zimports

Reformats Python imports so that they can pass flake8-import-order. This is roughly:

• one import per line
• alphabetically sorted, with stylistic options for how dots, case sensitivity, and dotted names are sorted
• grouped by builtin / external library / current application (also stylistically controllable)
• unused imports removed, using pyflakes to match "unused import" warnings to actual lines of code
• duplicate imports removed (note this does not yet include duplicate symbol names against different imports)
• no star imports (e.g. from <foo> import *); these are rewritten as explicit names, by importing all the names from each target module and then removing all the unused names
• support for TYPE_CHECKING import blocks.

The program currently bolts itself on top of flake8-import-order, in order to reuse the import classification and sorting styles that tool provides. Without options given, the script will look directly for a setup.cfg file with a [flake8] section and will consume flake8-import-order parameters "application-import- names", "application-package-names", and "import-order-style", to sort imports exactly as this linter then expects to find them. All of the single-line import styles, e.g. google, cryptography, pycharm, should just work.

Special classifications can be given to imports, as either a " # noqa" comment indicating the import should not be removed, and optionally the comment " # noqa nosort" which will place the import into a special "don't sort" category, placing all of the "nosort" imports in the order they originally appeared, grouped after all the sorted imports. This can be used for special situations where a few imports have to be in a certain order against each other (SQLAlchemy has two lines like this at the moment).

The application also does not affect imports that are inside of conditionals or defs, or otherwise indented in any way, with the exception of TYPE_CHECKING imports. This is also the behavior of flake8-import-order; only imports in column zero of the source file are counted, although imports that are on lines below other definitions are counted, which are moved up to the top section of the source file.

Note

This application runs in Python 3 only. It can reformat imports for Python 2 code as well but internally it uses library and language features only available in Python 3.

zzzeek why are you writing one of these, there are a dozen pep8 import fixers

I've just gone through a whole bunch. I need one that:

• works directly with flake8-import-order so we are guaranteed to have a match
• has shell capability, not only a plugin for vim or sublime text (Python Fix Imports, gratis)
• Removes unused imports, not just reformats them (importanize)
• Reformats imports, not just removes unused ones (autoflake)
• Doesn't miss removing an import that isn't used just because it's on a multiline import (autoflake)
• Breaks up all imports into individual lines, not just if the line is >80 char (importanize)
• Is still pretty simple (we're a bit beyond our original "extremely" simple baseline, because all problems are ultimately not that simple) because (since pyflakes and now flake8-import-order do most of the hard work) this is an extremely simple job, there's (still) no need for a giant application here.

But what about... isort ??

Since I developed zimports some years ago and now have it on all my projects, isort has come out and is widely becoming accepted as the de-facto import sorter, because it's actually super nice and has tons of features. It popped up turned on by default in my vscode IDE and it's under pycqa, it's clearly the winning tool in this space.

So I would like to use isort, and I've tried it out. I was able to get it 99% equivalent to how we sort our imports now, with the exception of a weird relative import issue that still wouldn't compare against flake8-import-order (it seemed like lexical sorting wasn't working correctly). Maybe we can get that little part working, but that's not the main issue.

The bigger shortcoming was IIUC it, like "importanize" mentioned previously, just reformats the imports that are present. It won't remove unused imports, nor does it have any ability to expand import * into individual imports, since it isn't looking at the rest of the code. zimports actually hangs on top of flake8 so that we can remove unused imports and it also uses flake8 output along with a module import path in order to expand out "*" imports. I use this feature all the time when I type out test scripts for SQLAlchemy, I just start with from sqlalchemy import * and have zimports clean it all up.

Maybe there would be a way to keep zimports for that part, and then use isort for the actual sorting. But then I'm still just using zimports, and while isort definitely does a better job at finding imports to sort (it does them inside method bodies, inside of cython files, wow), it's not really worth it right now for me to change everything when I still have to maintain zimports anyway.

TL;DR; yes go use isort, I have no desire to support zimports for other people! :) zimports does a few things that I personally like a lot, especially removing unused imports which is totally essential for my use cases.

Usage

The script can run without any configuration, options are as follows:

$ zimports --help
usage: zimports [-h] [-m APPLICATION_IMPORT_NAMES]
                [-p APPLICATION_PACKAGE_NAMES] [--style STYLE] [--multi-imports]
                [-k] [-kt] [--heuristic-unused HEURISTIC_UNUSED] [--statsonly]
                [-e] [--diff] [--stdout]
                filename [filename ...]

