python / mypy Public

…s" (but only for stubs) (#11915)

As discussed in #11895, mypy's test suite currently does not pass a mypy self-check when run on Windows. This should hopefully be fixed by python/typeshed#6812, but running a self-check on Windows as part of the CI tests should help avoid this issue in the future.

* Forbid extra `ParamSpec` arguments

* Change error message

Source commit:
python/typeshed@9116b29

* Fix self check

* update testDictWithStarStarSpecialCase

* update testTypedDictMappingMethods

* xfail testAsyncioGatherPreciseType

Co-authored-by: hauntsaninja <>

Closes #12062
Addresses feedback in #12066

Brings INI and TOML examples in line as noted.

Co-authored-by: Mehdi Drissi <mdrissi@snapchat.com>

If we change the error code of a message, it can result in existing
"type: ignore" comments being ignored. If this seems to be the case,
add a note that hints about changing the ignored error code.

Give a more specific message for a set of special cased error codes
where we know the original error code. In other cases give a more
vague message.

Make the join of a union type and Any commutative.
Previously the result depended on the order of the operands,
which was clearly incorrect.

Fix #12051, but other use cases are affected as well.

This change was split off from #12054.

When we erase last known values in an union with multiple
Instance types, make sure that the resulting union doesn't have
duplicate erased types. The duplicate items weren't incorrect as such,
but they could cause overly complex error messages and potentially
slow type checking performance.

This is one of the fixes extracted from #12054. Since some of the
changes may cause regressions, it's better to split the PR.

Work on #12051.

Co-authored-by: Nikita Sobolev <mail@sobolevn.me>

As far as I can tell, the nugget exposed by this commit lives only in PEP 544. This
copies that nugget closer to where it is likely to be spotted by the intended audience.

PEPs may not be accessible to customers who reasonably expect relevant information to be
surfaced in featured documentation. Even if customers are aware of PEPs, they may not
think to look there, and don't likely consider them primary sources. It is reasonable to
assume that is the job of the docs, with PEPs capturing more esoteric nuances,
rationales and other then-relevant details of decision-making, etc. It's also reasonable
to expect that where further study may be helpful, links from relevant sections of
primary sources to secondary materials like PEPs should exist. This commit fills in both
gaps for the subject shorthand.

Fixes #10189

Produce an error when encountering a yield expression (both `yield` and `yield from` clauses) in comprehensions and generator expressions. The latter is a syntax error in python 3.8+; see #10189 and also [python issue 10544](https://bugs.python.org/issue10544).

This should fix wheel builds. Our CI always targets Python 3.6
when doing a self check, which means that we won't catch issues
specific to more recent Python versions.

Fixes #11850

There are several discussions and comments describing the following problem 
(references can be found at the end of the PR summary):

```python
def func() -> str | NoReturn:
    ...

func().lower()
```

At the moment the code results in: `"NoReturn" of "Union[str, NoReturn]" has no 
attribute "lower"`

Make `Union[int, NoReturn]` equivalent to `int` in a return type, because in case 
the function returns it must be `int`.

Dataclasses that inherit from Any (actually another type that we don't
know) should assume an (almost) arbitrary constructor.

Fixes #11921

* Clarify stub-only packages need to be installed
* Add link to PEP 561

Allows PlaceholderNode objects before raising an error for the attempted
reuse of a member name inside an Enum.