This looks less safe to me, because now it can't be used in an expression like (Py_SET_TYPE(x, y), z), which is legal in 3.10.

I am going to back out of this discussion as I am out of my depth on c/c++ details and let you and Victor sort this out :)

Let's leave this open to track getting in line with python - we should match what ultimately gets merged in that PR.

Would it be OK to make it a milestones issue rather?

@seberg, this is valuable as an open PR because it means it's super easy to edit in-browser :)

This looks less safe to me, because now it can't be used in an expression like (Py_SET_TYPE(x, y), z), which is legal in 3.10.

Do you mean that python/cpython#20610 should suggest to use #define Py_SET_SIZE(obj, size) (Py_SIZE(obj) = (size)) ?

The variant I suggested over there using (void)0 is better than both the original here and the original there. Our version was unsafe because it allowed Py_SET_TYPE(x, y) = z, which obviously won't work in 3.9. Your version was overly strict. The version that is now in both patches should be exactly the right level of strict.

Did not expect as fast turn-around... I am half expecting that in the end, there may be some python header we can vendor in here instead. But for now all good as soon as tests pass...

I don't mind these small changes there are easy enough, just if there is a mass of tiny macro fixes we backport one-by-one it will get annoying.

Did not expect as fast turn-around... I am half expecting that in the end, there may be some python header we can vendor in here instead.

I started a thread on the capi-sig list to propose exactly that:
https://mail.python.org/archives/list/[email protected]/thread/HUCHXBYCHL6ZXPZYOEEATXTKMQLAYVEA/