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.
Note
YAPF is not an official Google product (experimental or otherwise), it is just code that happens to be owned by Google.
Contents
YAPF supports Python 2.7 and 3.4+.
To install YAPF from the source directory:
$ sudo python ./setup.py install
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] ...
Options:
usage: yapf [-h] [--style STYLE] [-d | -i] [-l START-END | -r] ...
Formatter for Python code.
positional arguments:
files
optional arguments:
-h, --help show this help message 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
-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
Note: after reformatting a chunk of code, YAPF verifies that it's correct (can be parsed by Python itself). This means that if you're reformatting Python 3 code, it's best to run YAPF itself under Python 3. The same goes for Python 2.
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 either
one of the predefined styles (currently "pep8" or "google"), or a path to a
configuration file that specifies the desired style. The 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).
We wanted to use clang-format's reformatting algorithm. It's very powerful and designed to come up with the best formatting possible. Existing tools were created with different goals in mind, and would require extensive modifications to convert to using clang-format's algorithm.
Please do! YAPF was designed to be used as a library as well as a command line tool. This means that a tool or IDE plugin is free to use YAPF.
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 # 2For 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!