bpo-45548: Remove _math.c workarounds for pre-C99 libm

[tiran]

The :mod:math and :mod:cmath implementation now require a C99 compatible
libm and no longer ship with workarounds for missing acosh, asinh,
expm1, and log1p functions.

The changeset also removes _math.c and moves the last remaining
workaround into _math.h. This simplifies static builds with
Modules/Setup and resolves symbol conflicts.

Co-authored-by: Mark Dickinson mdickinson@enthought.com
Co-authored-by: Brett Cannon brett@python.org
Signed-off-by: Christian Heimes christian@python.org

https://bugs.python.org/issue45548

Automerge-Triggered-By: GH:tiran

[brettcannon]

I created an alternative approach but just inlining into _math.h: #29181 .

[mdickinson]

I don't see it as a major problem. The functions are small.

Agreed. Looks fine to me. I somewhat prefer @brettcannon's approach (it feels odd to be including a .c file), but either seems fine, and it looks as though #29181 is having some issues with the Windows build.

[mdickinson]

A bit off-topic (and probably fodder for a separate PR), but it looks as though we ought to be able to get rid of most of the functionality in _math.c: if we can assume C99, then we can assume the existence of asinh, atanh, etc.

[mdickinson]

we ought to be able to get rid of most of the functionality in _math.c

Looking more closely, I think we can assume that all of HAVE_ACOSH, HAVE_ASINH, HAVE_ATANH, HAVE_EXPM1 and HAVE_LOG1P are defined. That would make almost all of _math.c redundant, with the sole exception of the extra zero-handling in _Py_log1p. But again, that can be fixed in a separate PR.

[tiran]

@mdickinson I implemented your suggestion, moved the log1p workaround into _math.h and deleted _math.c.

(Sorry for the squashed commit. I made an error while rewriting the commit message)

[mdickinson]

Mostly LGTM; two nitpicks and one question.