erlfmt

erlfmt is an opinionated Erlang code formatter. By automating the process of formatting the code, it frees your team up to focus on what the code does, instead of what it looks like.

erlfmt is feature complete and released as version 1.0 This means only backwards compatible changes and bug fixes can be adopted without very serious consideration. We do not want to put users in the position where they need to reformat code without a very good reason.

Before

Remember reading code before erlfmt and having arguments with co workers :(

what_is(Erlang) ->
case Erlang of movie->[hello(mike,joe,robert),credits]; language->formatting_arguments end
.

After

Now, with the new erlfmt, code is readable and you get along with your co workers :D

what_is(Erlang) ->
    case Erlang of
        movie -> [hello(mike, joe, robert), credits];
        language -> no_more_formatting_arguments
    end.

Disclaimer: erlfmt is just a code formatter, not a solution to all life's problems.

Comparison with other Erlang formatters
Usage
Line length
Design principles
Manual interventions
Respecting original format
Ignoring Formatting
Join the erlfmt community

Comparison with other Erlang formatters

See the comparison with other erlang formatters document for more details.

Usage

Rebar3

The easiest way to use erlfmt is as a rebar plugin, by adding to your rebar.config:

{project_plugins, [erlfmt]}.

This will provide a new rebar3 fmt task. All erlfmt command-line options can be configured with defaults in your rebar.config, for example:

{erlfmt, [write]}.

Now you can format all the files in your project by running:

$ rebar3 fmt

And you can add the following command in your CI to ensure your Erlang is formatted:

$ rebar3 fmt --check

For more usage instructions, see RebarUsage

Escript

Alternatively, you can build a standalone and portable escript and use erlfmt without rebar (it still requires Erlang to be installed on the target system).

$ rebar3 as release escriptize
$ _build/release/bin/erlfmt -h

You can then run it from the command line:

$ erlfmt -w './otp/lib/*/{src,include}/*.{erl,hrl}'

Requirements

erlfmt requires Erlang/OTP 21+ and works on all platforms.

Integrations

Visual Studio Code's Erlang Formatter extension.
How to integrate with doom emacs
Use erlfmt through rebar3_format

Add your integration here, by making a pull request.

Line length

erlfmt enforces a consistent style by parsing your code and re-printing it, while enforcing a selected maximum line-length.

For example, this line that exceeds the length limit:

scenario(dial_phone_number(),  ring(), hello(mike),hello(joe), hello(robert),   system_working(), seems_to_be())

will be re-printed automatically in a vertical style:

scenario(
    dial_phone_number(),
    ring(),
    hello(mike),
    hello(joe),
    hello(robert),
    system_working(),
    seems_to_be()
)

But this snippet:

hello(mike, joe, robert)

will be kept as-is, since it fits in a single line.

Note: The enforcing of line-length is best effort and will sometimes overrun the selected line length, because the algorithm is greedy.

Design principles

The formatter was designed with these main principles in mind:

First, the formatter never changes the semantics or structure of the code. This means the input AST and the output AST are equivalent. The formatter does not try to arbitrarily "improve" the code. For the most part it limits its behaviour to shifting whitespace around - it won't rewrite literals, add parentheses, reorder exports, etc.

The second principle is to provide as little configuration as possible. This removes contention points and makes it easier to achieve the goal of consistent code. Instead of providing configuration, the formatter respects a limited set of choices made in the original code to preserve intent and make it easier to achieve beautiful code based on contextual hints.

Furthermore, the formatter avoids special cases as much as possible. For example, there is no hard-coded behaviour specific to some function - all functions are laid out the same. There are some clear layout rules and general structures that are re-used as much as possible between different constructs. For example, the general layout of lists, functions, maps, records, and similar, all follow the same "container" rules.

Finally, the formatter should be idempotent. Formatting the code once should produce the same output as formatting it multiple times.

Manual interventions

In some cases, the formatter rules might lead to code that looks decent, but not perfect. Therefore some manual intervention to help the formatter out might be needed. For example, given the following code:

split_tokens([{TokenType, Meta, TokenValue} | Rest], TokenAcc, CommentAcc) ->
    split_tokens(Rest, [{TokenType, token_anno(erl_anno:to_term(Meta), #{}), TokenValue} | TokenAcc], CommentAcc).

Because the line-length is exceeded, the formatter will produce the following:

split_tokens([{TokenType, Meta, TokenValue} | Rest], TokenAcc, CommentAcc) ->
    split_tokens(
        Rest,
        [{TokenType, token_anno(erl_anno:to_term(Meta), #{}), TokenValue} | TokenAcc],
        CommentAcc
    ).

It might be more desirable, though, to extract a variable and allow the call to still be rendered in a single line, for example:

split_tokens([{TokenType, Meta, TokenValue} | Rest], TokenAcc, CommentAcc) ->
    Token = {TokenType, token_anno(erl_anno:to_term(Meta), #{}), TokenValue},
    split_tokens(Rest, [Token | TokenAcc], CommentAcc).

A similar situation could happen with long patterns in function heads, for example let's look at this function:

my_function(
    #user{name: Name, age: Age, ...},
    Arg2,
    Arg3
) ->
    ...

Even though the code is perfectly valid, you might prefer not to split the arguments across multiple lines and move the pattern extraction into the function body instead:

my_function(User, Arg2, Arg3) ->
    #user{name: Name, age: Age, ...} = User,
    ...

Such transformations cannot be automated since the formatter is not allowed to change the AST of your program. After running the formatter, especially if running it for the first time on a sizeable codebase, it's recommended to inspect the code manually to correct similar sub-par layouts.

Respecting original format

The formatter keeps the original decisions in two key places

when choosing between a "collapsed", "semi-expanded", and an "expanded" layout for containers
when choosing between single-line and multi-line clauses.

In containers

For containers like lists, tuples, maps, records, function calls, etc, there are three possible layouts - "collapsed" where the entire collection is printed in a single line; "semi-expanded" where the enclosing brackets/breaces/parentheses are printed on a line of their own, but all elements are printed in a single line; and "expanded" where each element is printed on a separate line. The formatter respects this choice, if possible. If there is a newline between the opening bracket/brace/parenthesis and the first element, the collection will be always printed "semi-

Erlfmt

Install / Use

README