yapf 0.2.3

Introduction

Most of the current formatters for Python — e.g., autopep8, and pep8ify — are made to remove lint errors from code. This has some obvious limitations. For instance, code that conforms to the PEP 8 guidelines may not be reformatted. But it doesn’t mean that the code looks good.

YAPF takes a different approach. It’s based off of ‘clang-format’, developed by Daniel Jasper. In essence, the algorithm takes the code and reformats it to the best formatting that conforms to the style guide, even if the original code didn’t violate the style guide. The idea is also similar to the ‘gofmt’ tool for the Go programming language: end all holy wars about formatting - if the whole code base of a project is simply piped through YAPF whenever modifications are made, the style remains consistent throughout the project and there’s no point arguing about style in every code review.

The ultimate goal is that the code YAPF produces is as good as the code that a programmer would write if they were following the style guide. It takes away some of the drudgery of maintaining your code.

Installation

To install YAPF from PyPI:

$ pip install yapf

YAPF is still considered in “alpha” stage, and the released version may change often; therefore, the best way to keep up-to-date with the latest development is to clone this repository.

Note that if you intend to use YAPF as a command-line tool rather than as a library, installation is not necessary. YAPF supports being run as a directory by the Python interpreter. If you cloned/unzipped YAPF into DIR, it’s possible to run:

$ PYTHONPATH=DIR python DIR/yapf [options] ...

Python versions

YAPF supports Python 2.7 and 3.4.1+.

YAPF requires the code it formats to be valid Python for the version YAPF itself runs under. Therefore, if you format Python 3 code with YAPF, run YAPF itself under Python 3 (and similarly for Python 2).

Usage

Options:

usage: yapf [-h] [--version] [--style-help] [--style STYLE] [--verify]
                 [-d | -i] [-l START-END | -r]
                 [files [files ...]]

Formatter for Python code.

positional arguments:
  files

optional arguments:
  -h, --help            show this help message and exit
  --version             show version number and exit
  --style-help          show style settings and exit
  --style STYLE         specify formatting style: either a style name (for
                        example "pep8" or "google"), or the name of a file
                        with style settings. The default is pep8 unless a
                        .style.yapf file located in one of the parent
                        directories of the source file (or current directory
                        for stdin)
  --verify              try to verify refomatted code for syntax errors
  -d, --diff            print the diff for the fixed source
  -i, --in-place        make changes to files in place
  -l START-END, --lines START-END
                        range of lines to reformat, one-based
  -r, --recursive       run recursively over directories

Formatting style

The formatting style used by YAPF is configurable and there are many “knobs” that can be used to tune how YAPF does formatting. See the style.py module for the full list.

To control the style, run YAPF with the --style argument. It accepts one of the predefined styles (e.g., pep8 or google), a path to a configuration file that specifies the desired style, or a dictionary of key/value pairs.

The config file is a simple listing of (case-insensitive) key = value pairs with a [style] heading. For example:

[style]
based_on_style = pep8
spaces_before_comment = 4
split_before_logical_operator = true

The based_on_style setting determines which of the predefined styles this custom style is based on (think of it like subclassing).

It’s also possible to do the same on the command line with a dictionary. For example:

--style='{based_on_style: google, indent_width: 4}'

This will take the google base style and modify it to have four space indentations.

Example

An example of the type of formatting that YAPF can do, it will take this ugly code:

x = {  'a':37,'b':42,

'c':927}

y = 'hello ''world'
z = 'hello '+'world'
a = 'hello {}'.format('world')
class foo  (     object  ):
  def f    (self   ):
    return       37*-+2
  def g(self, x,y=42):
      return y
def f  (   a ) :
  return      37+-+a[42-x :  y**3]

and reformat it into:

x = {'a': 37, 'b': 42, 'c': 927}

y = 'hello ' 'world'
z = 'hello ' + 'world'
a = 'hello {}'.format('world')


class foo(object):
    def f(self):
        return 37 * -+2

    def g(self, x, y=42):
        return y


def f(a):
    return 37 + -+a[42 - x:y ** 3]

Example as a module

The two main APIs for calling yapf are FormatCode and FormatFile, these share several arguments which are described below:

>>> from yapf.yapf_api import FormatCode  # reformat a string of code

>>> FormatCode("f ( a = 1, b = 2 )")
'f(a=1, b=2)\n'

A style_config argument: Either a style name or a path to a file that contains formatting style settings. If None is specified, use the default style as set in style.DEFAULT_STYLE_FACTORY.

