/ssort

Tool for automatically sorting python statements within a module

Primary LanguagePythonMIT LicenseMIT

SSort

Build Status Cocerage

The python source code sorter.

Sorts the contents of python modules so that statements are placed after the things they depend on, but leaves grouping to the programmer. Groups class members by type and enforces topological sorting of methods.

Makes old fashioned code navigation easier, you can always scroll up to see where something is defined, and reduces bikeshedding.

Compatible with and intended to complement isort and black.

Before:

from module import BaseClass

def function():
    return _dependency()

def _decorator(fn):
    return fn

@_decorator
def _dependency():
    return Class()

class Class(BaseClass):
    def public_method(self):
        return self

    def __init__(self):
        pass

After:

from module import BaseClass

class Class(BaseClass):
    def __init__(self):
        pass

    def public_method(self):
        return self

def _decorator(fn):
    return fn

@_decorator
def _dependency():
    return Class()

def function():
    return _dependency()

Installation

SSort can be installed manually using pip.

$ pip install ssort

Usage

To check that a file is correctly sorted use the --check flag. --diff can be passed to see what changes ssort would make.

$ ssort --check --diff path/to/python_module.py

To allow ssort to rearrange your file, simply invoke with no extra flags. If ssort needs to make changes to a black conformant file, the result will not necessarily be black conformant. The result of running black on an ssort conformant file will always be ssort conformant. We recommend that you reformat using isort and black immediately after running ssort.

$ ssort src/ tests/; isort src/ tests/; black src/ tests/

You can also setup ssort to run automatically before commit by setting up pre-commit, and registering ssort in your .pre-commit-config.yaml.

repos:
# ...
- repo: https://github.com/bwhmather/ssort
  rev: master
  hooks:
  - id: ssort
- repo: https://github.com/pycqa/isort
  rev: master
  hooks:
  - id: isort
    name: isort (python)
    args: [--profile=black]
- repo: https://github.com/psf/black
  rev: master
  hooks:
  - id: black

Output

ssort will sort top level statements and statements in class bodies.

When sorting top level statements, ssort follows three simple rules:

  • Statements must always be moved after the statements that they depend on, unless there is a cycle.
  • If there is a cycle, the order of statements within the cycle must not be changed.
  • If there is no dependency between statements then, to the greatest extent possible, the original order should be kept.

These rules result in low level building blocks being moved to the top of modules, with higher level logic going at the bottom. The FAQ goes into some detail about why this order was chosen.

The rules for sorting class bodies are more complicated. Class methods are generally only called from outside the class and so there aren't usually many interdependencies from which to derive structure. ssort therefore ignores (deferred) dependencies between d`under and public methods and instead divides up class statements into hard-coded groups that it arranges in the following order:

  • The class docstring.
  • Special attributes, i.e. __slots__ or __doc__.
  • Inner classes.
  • Regular attributes.
  • Lifecycle d'under methods, e.g. __init__ or __new__.
  • Public methods, and unused private methods.
  • Other d'under methods, e.g. __getattr__ or __len__.

Apart from the docstring, this order is essentially arbitrary. It is was chosen as being representative of current standard industry practice.

D'under methods are arranged in a hard coded order within their group. Statements in other groups are left in their original order.

Private methods should only be called from other methods in the class, and so are mixed in topologically.

If a class-definition-time dependency is detected between two statements preserving the relative order of the linked statements will take priority.

Frequently Asked Questions

Why does ssort sort bottom-up rather than top-down?

Python is a scripting language, which means that the body of each module is evaluated, statement by statement, from top to bottom. In almost all cases, things must be defined before they can be used. Attempting, in the subset of cases where it is possible, to reverse the order is difficult to do safely and leads to inconsistency with the cases where top-down ordering is impossible.

Top-down ordering is only possible when lookups are deferred

Top-down ordering is only possible when lookups are deferred, but in most cases, lookups happen immediately.

# Broken.

variable = dependency()

def dependency():
    ...

In this example python will try to find dependency in the locals() dict when the first line is evaluated, and fail because the statement that defines it has not been evaluated yet.

As far as I am aware, there is only one way to reference a variable that has not been bound yet, and that is to close over it in a function definition.

# Working.

def function():
    return dependency()

def dependency():
    ...

This is because the lookup is deferred until after function is called, which in this case doesn't happen until both functions are defined.

Top-down ordering fails unsafe

In cases where lookups are deferred, they may not be deferred sufficiently far to allow the dependant statement to be sorted before its dependencies.

Take the following example formatted in bottom-up order.

# Hidden runtime dependency example sorted bottom-up.

def _shared_dep():
    ...

def _decorator(fn):
    _shared_dep()
    return fn

@_decorator
def top_level():
    _shared_dep()

A naive analysis would suggest that _shared_dep is a runtime dependency and can safely be moved to the bottom of the script.

# Hidden runtime dependency example sorted top-down using naive analysis.

def _decorator(fn):
    _shared_dep()
    return fn

@_decorator
def top_level():
    _shared_dep()

def _shared_dep():
    ...

This will result in a NameError as _shared_dep will not have been bound when _decorator is invoked.

More powerful static analysls can mitigate this problem, but any missed hard references are likely to result in the program being broken. Bottom-up sorting can only force broken reorderings when static analysis misses a reference that results in a cycle.

Top-down ordering needs special cases for constants and imports

Even the most die hard proponent of top down ordering would not argue that import statements should be moved to the bottom of the file.

Take the following example:

from module import first_dep

def second_dep():
    ...

@decorator
def function():
    first_dep()
    second_dep()

A strict top-down sort would see it reordered with the first_dep import at the bottom of the file.

from other_module import decorator

@decorator
def function():
    first_dep()
    second_dep()

def second_dep():
    ...

from module import first_dep

Top-down ordering makes code navigation difficult

With bottom-up ordering, navigation is easy. If you want to find where a variable is defined you scroll up. If you want to find where a variable is used you scroll down. These rules are reliable, and straightforward for programmers to learn and apply.

With top-down order, navigation is more tricky. If you want to find where a variable is defined you scroll down, unless the variable is a constant or an import, or the variable is referenced here at import time, or the variable is referenced somewhere else at import time, or any of the many other special cases. If you want to find where a variable is used, you basically have to scan the whole file.

Every special case added to the sorting tool is a special case that programmers need to learn if they are to navigate quickly, and top-down ordering requires a lot of special cases.

Why doesn't ssort allow me to configure X?

ssort aims to bring about ecosystem wide consistency in how python source files are organised. If this can be achieved then it will help all programmers familiar with its conventions to navigate unfamiliar codebases, and it will reduce arguments between programmers who prefer different conventions. This only works if those conventions can't be changed.

Why was ssort created?

ssort exists because its author was too lazy to implement jump-to-definition in his text editor, and decided that it would be easier to just reformat all of the world's python code to make it possible to navigate by scrolling.

Links

License

The project is made available under the terms of the MIT license. See LICENSE for details.