[vstinner]

Use __extension__ on g++/clang++ to fix this g++ warning on Include/object.h:

error: ISO C++ prohibits anonymous structs [-Wpedantic]

• Issue: object.h:131:16: error: ISO C++ prohibits anonymous structs [-Wpedantic] #160208

[vstinner]

The change should be backported manually to 3.14 which doesn't have _Py_ANONYMOUS macro. I suppose that this macro should be backported to 3.14 as well.

[vstinner]

Without this change, two C++ tests of test_cext fails on OpenIndiana with g++ 14.3.0.

With this change, test_cext pass on the same machine.

[vstinner]

@mtelka: Would you mind to test this change?

cc @encukou

[vstinner]

This change is needed on OpenIndiana where <stdint.h> defines __STDC_VERSION__ macro even in C++.

[mtelka]

This change is needed on OpenIndiana where <stdint.h> defines __STDC_VERSION__ macro even in C++.

@vstinner Isn't this a bug in OpenIndiana (illumos) then? If so, it should be fixed there...

[mtelka]

This is related illumos bug report: https://www.illumos.org/issues/14920

[mtelka]

... and because of this: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57025#c13 I think this PR makes sense.

I'm just testing this change with Python 3.15.1 on OpenIndiana and I'll report back how it went.

[mtelka]

I can confirm that this PR fixes the test_cext failure on OpenIndiana.
Thank you!

[encukou]

I think we want a more robust approach here, given the comment https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57025#c4:

the C++ standard allows __STDC_VERSION to be defined by a C++ compiler.

Let me think about this a bit.

[vstinner]

Before my change, !(defined(__STDC_VERSION__) && __STDC_VERSION__ >= 201112L) is false with g++ on OpenIndiana and so __extension__ is not used.

With my change, (!defined(__STDC_VERSION__) || __STDC_VERSION__ < 201112L || defined(__cplusplus)) is true with g++ on OpenIndiana (since C++ compilers define __cplusplus macro), and so __extension__ is used as expected.

__extension__ is already used by Python with g++ and clang++ on all platforms but OpenIndiana.

I think we want a more robust approach here, given the comment https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57025#c4:

the C++ standard allows __STDC_VERSION to be defined by a C++ compiler.

Well, my change fix this exact issue.

[encukou]

Correction: turns out my other PR probably won't fix this issue -- that's about the _Py_ANONYMOUS definition, which is not used for the struct in object.h.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @encukou for commit 4fe34f2 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F160332%2Fmerge

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

[encukou]

Correction correction: Apparently my other PR will fix the issue, by coincidence. If not we'll work on top of it.
I will close this one. Thank you for the investigation, @mtelka & Victor.