The Black code style

Black reformats entire files in place. It is not configurable. It doesn’t take previous formatting into account. It doesn’t reformat blocks that start with # fmt: off and end with # fmt: on. # fmt: on/off have to be on the same level of indentation. It also recognizes YAPF’s block comments to the same effect, as a courtesy for straddling code.

How Black wraps lines

Black ignores previous formatting and applies uniform horizontal and vertical whitespace to your code. The rules for horizontal whitespace can be summarized as: do whatever makes pycodestyle happy. The coding style used by Black can be viewed as a strict subset of PEP 8.

As for vertical whitespace, Black tries to render one full expression or simple statement per line. If this fits the allotted line length, great.

# in:

l = [1,
     2,
     3,
]

# out:

l = [1, 2, 3]

If not, Black will look at the contents of the first outer matching brackets and put that in a separate indented line.

# in:

TracebackException.from_exception(exc, limit, lookup_lines, capture_locals)

# out:

TracebackException.from_exception(
    exc, limit, lookup_lines, capture_locals
)

If that still doesn’t fit the bill, it will decompose the internal expression further using the same rule, indenting matching brackets every time. If the contents of the matching brackets pair are comma-separated (like an argument list, or a dict literal, and so on) then Black will first try to keep them on the same line with the matching brackets. If that doesn’t work, it will put all of them in separate lines.

# in:

def very_important_function(template: str, *variables, file: os.PathLike, debug: bool = False):
    """Applies `variables` to the `template` and writes to `file`."""
    with open(file, 'w') as f:
        ...

# out:

def very_important_function(
    template: str,
    *variables,
    file: os.PathLike,
    debug: bool = False,
):
    """Applies `variables` to the `template` and writes to `file`."""
    with open(file, "w") as f:
        ...

You might have noticed that closing brackets are always dedented and that a trailing comma is always added. Such formatting produces smaller diffs; when you add or remove an element, it’s always just one line. Also, having the closing bracket dedented provides a clear delimiter between two distinct sections of the code that otherwise share the same indentation level (like the arguments list and the docstring in the example above).

If a data structure literal (tuple, list, set, dict) or a line of “from” imports cannot fit in the allotted length, it’s always split into one element per line. This minimizes diffs as well as enables readers of code to find which commit introduced a particular entry. This also makes Black compatible with isort with the following configuration.

A compatible .isort.cfg

[settings]
multi_line_output=3
include_trailing_comma=True
force_grid_wrap=0
use_parentheses=True
line_length=88

The equivalent command line is:

$ isort --multi-line=3 --trailing-comma --force-grid-wrap=0 --use-parentheses --line-width=88 [ file.py ]

Line length

You probably noticed the peculiar default line length. Black defaults to 88 characters per line, which happens to be 10% over 80. This number was found to produce significantly shorter files than sticking with 80 (the most popular), or even 79 (used by the standard library). In general, 90-ish seems like the wise choice.

If you’re paid by the line of code you write, you can pass --line-length with a lower number. Black will try to respect that. However, sometimes it won’t be able to without breaking other rules. In those rare cases, auto-formatted code will exceed your allotted limit.

You can also increase it, but remember that people with sight disabilities find it harder to work with line lengths exceeding 100 characters. It also adversely affects side-by-side diff review on typical screen resolutions. Long lines also make it harder to present code neatly in documentation or talk slides.

If you’re using Flake8, you can bump max-line-length to 88 and forget about it. Alternatively, use Bugbear’s B950 warning instead of E501 and keep the max line length at 80 which you are probably already using. You’d do it like this:

[flake8]
max-line-length = 80
...
select = C,E,F,W,B,B950
ignore = E501

You’ll find Black’s own .flake8 config file is configured like this. If you’re curious about the reasoning behind B950, Bugbear’s documentation explains it. The tl;dr is “it’s like highway speed limits, we won’t bother you if you overdo it by a few km/h”.

Empty lines

Black avoids spurious vertical whitespace. This is in the spirit of PEP 8 which says that in-function vertical whitespace should only be used sparingly.

Black will allow single empty lines inside functions, and single and double empty lines on module level left by the original editors, except when they’re within parenthesized expressions. Since such expressions are always reformatted to fit minimal space, this whitespace is lost.

It will also insert proper spacing before and after function definitions. It’s one line before and after inner functions and two lines before and after module-level functions and classes. Black will not put empty lines between function/class definitions and standalone comments that immediately precede the given function/class.

Black will enforce single empty lines between a class-level docstring and the first following field or method. This conforms to PEP 257.

Black won’t insert empty lines after function docstrings unless that empty line is required due to an inner function starting immediately after.

Trailing commas

Black will add trailing commas to expressions that are split by comma where each element is on its own line. This includes function signatures.

Unnecessary trailing commas are removed if an expression fits in one line. This makes it 1% more likely that your line won’t exceed the allotted line length limit. Moreover, in this scenario, if you added another argument to your call, you’d probably fit it in the same line anyway. That doesn’t make diffs any larger.

One exception to removing trailing commas is tuple expressions with just one element. In this case Black won’t touch the single trailing comma as this would unexpectedly change the underlying data type. Note that this is also the case when commas are used while indexing. This is a tuple in disguise: numpy_array[3, ].

One exception to adding trailing commas is function signatures containing *, *args, or **kwargs. In this case a trailing comma is only safe to use on Python 3.6. Black will detect if your file is already 3.6+ only and use trailing commas in this situation. If you wonder how it knows, it looks for f-strings and existing use of trailing commas in function signatures that have stars in them. In other words, if you’d like a trailing comma in this situation and Black didn’t recognize it was safe to do so, put it there manually and Black will keep it.

