No error on new union syntax with union of fake types (TypeAlias and None)

[KotlinIsland]

from typing import TypeAlias

AMONGUS: TypeAlias = "int"
IMPOSTER: TypeAlias = "str"

a: AMONGUS | IMPOSTER = 0  # SUS ALERT
b: AMONGUS | None = None  # SUS ALERT

Any types in a new union syntax that are fake(looking at you None) will be runtime errors.

[sobolevn]

Interesting case 🤔

Forward references are allowed in https://www.python.org/dev/peps/pep-0613/#forward-references

But, I think that AMONGUS | IMPOSTER should be an error in mypy in this case.
We have to track what value was actually used for creating a type alias: real or forward ref.
And when using new-style unions we can check that | operator is supported for these types.

[sobolevn]

This is quite hard to fix. Because AssignmentStmt does not have something like .annotation_expr.

Fields that are present at the moment:

mypy/mypy/nodes.py

Lines 1104 to 1121 in b47245a

	lvalues: List[Lvalue]
	# This is a TempNode if and only if no rvalue (x: t).
	rvalue: Expression
	# Declared type in a comment, may be None.
	type: Optional["mypy.types.Type"]
	# Original, not semantically analyzed type in annotation (used for reprocessing)
	unanalyzed_type: Optional["mypy.types.Type"]
	# This indicates usage of PEP 526 type annotation syntax in assignment.
	new_syntax: bool
	# Does this assignment define a type alias?
	is_alias_def: bool
	# Is this a final definition?
	# Final attributes can't be re-assigned once set, and can't be overridden
	# in a subclass. This flag is not set if an attempted declaration was found to
	# be invalid during semantic analysis. It is still set to `True` if
	# a final declaration overrides another final declaration (this is checked
	# during type checking when MROs are known).
	is_final_def: bool

So, my plan is:

1. Add annotation_expr field to AssignmentStmt, assign it during fastparse.py
2. Analyze it by slightly different rules during semanal.py
3. Type check this new expression in checker.py stage
4. Probably I would need to also modify things like ast merger / differ, fixup, cache, etc

I am not going to touch anything related to .type or .unanalyzed_type, so it should be fine.

[sobolevn]

Moreover, we need to change how semanal binds types somehow for this case.
Because we need to make sure that semanal understands that AMONGUS: TypeAlias = "int" is str, not TypeAlias[int] when analyzed as an annotation.

[sobolevn]

@KotlinIsland take a look at #11611

[KotlinIsland]

@sobolevn Looks good! but does that handle None (None and strs are the only two fake types that I know of).
None | None currently is okay with mypy.