Correct usage of multi-phase init might now get users stuck when they start needing per-module state. See PEP 573 "Module State Access from C Extension Methods" for the (hopefully) last thing that prevents me from generally recommending multi-phase init. That PEP is on my list for after PRP 590 is done. Before it's implemented, I'm happy leaving PyModuleDef_Init to experts -- that is, have it in the reference docs only.
Changed target version as per Petr's comment (PEP 573 is close to being accepted for 3.9 - it just needs some editing to improve clarity in the PEP itself, rather than needing any changes to the technical proposal)
PEP 630 is now Active. I believe the vast majority of modules can now be ported, and it should be helpful for early adopters (including CPython stdlib). But there are still Open Issues and the API not battle-tested on all the edge cases. The PEP is written to be easily integrated into official docs when the time comes.