Strings

Black prefers double quotes (" and """) over single quotes (' and '''). It will replace the latter with the former as long as it does not result in more backslash escapes than before.

Black also standardizes string prefixes, making them always lowercase. On top of that, if your code is already Python 3.6+ only or it’s using the unicode_literals future import, Black will remove u from the string prefix as it is meaningless in those scenarios.

The main reason to standardize on a single form of quotes is aesthetics. Having one kind of quotes everywhere reduces reader distraction. It will also enable a future version of Black to merge consecutive string literals that ended up on the same line (see #26 for details).

Why settle on double quotes? They anticipate apostrophes in English text. They match the docstring standard described in PEP 257. An empty string in double quotes ("") is impossible to confuse with a one double-quote regardless of fonts and syntax highlighting used. On top of this, double quotes for strings are consistent with C which Python interacts a lot with.

On certain keyboard layouts like US English, typing single quotes is a bit easier than double quotes. The latter requires use of the Shift key. My recommendation here is to keep using whatever is faster to type and let Black handle the transformation.

If you are adopting Black in a large project with pre-existing string conventions (like the popular “single quotes for data, double quotes for human-readable strings”), you can pass --skip-string-normalization on the command line. This is meant as an adoption helper, avoid using this for new projects.

Numeric literals

Black standardizes most numeric literals to use lowercase letters for the syntactic parts and uppercase letters for the digits themselves: 0xAB instead of 0XAB and 1e10 instead of 1E10. Python 2 long literals are styled as 2L instead of 2l to avoid confusion between l and 1. In Python 3.6+, Black adds underscores to long numeric literals to aid readability: 100000000 becomes 100_000_000.

For regions where numerals are grouped differently (like India and China), the -N or --skip-numeric-underscore-normalization command line option makes Black preserve underscores in numeric literals.

Line breaks & binary operators

Black will break a line before a binary operator when splitting a block of code over multiple lines. This is so that Black is compliant with the recent changes in the PEP 8 style guide, which emphasizes that this approach improves readability.

This behaviour may raise W503 line break before binary operator warnings in style guide enforcement tools like Flake8. Since W503 is not PEP 8 compliant, you should tell Flake8 to ignore these warnings.

Slices

PEP 8 recommends to treat : in slices as a binary operator with the lowest priority, and to leave an equal amount of space on either side, except if a parameter is omitted (e.g. ham[1 + 1 :]). It also states that for extended slices, both : operators have to have the same amount of spacing, except if a parameter is omitted (ham[1 + 1 ::]). Black enforces these rules consistently.

This behaviour may raise E203 whitespace before ':' warnings in style guide enforcement tools like Flake8. Since E203 is not PEP 8 compliant, you should tell Flake8 to ignore these warnings.

Parentheses

Some parentheses are optional in the Python grammar. Any expression can be wrapped in a pair of parentheses to form an atom. There are a few interesting cases:

  • if (...):
  • while (...):
  • for (...) in (...):
  • assert (...), (...)
  • from X import (...)
  • assignments like:
    • target = (...)
    • target: type = (...)
    • some, *un, packing = (...)
    • augmented += (...)

In those cases, parentheses are removed when the entire statement fits in one line, or if the inner expression doesn’t have any delimiters to further split on. If there is only a single delimiter and the expression starts or ends with a bracket, the parenthesis can also be successfully omitted since the existing bracket pair will organize the expression neatly anyway. Otherwise, the parentheses are added.

Please note that Black does not add or remove any additional nested parentheses that you might want to have for clarity or further code organization. For example those parentheses are not going to be removed:

return not (this or that)
decision = (maybe.this() and values > 0) or (maybe.that() and values < 0)

Call chains

Some popular APIs, like ORMs, use call chaining. This API style is known as a fluent interface. Black formats those by treating dots that follow a call or an indexing operation like a very low priority delimiter. It’s easier to show the behavior than to explain it. Look at the example:

def example(session):
    result = (
        session.query(models.Customer.id)
        .filter(
            models.Customer.account_id == account_id,
            models.Customer.email == email_address,
        )
        .order_by(models.Customer.id.asc())
        .all()
    )

Typing stub files

PEP 484 describes the syntax for type hints in Python. One of the use cases for typing is providing type annotations for modules which cannot contain them directly (they might be written in C, or they might be third-party, or their implementation may be overly dynamic, and so on).

To solve this, stub files with the .pyi file extension can be used to describe typing information for an external module. Those stub files omit the implementation of classes and functions they describe, instead they only contain the structure of the file (listing globals, functions, and classes with their members). The recommended code style for those files is more terse than PEP 8:

  • prefer ... on the same line as the class/function signature;
  • avoid vertical whitespace between consecutive module-level functions, names, or methods and fields within a single class;
  • use a single blank line between top-level class definitions, or none if the classes are very small.

Black enforces the above rules. There are additional guidelines for formatting .pyi file that are not enforced yet but might be in a future version of the formatter:

  • all function bodies should be empty (contain ... instead of the body);
  • do not use docstrings;
  • prefer ... over pass;
  • for arguments with a default, use ... instead of the actual default;
  • avoid using string literals in type annotations, stub files support forward references natively (like Python 3.7 code with from __future__ import annotations);
  • use variable annotations instead of type comments, even for stubs that target older versions of Python;
  • for arguments that default to None, use Optional[] explicitly;
  • use float instead of Union[int, float].