Christian's general approach looks fine to me - consolidating the "kind" hashes (i.e. byte sequences, numbers and pointers) into one place independent of any particular type implementation makes sense to me, and the clear abstraction of "What is a hash function?" from Python's point of view is a good thing for embedding purposes.

If you get the PEP updated accordingly, we should be able to get that formally accepted in fairly short order. (I had some other suggestions in the review, but they aren't relevant to accepting the PEP).

Regarding the concerns about a potential performance impact for unaligned memory access, I think that can be better assessed *after* the simpler patch is merged and it's easier for people to get hold of the new hash implementation. However, the PEP should mention the concern, and note it as something we will be keeping a close eye on during the beta period.