Why not using #typecheck:: comments instead of if typing.TYPE_CHECKING?

[xareelee]

If you need to know what the exact type of a variable is by type inferencing, you need to use typing.TYPE_CHECKING:

c: list[int] = [1, 2, 3]
if typing.TYPE_CHECKING:
        print('this line will not be printed')    # this line will not be executed
        reveal_type(c)   # This line only works for mypy

Why not using comment style, just like a hint for linter?

c: list[int] = [1, 2, 3]
# typecheck::reveal_type(c)    # mypy reads any comment starting with `typecheck::` and execute it when analyzing.

To turn off mypy function call, just make the comment from # typecheck:: to #! typecheck::

c: list[int] = [1, 2, 3]
#! typecheck::reveal_type(c)    # mypy won't read this line

Why do we need an if typing.TYPE_CHECKING statement for mypy?

[erictraut]

The reveal_type call was recently added to the runtime. You can import it from typing_extensions, and it will also be available in typing as of Python 3.11. Note that reveal_type is not specific to mypy; it works with all popular Python type checkers.

If you use a type checker in your code editor rather than running it as a command-line tool, you can get "live" feedback about a symbol's type simply by hovering over it with your mouse pointer. Mypy wasn't designed for use a language server, but pyright (the type checker I work on) is designed for this use. This eliminates the need for using reveal_type.

I'll also note that I generally recommend against using typing.TYPE_CHECKING in your code because it "lies" to the type checker and potentially obscures runtime errors that would otherwise be caught.

[xareelee]

The reveal_type call was recently added to the runtime. You can import it from typing_extensions, and it will also be available in typing as of Python 3.11. Note that reveal_type is not specific to mypy; it works with all popular Python type checkers.

May I ask what reveal_type will do at runtime? Do I need to remove them in production code?

I remember mypy and many other linters/checkers support comment syntax annotation. I think this feature is not hard to implement.

Using comment annotation ensures the code will not be execute at runtime, and you don't need to use import statement to use that. This is coding friendly for other developers.

[JelleZijlstra]

if TYPE_CHECKING has other usages than reveal_type(). For reveal_type() specifically, you can instead use from typing_extensions import reveal_type (or from typing import reveal_type on 3.11) to avoid the condition.

I wasn't around when if TYPE_CHECKING was designed, but it has a couple of advantages over a magic comment: it uses normal Python syntax, so it's easier to understand for other tools, and it naturally supports more than one line of code.

[Akuli]

if TYPE_CHECKING is also super easy to implement. Type checkers have to understand things like if foo is not None or if sys.platform == "win32" anyway, and if TYPE_CHECKING is just like those, but even simpler.

[xareelee]

Currently, I write reveal_type using one line code as the following:

(reveal_type(c) if typing.TYPE_CHECKING else None)

Does this statement have any issue in production?

[erictraut]

reveal_type simply returns the value it was passed. This allows it to be used within other expressions and without the use of typing.TYPE_CHECKING.