bpo-36020: Remove snprintf macro in pyerrors.h

[vstinner]

On Windows, #include "pyerrors.h" no longer defines "snprintf" and
"vsnprintf" macros.

PyOS_snprintf() and PyOS_vsnprintf() should be used to get portable
behavior.

Replace snprintf() calls with PyOS_snprintf() and replace vsnprintf()
calls with PyOS_vsnprintf().

https://bugs.python.org/issue36020

[vstinner]

Previous attempts:

• PR bpo-36020: Define std::snprintf only for the previous versions of MSVC++ 14.0 #11913
• PR bpo-36020: snprintf definition #17980

[vstinner]

PyOS_snprintf() and PyOS_vsnprintf() should be used to get portable behavior.

https://docs.python.org/dev/c-api/conversion.html#c.PyOS_snprintf

"PyOS_snprintf() and PyOS_vsnprintf() wrap the Standard C library functions snprintf() and vsnprintf(). Their purpose is to guarantee consistent behavior in corner cases, which the Standard C functions do not."

[skrah]

Which snprintf implementations are broken?

[skrah]

The original PyOS_snprintf is from 2001. Is the stdlib snprintf really still broken on platforms?

[vstinner]

Which snprintf implementations are broken?

This PR is about removing "snprintf" and "vsnprintf" macros from Python.h.

I consider proposing a follow-up macro to deprecate PyOS_ functions and use directly snprintf() and vsnprintf() functions. But I didn't have time to dig into the history and try to understand the rationale behind these wrappers. So for now, it's safer to use PyOS_ wrappers.

Currently, PyOS_vsnprintf() has a fallback implementation with this:

    else if ((size_t)len >= size + _PyOS_vsnprintf_EXTRA_SPACE) {
        _Py_FatalErrorFunc(__func__, "Buffer overflow");
    }

Maybe this code is now dead since all platforms that we support are providing vsnprintf(). My plan is first to try to remove the fallback and see if it breaks any buildbot :-)

[vstinner]

Ah, also, this PR is intended to land into 3.9, so I prefer to take the safest approach for this PR.

For the follow-up PR, we can break stuff :-)

[skrah]

So for now, it's safer to use PyOS_ wrappers.

I kind of disagree, because snprintf has been extremely heavily tested in _decimal and the other function has not.

[vstinner]

I reverted changes in the _decimal module.

[skrah]

Thanks, I've definitely always compiled _decimal.c with -std=c99, so e.g. this passage in Python/mysnprintf.c makes me uneasy:

CAUTION: Unlike C99, str != NULL and size > 0 are required.

[miss-islington]

Thanks @vstinner for the PR 🌮🎉.. I'm working now to backport this PR to: 3.9.
🐍🍒⛏🤖

[bedevere-bot]

GH-20897 is a backport of this pull request to the 3.9 branch.