Empty typing.Tuple

[serhiy-storchaka]

BPO	46981
Nosy	@gvanrossum, @serhiy-storchaka, @JelleZijlstra, @Fidget-Spinner
PRs	bpo-46981: Remove typing._TypingEmpty #31836

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2022-03-17.08:20:59.454>
created_at = <Date 2022-03-11.08:21:12.398>
labels = ['library', '3.11']
title = 'Empty typing.Tuple'
updated_at = <Date 2022-04-08.05:07:15.239>
user = 'https://github.com/serhiy-storchaka'

bugs.python.org fields:

activity = <Date 2022-04-08.05:07:15.239>
actor = 'JelleZijlstra'
assignee = 'none'
closed = True
closed_date = <Date 2022-03-17.08:20:59.454>
closer = 'serhiy.storchaka'
components = ['Library (Lib)']
creation = <Date 2022-03-11.08:21:12.398>
creator = 'serhiy.storchaka'
dependencies = []
files = []
hgrepos = []
issue_num = 46981
keywords = ['patch']
message_count = 5.0
messages = ['414892', '414947', '414986', '415391', '416958']
nosy_count = 4.0
nosy_names = ['gvanrossum', 'serhiy.storchaka', 'JelleZijlstra', 'kj']
pr_nums = ['31836']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue46981'
versions = ['Python 3.11']

[serhiy-storchaka]

There are two empty typing.Tuple. They have the same repr but are not equal.

>>> from typing import *
>>> t1 = Tuple[()]
>>> t2 = t1.copy_with(())
>>> t1
typing.Tuple[()]
>>> t2
typing.Tuple[()]
>>> t1 == t2
False
>>> t1.__args__
((),)
>>> t2.__args__
()

The only differences is that one has empty __args__, while other has __args__ containing an empty tuple. There is a code purposed to make __args__ containing an empty tuple in this case. What is the purpose?

It is not pure theoretical question. This affects unpacked TypeVarTuple substitution. With natural implementation Tuple[Unpack[Ts]][()] is not equal to Tuple[()] and I still have not figured which and where code should be added to handle this special case. It would be easier if such special case did not exist.

Built-in tuple does not have a special case:

>>> tuple[()].__args__
()

[gvanrossum]

Alas, I have no idea. I don't even recall what copy_with() is for (it was apparently introduced in 3.7). Possibly vopy_with() is wrong here? I imaging some of this has to do with the special casing needed so that repr() of an empty Tuple type doesn't print "Tuple[]" (which IIRC it did, long ago).

[serhiy-storchaka]

If for repr(Tuple[()]), it is no longer needed.

[serhiy-storchaka]

New changeset 15df8f8 by Serhiy Storchaka in branch 'main':
bpo-46981: Remove typing._TypingEmpty (GH-31836)
15df8f8

[JelleZijlstra]

I tried out 3.11 on my pyanalyze type checker and got some failures because of this change, because my previous trick for distinguishing between Tuple and Tuple[()] failed.

3.10:

>>> from typing import get_args, Tuple
>>> get_args(Tuple[()])
((),)
>>> get_args(Tuple)
()

3.11:

>>> from typing import get_args, Tuple
>>> get_args(Tuple[()])
()
>>> get_args(Tuple)
()

However, the new behavior is more consistent: get_args(tuple[()]) always returned (). It's also easy enough to work around (just check ... is Tuple).

I'll put a note in the What's New for 3.11 about this change.

[serhiy-storchaka]

We get a report about broken pickling of empty typing.Tuple #94245. Shallow and deep copying are broken too.

>>> copy.copy(Tuple[()])
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/serhiy/py/cpython3.10/Lib/copy.py", line 102, in copy
    return _reconstruct(x, None, *rv)
  File "/home/serhiy/py/cpython3.10/Lib/copy.py", line 265, in _reconstruct
    y = func(*args)
  File "/home/serhiy/py/cpython3.10/Lib/typing.py", line 312, in inner
    return func(*args, **kwds)
  File "/home/serhiy/py/cpython3.10/Lib/typing.py", line 1233, in __getitem__
    params = tuple(_type_check(p, msg) for p in params)
  File "/home/serhiy/py/cpython3.10/Lib/typing.py", line 1233, in <genexpr>
    params = tuple(_type_check(p, msg) for p in params)
  File "/home/serhiy/py/cpython3.10/Lib/typing.py", line 176, in _type_check
    raise TypeError(f"{msg} Got {arg!r:.100}.")
TypeError: Tuple[t0, t1, ...]: each t must be a type. Got ().
>>> copy.deepcopy(Tuple[()])
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/serhiy/py/cpython3.10/Lib/copy.py", line 172, in deepcopy
    y = _reconstruct(x, memo, *rv)
  File "/home/serhiy/py/cpython3.10/Lib/copy.py", line 265, in _reconstruct
    y = func(*args)
  File "/home/serhiy/py/cpython3.10/Lib/typing.py", line 312, in inner
    return func(*args, **kwds)
  File "/home/serhiy/py/cpython3.10/Lib/typing.py", line 1233, in __getitem__
    params = tuple(_type_check(p, msg) for p in params)
  File "/home/serhiy/py/cpython3.10/Lib/typing.py", line 1233, in <genexpr>
    params = tuple(_type_check(p, msg) for p in params)
  File "/home/serhiy/py/cpython3.10/Lib/typing.py", line 176, in _type_check
    raise TypeError(f"{msg} Got {arg!r:.100}.")
TypeError: Tuple[t0, t1, ...]: each t must be a type. Got ().

All this can be fixed by backporting this change. But it will break pyanalyze.

[amitbad]

I was using python 3.10 and getting the same error while using Gemini-Pro-Vision.

This resolved my issue - pip install pydantic==1.10.13