I have been bitten by that as well. I think the doc should mention to verify that the given string is normalized, not that it **should** be normalized.

Agreed that If isidentifier could also possibly grow a `allow_non_nfkc=True` default parameter that would allow to deactivate internal normalisation and return False/Raise on Non NKFC that would be great. 

I'm also interested on having an option on ast.parse or compile to not normalize to at least be able to lint wether users are using non NFKC form, but that's another issue.

I'll see if I can come up with – at least – a documentation patch.