Consider not wrapping a single import when the line is too long

[yilei]

Describe the style change

When an import line is too long, and it only has a single imported symbol, it's more readable to NOT wrap the line.

Examples in the current Black style

from company.organization.products.my_awesome_learning_system.backend.handlers import (
    register,
)
from company.organization.products.my_awesome_learning_system.common import logging
from company.organization.products.my_awesome_learning_system.utils import (
    authentication,
)

Desired style

from company.organization.products.my_awesome_learning_system.backend.handlers import register
from company.organization.products.my_awesome_learning_system.common import logging
from company.organization.products.my_awesome_learning_system.utils import authentication

Additional context

The current behavior works very well in the majority cases. If the import line has more than one imports, it should continue wrapping them:

from company.organization.products.my_awesome_learning_system.backend.handlers import (
    enroll,
    login,
    register,
)

But for code bases that follow the Google Python Style Guide to only import modules/packages, never individual functions/classes from a module (it won't import multiple symbols in the same line, except from typing), Black can make the imports very awkward.

[Anupam-USP]

Hey, I understood the issue, but don't know how to proceed on this right now. Can you please help me to initialize this??

[yilei]

@Anupam-USP This is a style change proposal, so the first step is to have buy-in from Black maintainers that this is a good change.

[felix-hilden]

Thanks for the suggestion 👍 I'm a bit ambivalent:

Your suggestion is good for the reduced number of extra lines. It is also much better when the imports share module paths. But if the line is really long, it could be beneficial to see the final imported name on its own line so you don't have to horizontal scroll.

To have an adversarial example:

from really.long_package_name.with_submodules.that_should_probably.be_refactored_to.a_common_top_level_api.before_publishing_anything.to_pypi import this_function
from a_different_package.that_does_not_share.submodules_with_the_previous.so_you_still_need_to.read_the_path import another_function
from another_goddamn_package.that_does_not_know.how_to_manage_comes.this_module_that_is.slightly_longer_than.previous_so_your_eyes.dont_know_where.to_look import ThisClass

# VS

from really.long_package_name.with_submodules.that_should_probably.be_refactored_to.a_common_top_level_api.before_publishing_anything.to_pypi import (
    this_function
)
from a_different_package.that_does_not_share.submodules_with_the_previous.so_you_still_need_to.read_the_path import (
    another_function
)
from another_goddamn_package.that_does_not_know.how_to_manage_comes.this_module_that_is.slightly_longer_than.previous_so_your_eyes.dont_know_where.to_look import (
    ThisClass
)

Just playing the devil's advocate here. This is an unlikely scenario and I don't really feel strongly 😄

[yilei]

Yeah this is a relatively unlikely scenario if you look at the overall picture. When you import almost all packages on PyPI, you don't end up such long import paths.

But when a codebase is impacted, it can be pretty bad. The long import paths usually come from internal code e.g. in a mono repo. For us, such long import paths are almost everywhere.

One thing do note that since imports are sorted, it's much, much more common for adjacent imports to share module paths, than your example. Hence I think we should prioritize the relatively more common scenario and have them not wrapped.

I'll send a PR once I find myself some time.

[nameer]

I find the current one more readable and less likely to break the line length.

[sanmai-NL]

Too long lines trigger linters, which is a significant downside of keeping them as is.

[cobaltt7]

IMO, you shouldn't need to jump through hoops and formst your code a specific way in order to satisfy a linter. Maybe the real lint error in this case is the imported package and not your code.

An alternative solution I haven't seen mentioned here is to just import a higher level namespace and access the individual thing you want in a variable declaration later on, i.e.:

from company.organization.products import my_awesome_learning_system as system
register = system.backend.handlers.register
logging = system.common.logging
authentication = system.utils.authentication