Conversation
|
Not sure why that test is failing only on 3.9; it's a bit weird. I'll try to take another look tomorrow, but any suggestions are welcome :) |
I figured it out -- |
|
Sorry for being a pain, but could we split out the flake8 changes for a cleaner history? |
|
srittau
left a comment
srittau
left a comment
There was a problem hiding this comment.
My understanding of typing internals is limited, but this LGTM, apart from the comment about flake8 and a question below.
| return NotImplemented | ||
| if not getattr(cls, '_is_runtime_protocol', False): | ||
| if _caller(depth=3) in {'abc', 'functools'}: | ||
| if _allow_reckless_class_checks(): |
There was a problem hiding this comment.
Is it correct that the depth is changed from 3 to 4?
There was a problem hiding this comment.
Yes, because by adding a new function that the code has to pass through before it gets to the sys._getframe call, the call stack becomes "another frame deep"
There was a problem hiding this comment.
I could just have looked at _caller and figured that out myself. 🤦
| " @runtime protocols") | ||
| if not _is_callable_members_only(cls): | ||
| if _caller(depth=3) in {'abc', 'functools'}: | ||
| if _allow_reckless_class_checks(): |
|
Thanks! |
2 participants
This backports the CPython PR python/cpython#26067, which was backported to Python 3.10 but no earlier in the CPython repo.
(It was initially backported to 3.9, but the backport was reverted: python/cpython#26077.)
Since we're now backporting a bugfix to <=3.9, this means that we re-implement
Protocol/runtime_checkableon <=3.9, rather than on just 3.7. Doing this requires a change to_get_protocol_attrs; I also tweakedtyping_extensions.runtime_checkableso that it also works ontyping.Protocol, iftyping.Protocolexists.