pull_request) Failing after 19m

Error:

LINK : fatal error LNK1104: cannot open file 'python311.lib'

[steve-s]

FYI we were/are trying to address the same issue with HPy headers. My understanding is that this only tests the static inline functions and macros that are used in the example extension. It is still possible that some other macro or static inline function could cause warnings or for macros even errors. So my conclusion so far has been that to have a fully C++ compat tested headers, one has to write a C++ extension(s) that use all the static inline functions and macros from the headers. Something tested is way better than nothing tested, so no objections to this PR -- just wanted to check if you see it the same way?

[vstinner]

@steve-s:

FYI we were/are trying to address the same issue with HPy headers. My understanding is that this only tests the static inline functions and macros that are used in the example extension. It is still possible that some other macro or static inline function could cause warnings or for macros even errors. So my conclusion so far has been that to have a fully C++ compat tested headers, one has to write a C++ extension(s) that use all the static inline functions and macros from the headers. Something tested is way better than nothing tested, so no objections to this PR -- just wanted to check if you see it the same way?

Sadly, my experience is the same: a C++ compiler doesn't complain when a static inline function is defined. Only when it's being used.

My practical plan is:

• Write a bare minimum C++ extension which does nothing and check in the Python test suite that we can build it for C++11.
• First, I tried to load the extension but my code didn't work when Python is installed. Maybe we can try to load and test the built C++ extension later. But for now, I prefer to skip that.
• Slowly test more and more static inline functions in this C++ extension.
• Test more C++ versions.

Right now, my main blocker issue is to test if Python has a working C++ compiler. For example, the "Tests / Windows (x64)" fails with cannot open file 'python311.lib'. This CI has a C++ compiler but not the required python library to link the C++ extension :-( Maybe it's just a matter of using the right compiler command line. I don't know.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @vstinner for commit 31b7a4f7ab6092998865e06dd0f4b0d5ea79b13f 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[vstinner]

I simplified Modules/_testcppext.cpp to make it as small as possible. I prefer to start with a small extension and then extend it slowly. As I wrote, right now, my practical issue is more to ensure that there is a working C++ compiler all on CIs used by Python.

[vstinner]

Oh, when Python is installed, the test fails with:

error: could not create 'build': Permission denied

For example, it failed on these buildbots:

• AMD64 Fedora Stable Clang Installed PR
• s390x Fedora Clang Installed PR

[vstinner]

I rewrote the PR to build the C++ extension in a temporary directory and fix the test on installed Python: _testcppext.cpp is now part of Lib/test/ and so is installed.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @vstinner for commit e067677 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[vstinner]

buildbot/s390x RHEL7 PR — Build done.

The error is unrelated to my change:

1 test altered the execution environment: test_threading

[encukou]

reinterpret_cast breaks existing code, see #94731.
#94782 is a PR to revert back to C-style casts.