Adding Py_HashBuffer to the Stable ABI

[vstinner]

It was proposed to add Py_HashBuffer() to the Stable ABI. API:

Py_hash_t Py_HashBuffer(const void *ptr, Py_ssize_t len)
This function always succeeds.

The function is properly documented and tested (by test_capi.test_hash). It has a well defined API. IMO it's a good candidate for the stable ABI.

The function was added to Python 3.14 to replace the private _Py_HashBytes() API.

Vote:

• @encukou
• @erlend-aasland
• @serhiy-storchaka
• @vstinner
• @zooba

[encukou]

This function always succeeds.

The function should be allowed to fail; (return -1 with an error set; -1 is the reserved for errors in hash function).

Can we keep infallibility as a CPython implementation detail only?

[vstinner]

Py_HashPointer() and PyObject_GenericHash() have the same property: they cannot fail.

Can we keep infallibility as a CPython implementation detail only?

I'm not sure that it's useful. The function is quite basic and I don't think that an implementation can fail. Why would it fail?

[encukou]

If you pass NULL, a negative size, or the memory doesn't belong to you, raising an exception would be more user-friendly than what CPython does (crashing or giving a “wrong” answer).

And there's the ever-present possibility of a (deprecation) warning turned into an error.