False positive for unused import when bringing things forward in __init__.py

[robvdl]

trafficstars

In myapp/forms/__init__.py:

from .login import LoginForm

In myapp/views/auth.py:

from myapp.forms import LoginForm`

ruff says that LoginForm in forms/__init__.py is unused, it's totally wrong because views/auth.py imports and uses it

[robvdl]

Example code here I was trying to add ruff to which triggered this bug https://github.com/wainuiomata/sambal

[trag1c]

It's correct that it's unused in myapp/forms/__init__.py. The fix would be to define a __all__ list/tuple in there too (as Ruff treats that as usage):

from .login import LoginForm

__all__ = ("LoginForm",)

[robvdl]

I'm not sure I agree with this, other linters will notice it unusued.

You appear to be defending a bug, I'm OK with doing the __all__ thing but don't defend a bug, ruff should also should realise it's actually used if it really is.

[trag1c]

I simply don't think it's a bug, I always define an __all__ in my __init__.py files and it's helpful to know I missed an object somewhere.

[robvdl]

It is a bug, it's ruff not able to track this import in views/auth.py properly.

Both pyflakes and PyCharms's linter can do this, so please don't defend this as NOT being a bug.

It's not that I mind doing the __all__ thing but it is still a bug.

[trag1c]

That's your opinion :thumbsup:

[tjkuson]

For what it's worth, I think mypy and Pyright on their respective strict modes will also complain about not explicitly exporting via the __all__ attribute.

[robvdl]

I've never heard of Pyright, but so far I have found 3 linters that can track an import being forwarded in __init__.py.

• pyflakes
• pylint
• PyCharm

So it doesn't look good if both pyflakes and pylint can track this import, but not ruf. You are trying to get users right, because this "I'm right, you're wrong" behaviour is annoying.

[tjkuson]

To be clear, @trag1c and I are not members of Astral, we're just fellow users trying to contribute! Sorry to hear you find my opinion annoying, though.

[robvdl]

Sorry I'm trying to legitimately report a bug here and I felt I was getting fight back, I didn't realise you weren't official devs.

[robvdl]

In the mean time I have put the __all__ thing in it, I just want Astral to look at this rather than sweep it under the carpet.

[charliermarsh]

Hey @robvdl thanks for the report. Unfortunately we don't track imports across files, so the assumption is that you either disable F401 in __init__.py files, or use some other mechanism to make it explicit that this is a re-export -- for example, adding to __all__, or using the re-export convention of from .login import LoginForm as LoginForm. (I know you already heard some of this above, but perhaps the re-export convention will be interesting to you.)

Ideally we would be able to track these across files, so it's fine to call it a bug (and we're planning to explore multi-file analysis over the next few months). The only caveat is that... even then it's not totally clear that it'd be safe to remove from .login import Login, since you're defining a public interface, and you could have users of your package that rely on Login being exported there.

As an aside, PyFlakes would also flag this. I'm not sure how you tested it, but PyFlakes doesn't look across files either:

❯ flake8 foo/
foo/__init__.py:1:1: F401 '.login.LoginForm' imported but unused

Regardless, thanks for the report! That we don't detect these across files is a limitation.

[robvdl]

Thank you, yeah in the mean time I have gone with the __all__ method which I am a little more comfortable with than doing something like this:

# 4. Ignore `E402` (import violations) in all `__init__.py` files.
[tool.ruff.lint.per-file-ignores]
"__init__.py" = ["E402"]

I'll leave it at that.

[charliermarsh]

We've also considered not enforcing unused imports in __init__.py files by default.

[umarbutler]

@charliermarsh Any movement on this? I agree with @robvdl that this is a bug. Having to do __all__ goes against DRY, is unintuitive, and is cumbersome.

@robvdl's fix unfortunately doesn't work for me, however, this fix does (in case someone else wants it):

[tool.ruff.lint.per-file-ignores]
"__init__.py" = ["F401"]

[MichaReiser]

Ruff doesn't support multifile analysis yet but we're working towards bringing this capability to Ruff. But it will take a while and we then also have to change this specific rule implementation. TLDR: We're moving toward a fix but it is a rather large undertaking.

[umarbutler]

Ruff doesn't support multifile analysis yet but we're working towards bringing this capability to Ruff. But it will take a while and we then also have to change this specific rule implementation. TLDR: We're moving toward a fix but it is a rather large undertaking.

Couldn’t this just be made the default?

[tool.ruff.lint.per-file-ignores]
"__init__.py" = ["F401"]

If you’re writing a Python package, you may not have any files in the package that reference imports from init so multifile analysis may not help there anyways right?

[MichaReiser]

yeah, we could consider adding an option or splitting the __init__.py behavior into its own rule.

[Avasam]

I'm surprised to not see anyone mention of

from .login import LoginForm as LoginForm

Which is a canonical way to mark intended re-exports that type-checkers understand, and especially popular in stubs Edit: I missed Charlie did mention it in https://github.com/astral-sh/ruff/issues/10253#issuecomment-1981682998

That way you don't have to explicitly list and maintain all names in __all__

Of course if you use useless-import-alias (PLC0414) now that will trigger, so you have to do:

[tool.ruff.lint.per-file-ignores]
"__init__.py" = ["PLC0414"]

So yeah, feel free to choose how you want to handle your package re-exports, and ignore the relevant code for __init__.py files.

---

We've also considered not enforcing unused imports in __init__.py files by default.

As long as it's an option, I'd still prefer being explicit about my intended re-exports.

[robvdl]

I've never seen this method "from .login import LoginForm as LoginForm" but I can already say I hate it. It's like renaming an import to the same name it already is, it looks awkward and some other developer is going to come along and see it and go "what the heck is this" and remove them all, so you'd then have to place a comment to say why your'e doing that.

I'd rather use the __all__ method, but I stand firm by my statement that we shouldn't have to bring imports forward in order for them to be included for linting.

[umarbutler]

Personally, now that I have my workaround have started adding to all of my project's ruff configs, I'm not as annoyed by this, however, for a new user, there is a real potential for this to be confusing when you literally have 60+ error show up for init.py imports (yes, I had quite a lot of init.pys in my project) but your Pylance has never complained about that as an error.