conforming to flake8 appnexus

[JackDra]

appnexus requires relative imports to have current directory above parent directory. E.g.

from .fileA import a_var
from ..fileB import b_var

but isort sorts them in order:

from ..fileB import b_var
from .fileA import a_var

Is there any way to reverse this condition? (i might have missed it in the docs somewhere...).

Currently it is reversed with the current tox.ini setup:

[flake8]
ignore = E123,E133,E226,E241,E262,E265,W503,W504
application-import-names = ...
application-package-names = ...
import-order-style = appnexus
exclude = build

[isort]
force_sort_within_sections = true
order_by_type = false
case_sensitive = false
multi_line_output = 5
sections=FUTURE,STDLIB,THIRDPARTY,FIRSTPARTY,APPLICATION,LOCALFOLDER
known_third_party = ...
known_first_party = ...
known_application = ...
lines_after_imports = 2
no_lines_before=LOCALFOLDER

[timothycrosley]

Hi @JackDra,

This is indeed supported by using the reverse_relative config option, see: https://timothycrosley.github.io/isort/docs/configuration/options/#reverse-relative.

Thanks!

~Timothy

[JackDra]

Ah, I wasn't sure if that was for that.... My bad

It seems to still not be working. I found the issue to be a conflict(?) between force_sort_within_sections and reverse_relative.
If I have both on, it ignores reverse_relative. Is this just a limitation with the logic? or a bug?

[JackDra]

From what I can tell, it seem like here:

https://github.com/timothycrosley/isort/blob/66bc89ff5290d28571b1d8af620644bcba5084f9/isort/output.py#L125-L134

The ordering of the '..' and '.' imports are flipped back to the regular ordering.

adding print statements to the sorting.naturally function called:

def naturally(to_sort: Iterable[str], key: Optional[Callable[[str], Any]] = None) -> List[str]:
    """Returns a naturally sorted list"""
    if key is None:
        key_callback = _natural_keys
    else:

        def key_callback(text: str) -> List[Any]:
            return _natural_keys(key(text))  # type: ignore

    show_sort = False
    for this_sort in to_sort:
        if any(['..' in ikey for ikey in to_sort]):
            show_sort = True
    if show_sort:
        print(to_sort)
        print(sorted(to_sort, key=key_callback))
        print()
    return sorted(to_sort, key=key_callback)

shows in the case where there are both . and .. imports:

OrderedDict([('..fields', OrderedDict([('UnitScalarField', True)])), ('..helpers', OrderedDict([('reserialize_date', True)])), ('..strict_schema', OrderedDict([('StrictSchema', True)])), ('.compound', OrderedDict([('BaseCompoundField', True)])), ('.compound_reagent', OrderedDict([('ICompoundReagentField', True)])), ('.compound_sample', OrderedDict([('CustomPHRField', True)])), ('.custom_fill_factor', OrderedDict([('CustomFillFactorField', True)]))])
['.compound', '.compound_reagent', '.compound_sample', '.custom_fill_factor', '..fields', '..helpers', '..strict_schema']

[timothycrosley]

@JackDra,

Thanks for the additional information and to digging into the cause of the issue. I was able to reproduce, and can confirm that reverse_relative doesn't currently work when combined with force_sort_within_sections. I've fixed this in develop and it will make it's way out in the next release of isort. Additionally, a regression test has been added with your exact error case to ensure it is fixed for good.

Thanks!

~Timothy

[JackDra]

Awsome, thanks!

I have put a request into our platform team to update our package manager (edm) with the latest isort version. I will tell them to hold off until the new release is out.

[timothycrosley]

This change has just been deployed to PyPI in version 5.5.0

Thanks!

~Timothy