Typeshed PR here: python/typeshed#6984

Python 3.7 and 3.8 fail the CI because they don't have ast.Constant. Actually they have it, but Python just doesn't use it in the AST for whatever reason.

I added print(ast.dump(literal)) just under for literal in literals_in_union:, and here's how the AST looks:

akuli@akuli-desktop:~/flake8-pyi$ cat a.pyi
from typing import Union
from typing_extensions import Literal

x: Union[str, Literal['foo'], Literal['bar']]
akuli@akuli-desktop:~/flake8-pyi$ env37/bin/flake8 a.pyi
Index(value=Str(s='foo'))
Index(value=Str(s='bar'))
a.pyi:4:10: Y030 Multiple Literal members in a union. Combine them into one, e.g. "Literal[]".

akuli@akuli-desktop:~/flake8-pyi$ env310/bin/flake8 a.pyi
Constant(value='foo')
Constant(value='bar')
a.pyi:4:10: Y030 Multiple Literal members in a union. Combine them into one, e.g. "Literal['foo', 'bar']".

We should be prepared to handle Literals containing "literal ints, byte and unicode strings, bools, Enum values and None", according to PEP 586.

We should be prepared to handle Literals containing "literal ints, byte and unicode strings, bools, Enum values and None", according to PEP 586.

Ugh bytes and enums. Great.

Thinking about it a bit more, I think the logic can be quite simple: if it's a Tuple, append the elts, and if it's anything else, append the object as is, regardless of whether it's ast.Constant or ast.Str.

Thinking about it a bit more, I think the logic can be quite simple: if it's a Tuple, append the elts, and if it's anything else, append the object as is, regardless of whether it's ast.Constant or ast.Str.

Yeah, mypy can reliably complain if a user puts in an incompatible value, we just need to handle the formatting

I added one more test case, because I wasn't sure whether it would work in different Python versions. It seems to be fine though.

I added one more test case, because I wasn't sure whether it would work in different Python versions. It seems to be fine though.

Thanks! I was having my supper :)

Shall I hold off on merging until #113 is in?

I think I'll merge this first and fix the conflicts in my PR then.

Thanks for the quick implementation here @AlexWaygood!

Thanks for the quick implementation here @AlexWaygood!

Pleasure 😀