PyUnstable_AtExit isn't well tested and undocumented

[ZeroIntensity]

Bug report

Bug description:

While working on #126908, @vstinner noted that this part of PyUnstable_AtExit looks wrong:

cpython/Modules/atexitmodule.c

Lines 45 to 47 in cef0a90

	else {
	state->last_ll_callback->next = callback;
	}

This will result in loss of callbacks after one has been stored, because the second-to-last one is always overwritten. Noted in gh-118915, PyUnstable_AtExit is also undocumented. I'll work on fixing both.

CPython versions tested on:

3.13, 3.14, CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-127791: Fix, document, and test PyUnstable_AtExit #127793
• [3.13] gh-127791: Fix, document, and test PyUnstable_AtExit (GH-127793) #127819

[disconnect3d]

This will result in loss of callbacks after one has been stored (...)

Hi, just out of curiosity: what does this exactly mean? Is this a memory leak or an UAF?

[ZeroIntensity]

Memory leak, I think? The next pointer in the linked list is overwritten.