Fix isinstance with unions of tuples

[hauntsaninja]

No description provided.

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

hydpy (https://github.com/hydpy-dev/hydpy)
+ hydpy/core/objecttools.py:1743: error: Unused "type: ignore" comment  [unused-ignore]

psycopg (https://github.com/psycopg/psycopg)
+ psycopg/psycopg/types/numeric.py:407: error: No overload variant of "int" matches argument type "object"  [call-overload]
+ psycopg/psycopg/types/numeric.py:407: note: Possible overload variants:
+ psycopg/psycopg/types/numeric.py:407: note:     def int(str | Buffer | SupportsInt | SupportsIndex = ..., /) -> int
+ psycopg/psycopg/types/numeric.py:407: note:     def int(str | bytes | bytearray, /, base: SupportsIndex) -> int

Tanjun (https://github.com/FasterSpeeding/Tanjun)
- tanjun/context/slash.py:982: error: List item 0 has incompatible type "_OtherT"; expected "_T"  [list-item]

pandas (https://github.com/pandas-dev/pandas)
+ pandas/core/indexes/datetimelike.py:892: error: Unused "type: ignore" comment  [unused-ignore]

pylint (https://github.com/pycqa/pylint)
+ pylint/checkers/utils.py:1969: error: Unused "type: ignore" comment  [unused-ignore]

cryptography (https://github.com/pyca/cryptography)
+ src/cryptography/x509/extensions.py:1561: error: Incompatible return value type (got "list[str | Name | IPv4Address | IPv6Address | IPv4Network | IPv6Network | bytes | ObjectIdentifier]", expected "list[IPv4Address | IPv6Address | IPv4Network | IPv6Network] | list[str] | list[OtherName] | list[Name] | list[ObjectIdentifier]")  [return-value]
+ src/cryptography/x509/extensions.py:1562: error: Incompatible return value type (got "list[DNSName | DirectoryName | IPAddress | OtherName | RFC822Name | RegisteredID | UniformResourceIdentifier]", expected "list[IPv4Address | IPv6Address | IPv4Network | IPv6Network] | list[str] | list[OtherName] | list[Name] | list[ObjectIdentifier]")  [return-value]

jinja (https://github.com/pallets/jinja)
+ src/jinja2/nodes.py:203: error: Unused "type: ignore" comment  [unused-ignore]

[hauntsaninja]

Most of these are positive.
The psycopg one may potentially be fixable in a future PR
The cryptography one is unfortunate but it is what it is, we would need to somehow magically infer Iterable[X] | Iterable[Y] instead of Iterable[X | Y]

[JukkaL]

I agree that this seems like an improvement overall.

[jelle-openai]

hm this seems wrong? Now if you have t: tuple[type[int]] | tuple[type[str]] or something and you do if isinstance(x, t):, won't we incorrectly narrow t in the else branch?

[hauntsaninja]

Thanks! We won't, but it is possible we might do the incorrect narrowing you suggest for final types in the future. I will add some more test cases to #20675.

The right fix is probably elsewhere, in conditional_types. Over here we are returning a list of types, not a union, and we should distribute over proposed type ranges instead of union-ing here:

mypy/mypy/checker.py

Lines 8380 to 8388 in 8bf049f

	proposed_precise_type = UnionType.make_union(
	[type_range.item for type_range in proposed_type_ranges if not type_range.is_upper_bound]
	)
	remaining_type = restrict_subtype_away(
	current_type,
	proposed_precise_type,
	consider_runtime_isinstance=consider_runtime_isinstance,
	)
	return proposed_type, remaining_type

[hauntsaninja]

Tests added here: https://github.com/python/mypy/pull/20675/files#diff-e3de7a75a8a107b4f462b164cdf4945d50505c5e9f7092b753c4add0c01530bbR3006