AC changes look good, but I would try to investigate further which compilers/build environments this will affect (if any); buildbots do not cover everything.

AC changes look good, but I would try to investigate further which compilers/build environments this will affect (if any); buildbots do not cover everything.

It is unlikely as there is already code which uses variadic args and does not checks for HAVE_STDARG_PROTOTYPES and is working.

🤖 New build scheduled with the buildbot fleet by @kumaraditya303 for commit c9a6cc4 🤖

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

So far all buildbots passed, I added a news item.

cc @vstinner

@erlend-aasland: va_start macro requires two arguments1 and if you try to replace va_start(vargs, format); with va_start(vargs); then it won't even compile. I tried it and got this compilation error proving that va_start with one argument is wrong. Hence this PR is correct and fixes the issue.

Footnotes

1. https://www.cplusplus.com/reference/cstdarg/va_start/ ↩

Many files already call va_start() with 2 arguments: Objects/call.c, PC/launcher.c and PC/launcher2.c, Parser/pegen_errors.c, etc. (Without testing if the HAVE_STDARG_PROTOTYPES macro is defined.)

Please remove also this line from PC/pyconfig.h:

Many files already call va_start() with 2 arguments: Objects/call.c, PC/launcher.c and PC/launcher2.c, Parser/pegen_errors.c, etc. (Without testing if the HAVE_STDARG_PROTOTYPES macro is defined.)

Done

Merged, thanks.

This broke cython which appears to have been relying on HAVE_STDARG_PROTOTYPES to be defined by CPython. I am going to open a PR an the cython side to always go with the 2 argument version.