positional arguments:
  filename              Python filename(s) or directories

optional arguments:
  -h, --help            show this help message and exit
  -m APPLICATION_IMPORT_NAMES, --application-import-names APPLICATION_IMPORT_NAMES
                        comma separated list of names that should be
                        considered local to the application. reads from
                        [flake8] application-import-names by default.
  -p APPLICATION_PACKAGE_NAMES, --application-package-names APPLICATION_PACKAGE_NAMES
                        comma separated list of names that should be
                        considered local to the organization. reads from
                        [flake8] application-package-names by default.
  --style STYLE         import order styling, reads from [flake8] import-
                        order-style by default, or defaults to 'google'
  --multi-imports       If set, multiple imports can exist on one line
  -k, --keep-unused     keep unused imports even though detected as unused.
                        Implies keep-unused-type-checking
  -kt, --keep-unused-type-checking
                        keep unused imports even though detected as unused in
                        type checking blocks. zimports does not detect type usage
                        in comments or when used as string
  --heuristic-unused HEURISTIC_UNUSED
                        Remove unused imports only if number of imports is
                        less than <HEURISTIC_UNUSED> percent of the total
                        lines of code. Ignored in type checking blocks
  --statsonly           don't write or display anything except the file stats
  -e, --expand-stars    Expand star imports into the names in the actual
                        module, which can then have unused names removed.
                        Requires modules can be imported
  --diff                don't modify files, just dump out diffs
  --stdout              dump file output to stdout

Configuration is currently broken up between consumption of flake8 parameters from setup.cfg, and then additional zimports parameters in pyproject.toml (as of version 0.5.0) - unification of these two files will be in a future release, possibly when flake8 adds toml support:

# setup.cfg

[flake8]
enable-extensions = G
ignore =
    A003,
    E203,E305,E711,E712,E721,E722,E741,
    F841,
    N801,N802,N806,
    W503,W504
import-order-style = google
application-import-names = sqlalchemy,test

# pyproject.toml, integrated with black

[tool.black]
line-length = 79
target-version = ['py37']


[tool.zimports]
black-line-length = 79
keep-unused-type-checking = true

# other options:
# multi-imports = true
# keep-unused = true

Then, a typical run on a mostly clean source tree looks like:

$ zimports lib/
[Unchanged]     lib/sqlalchemy/inspection.py (in 0.0058 sec)
[Unchanged]     lib/sqlalchemy/log.py (in 0.0221 sec)

...

[Unchanged]     lib/sqlalchemy/orm/attributes.py (in 0.2152 sec)
[Unchanged]     lib/sqlalchemy/orm/base.py (in 0.0363 sec)
[Writing]       lib/sqlalchemy/orm/relationships.py ([2% of lines are imports] [source +0L/-2L] [3 imports removed in 0.3287 sec])
[Unchanged]     lib/sqlalchemy/orm/strategies.py (in 0.2237 sec)

The program has two general modes of usage. One is that of day-to-day usage for an application that already has clean imports. Running zimports on the source files of such an application should produce no changes, except for whatever source files were recently edited, and may have some changes to imports that need to be placed into the correct order. This usage model is similar to that of Black, where you can run "zimports ." and it will find whatever files need adjusting and leave the rest alone.

The other mode of usage is that of the up-front cleaning up of an application that has un- organized imports. In this mode of usage, the goal is to get the source files to be cleaned up so that zimports can be run straight without any modifications to the file needed, including that all necessary imports are either used locally or marked as not to be removed.

Problems that can occur during this phase are that some imports are unused and should be removed, while other imports that are apparently unused are still in fact imported by other parts of the program. Another issue is that changing the ordering of imports in complex cases may cause the application to no longer run due to the creation of unresolvable import cycles. Finally, some programs have use of import *, pulling in a large list of names for which an unknown portion of them are needed by the application. The options --keep-unused, --heuristic-unused and --expand-stars are provided to assist in working through these issues until the code can be fully reformatted such that running zimports no longer produces changes.

The issue of apparently unused imports that are externally imported can be prominent in some applications. In order to allow imports that aren't locally used to remain in the source file, symbols that are part of __all__ will not be removed, as will imports that are followed by a `` # noqa`` comment. Either of these techniques should be applied to imports that are used from other modules but not otherwise referenced within the immediate source file. For the less common case that a few imports really need a very specific import order for things to work, those imports can be followed by a `` # noqa nosort`` comment that will add these lines to a special group at the end of all imports, where they will not be removed and their order relative to each other will be maintained.

The program does currently require that you pass it at least one file or directory name as an argument. It also does not have the file caching feature that Black has, which can allow it to only look at files that have changed since the last run. The plan is to have it check that it's inside a git repository where it will run through files to be committed if no filenames are given.

Usage as a git hook

zimports can be used with the pre-commit git hooks framework. To add the plugin, add the following to your .pre-commit-config.yaml. Note the rev: attribute refers to a git tag or revision number of zimports to be used, such as "master" or "0.1.3":

repos:
-   repo: https://github.com/sqlalchemyorg/zimports/
    rev: v0.4.5
    hooks:
    -   id: zimports