After learning about TYPE_CHECKING i made it a habit to put all imports that were only needed for type checking into an if TYPE_CHECKING: guard. But now I am wondering if that is actually intended to be used like that. Checking whether an import is only needed at type checking time can get quite tedious and sometimes you run into situations were you introduced some code that made the import a requirement at runtime.
How do you use TYPE_CHECKING? Whenever it is possible or only when using it actually solves a circular import?
I only use it to avoid circular imports. Otherwise, I can import the type plainly.
Same, I think it’s more common only to use when necessary.
The main case I can think of to use it more is for performance to save an import at runtime, but I don’t think that’s really valid, especially since the fact you’re using the type annotation suggest the module would have been used elsewhere anyway so the import would be cached.
The argument against using anywhere is that it could be misleading as your editor may indicate that the import is definited even if it wouldn’t be at runtime. Not sure if things like pylance have special handling to avoid this, would have to check…
I don’t like having to quote the types, so I use it exclusively for avoiding circular imports.
from __future__ import annotationsThanks for the tip
Why not use
from __future__ import annotations?You still need to import the type before using it in a stringified type annotation for it to be valid though, so you’d need the import in an
if TYPE_CHECKING:block either way, no?Yes, but if it’s in a TYPE_CHECKING block I can ONLY use the annotation with quotes*, which is why I only use that method if I must.
- except with
from __future__ import annotationsas I’ve just learned.
Ah yeah, I see what you meant.
- except with
That’s a very cool feature, had no clue about it!
If it doesn’t have any visible downsides, it’s be nice use it whenever possible. This should provide the additional benefit of having the imports clearly separated.
The tediousness aspect of it makes me wonder though. I’d probably just only use it when I’m specifically importing something only for typing .
Maybe could be a cool feature request for an lsp as well.
Any time you need different behavior between static type checking and runtime.
in 3.10 I’m using it to work around issue with NamedTuple generics. typing_extensions.NamedTuple allows Generics at runtime but typing.NamedTuple doesn’t. But the type checker we are using doesn’t support typing_extensions.NamedTuple like it does for the typing version so we lie at type checking time to get the typing to make sense but have different runtime type because otherwise its a TypeError
A cheeky answer: whenever Ruff/flake8-type-checking tells me to. Though I’d only enable that check now that there’s an autofix in Ruff as well.
You should have part of your test harness perform a separate import of every module. If your module is idempotent (most good code is) you could do this in a single process by cleaning
sys.modulesI guess … but it still won’t be part of yourpytestprocess.Static analyzers can only detect some cases, so can’t be fully trusted.
I’ve also found there are a lot of cases where performant Python code has to be implemented in a distinct way from what the type-checker sees. You can do this with aggressive
type: ignorebut I often find it cleaner to use separateifblocks.
