homepage

Consistent with [PEP 3131](https://www.python.org/dev/peps/pep-3131/) and NFKC normalization of identifiers, these two last lines yield an error, since `τ` (U+1D70F) is automatically converted to `τ` (U+03C4).
 class Base(object):
 def __init__(self):
 self.τ = 5 # defined with U+1D70F
 a = Base()
 print(a.τ) # 5 # (U+1D70F) expected and intuitive
 print(a.τ) # 5 as well # (U+03C4) normalized version, okay.
 d = a.__dict__ # {'τ': 5} # (U+03C4) still normalized version
 print(d['τ']) # 5 # (U+03C4) consistent with normalization
 assert hasattr(a, 'τ') # (U+03C4) consistent with normalization
 # But if I want to retrieve it the way I entered it because I can type (U+1D70F)
 print(d['τ']) # KeyError: 'τ' # (U+1D70F) counterintuitive
 assert hasattr(a, 'τ') # Fails # (U+1D70F) counterintuitive
I've described and undestood the problem in [this post](https://stackoverflow.com/questions/48063082/).
Nothing is unconsistent here. However, I am worried that:
- this behaviour might be counterintuitive and misleading, especially if it occurs that the character user can easily enter for some reason (e.g. U+1D70F) is not equivalent to its NFKC normalization (e.g. U+03C4)
- this behaviours makes it more difficult to enjoy python's reflective `__dict__`, `hasattr` and `getattr` features in this particular case.
Maybe it is user's responsibility to be aware of this limitation, and to keep considering utf-8 coding a bad practice. In this case, maybe this particular reflective limitation could be made explicit in PEP 3131.
Or maybe it is python's responsibility to ensure intuitive and consistent behaviour even in tricky-unicode-cases. So reflective features like `__dict__.__getitem__`, `hasattr` or `getattr` would NFKC-convert their arguments before searching just like `a.τ` does, so that:
 getattr(a, 'τ') is gettatr(a, 'τ')
always yields True.
I actually have no idea of the philosophy to stick to. And the only purpose of this post is to inform the community about this particular, low-priority case.
Thank you for supporting Python anyway, cheers for your patience.. and happy 2018 to everyone :)
--
Iago-lito

I just found out about [this](https://bugs.python.org/issue13793) very close issue. Much of the philosophy has been made very clear there.
Since the solution to issue13793 is to *document* much this NFKC normalization. Then I think I'd be a good thing to make an explicit statement about these particular reflective limitations in PEP 3131 :)

We don't generally update finalized PEPs. The official documentation for a feature is in the Python docs. Feel free to propose a PR if you think it could be improved.