>>> FormatCode("def g():\n  return True", style_config='pep8')
'def g():\n    return True\n'

A lines argument: A list of tuples of lines (ints), [start, end], that we want to format. The lines are 1-based indexed. It can be used by third-party code (e.g., IDEs) when reformatting a snippet of code rather than a whole file.

>>> FormatCode("def g( ):\n    a=1\n    b = 2\n    return a==b", lines=[(1, 1), (2, 3)])
'def g():\n    a = 1\n    b = 2\n    return a==b\n'

A print_diff (bool): Instead of returning the reformatted source, return a diff that turns the formatted source into reformatter source.

>>> print(FormatCode("a==b", filename="foo.py", print_diff=True))
--- foo.py (original)
+++ foo.py (reformatted)
@@ -1 +1 @@
-a==b
+a == b

Note: the filename argument for FormatCode is what is inserted into the diff, the default is <unknown>.

FormatFile returns reformatted code from the passed file along with its encoding:

>>> from yapf.yapf_api import FormatFile  # reformat a file

>>> print(open("foo.py").read())  # contents of file
a==b

>>> FormatFile("foo.py")
('a == b\n', 'utf-8')

The in_place argument saves the reformatted code back to the file:

>>> FormatFile("foo.py", in_place=True)
(None, 'utf-8')

>>> print(open("foo.py").read())  # contents of file (now fixed)
a == b

(Potentially) Frequently Asked Questions

# yapf: disable
FOO = {
    # ... some very large, complex data literal.
}

BAR = [
    # ... another large data literal.
]
# yapf: enable

BAZ = {
    (1, 2, 3, 4),
    (5, 6, 7, 8),
    (9, 10, 11, 12),
}  # yapf: disable

Gory Details

Algorithm Design

The main data structure in YAPF is the UnwrappedLine object. It holds a list of FormatTokens, that we would want to place on a single line if there were no column limit. An exception being a comment in the middle of an expression statement will force the line to be formatted on more than one line. The formatter works on one UnwrappedLine object at a time.

An UnwrappedLine typically won’t affect the formatting of lines before or after it. There is a part of the algorithm that may join two or more UnwrappedLines into one line. For instance, an if-then statement with a short body can be placed on a single line:

if a == 42: continue

YAPF’s formatting algorithm creates a weighted tree that acts as the solution space for the algorithm. Each node in the tree represents the result of a formatting decision — i.e., whether to split or not to split before a token. Each formatting decision has a cost associated with it. Therefore, the cost is realized on the edge between two nodes. (In reality, the weighted tree doesn’t have separate edge objects, so the cost resides on the nodes themselves.)

For example, take the following Python code snippet. For the sake of this example, assume that line (1) violates the column limit restriction and needs to be reformatted.

def xxxxxxxxxxx(aaaaaaaaaaaa, bbbbbbbbb, cccccccc, dddddddd, eeeeee):  # 1
    pass                                                               # 2

For line (1), the algorithm will build a tree where each node (a FormattingDecisionState object) is the state of the line at that token given the decision to split before the token or not. Note: the FormatDecisionState objects are copied by value so each node in the graph is unique and a change in one doesn’t affect other nodes.

Heuristics are used to determine the costs of splitting or not splitting. Because a node holds the state of the tree up to a token’s insertion, it can easily determine if a splitting decision will violate one of the style requirements. For instance, the heuristic is able to apply an extra penalty to the edge when not splitting between the previous token and the one being added.

There are some instances where we will never want to split the line, because doing so will always be detrimental (i.e., it will require a backslash-newline, which is very rarely desirable). For line (1), we will never want to split the first three tokens: def, xxxxxxxxxxx, and (. Nor will we want to split between the ) and the : at the end. These regions are said to be “unbreakable.” This is reflected in the tree by there not being a “split” decision (left hand branch) within the unbreakable region.

Now that we have the tree, we determine what the “best” formatting is by finding the path through the tree with the lowest cost.

And that’s it!