test suite characterizing dict.get()

[tungol]

While working on understanding the subtleties of dict.get(), I read this comment from Jelle:

It would also be good to add some more test cases to cover dict.get; it's been a very tricky function to cover in the stubs.

Here's those test cases which reflect the current behavior in several different contexts.

I also had this one, but I'm convinced this is just a mypy bug. It'd probably be a good regression test there, if this version of TypeOfAny.special_form is ever split to a new type of Any as suggested by python/mypy#15497 (comment).

# Bonus weirdness with TypeAlias
MyType: TypeAlias = Dict[str, Any]
d_type: MyType = {}
special_any = d_type["key"]

assert_type(d_any.get("key", any_value), Any)
assert_type(d_any.get("key", special_any), Any)

assert_type(d_str.get("key", any_value), Any)
assert_type(d_str.get("key", special_any), str)  # ?

[tungol]

For additional context: it's not obvious that all of these variants are needed, but some of them change in non-obvious ways with #13222 so I think it's reasonable.

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[tungol]

Discovered a new type of edge case for dict.get that I wasn't thinking about before: #13229 (comment)

Analysis of some mypy primer results linked, but the theme is what happens if the default value is an empty container type. Right now for dict[str, Any], d_any.get("key", []) returns Any but any change that makes it return Ant | list[Any] opens up a huge amount of noise in mypy primer. It's even worse for nested d_any.get("key1, {}).get("key2", {}). It might be worth adding something in this genre to the tests as well.

Note that for the fully concrete dict[str, str], d_str.get("key", []) already returns str | list[Any] any mypy complains about the unknown container type. For dict[str, list[str]], d_lstr.get("key", []) returns list[str] without issue.