Automatically rewrite NamedTuple and TypedDict with newer syntaxes

[charliermarsh]

From pyupgrade:

This could be a good use-case for the codegen module.

[martinlehoux]

I'll try this one if that's okay

[martinlehoux]

A few questions on top of my mind @charliermarsh

• Should we handle from typing import TypedDict as CustomTypedDict?
• Should we handle MyType = TypedDict("MyType", dict(a=str, b=int))?
• What should we do for MyType = TypedDict("OtherName", ...)?
• And overall, do you think TypedDict rule and NamedTuple rule should be the same, or two different rules?

[charliermarsh]

Good questions!

Should we handle from typing import TypedDict as CustomTypedDict?

Na, I think we should ignore (leave it unchanged).

Should we handle MyType = TypedDict("MyType", dict(a=str, b=int))?

I think it's fine not to handle this case. It'd be nice to include it, but I don't see it as a requirement.

What should we do for MyType = TypedDict("OtherName", ...)?

Let's ignore this case, if possible (leave it unchanged).

And overall, do you think TypedDict rule and NamedTuple rule should be the same, or two different rules?

I would vote for two different rules.

[martinlehoux]

Regarding NamedTuples:

• Support rename (that converts invalid identifiers to _<k> keys, and duplicated too) => I don't see a technical reason for why not, but doesn't seem very useful
• Support defaults (that list fields defaults from right to left) => I guess this one is useful
• Support module (I don't event know if possible with class) => I'd say we drop that

[charliermarsh]

Completed by @martinlehoux! Thank you!