tipsy

python3.10 introduced user-defined TypeGuards. Finally giving users the ability to actually code type narrowing functions themselves and avoid shortfalls of mypy. However, for most builtin basic types automatic TypeGuards are still missing. Writing these TypeGuards yourself for each of your types from scratch can feel redundant and frustrating. tipsy generates these TypeGuards automatically for you based solely on the type annotations.

Example 1: TypeGuards

Here's an example of how you can use tipsy to generate TypeGuards for a TypedDict. Let's suppose you have a TypedDict like this:

class Person(TypedDict):
    uid: int
    name: str

You might already have observed that you can not use isinstance to check whether a variable is of that type.

isinstance({"uid": 5, "name": "jeeves"}, Person)
# TypeError: TypedDict does not support instance and class checks

So how can you do type narrowing in that case? Well, you need to write a TypeGuard yourself.

def is_person(data: Any) -> TypeGuard[Person]:
    return (isinstance(data, dict)
        and set(data.keys()) == {"uid", "name"}
        and isinstance(data["uid"], int)
        and isinstance(data["name"], str)
    )

is_person({"uid": 5, "name": "jeeves"}) # True
is_person({"name": "jeeves"}) # False
is_person({"uid": 5.0, "name": "jeeves"}) # False
is_person({"uid": 5, "name": b"jeeves"}) # False

As you can see this works perfectly, but these TypeGuards really feel like you are just duplicating the information that is already contained in the type, just to make mypy happy. Well, it's because it is actually duplicated. And even worse, for more complex nested types this gets out of hand quickly. tipsy's is_type uses just the information from the type to automatically generate such TypeGuards for you.

is_type({"uid": 5, "name": "jeeves"}, Person) # True
is_type({"name": "jeeves"}, Person) # False
is_type({"uid": 5.0, "name": "jeeves"}, Person) # False
is_type({"uid": 5, "name": b"jeeves"}, Person) # False

That's it, it's literally the same function, but automatically generated for you by tipsy.

Example 2: casting

Another use-case is casting, maybe you know that only one type is valid in a certain scenario. Let's say you again have the same TypedDict:

class Person(TypedDict):
    uid: int
    name: str

with tipsy, this is a breeze:

jeeves = cast_to_type({"uid": 5, "name": "jeeves"}, Person)