A code formatter with rules set in stone.

Runic is a formatter for the Julia programming language built on top of JuliaSyntax.jl.

Similarly to gofmt, Runic have no configuration. The formatting rules are set in stone (although not yet complete). This approach is something that is appreciated by most Go programmers, see for example the following quote:

Gofmt's style is no one's favorite, yet gofmt is everyone's favorite.

• Installation
• Usage
  • CLI
  • Editor integration
• Checking formatting
  • Github Actions
  • Git hooks
  • Ignore formatting commits in git blame
• Formatting specification

Runic can be installed with Julia's package manager:

julia -e 'using Pkg; Pkg.add(url = "https://github.com/fredrikekre/Runic.jl")'

For CLI usage and editor integration (see Usage) it is recommended to install Runic in a separate project such as e.g. the shared project @runic:

julia --project=@runic -e 'using Pkg; Pkg.add(url = "https://github.com/fredrikekre/Runic.jl")'

The main interface to Runic is the command line interface (CLI) through the main function invoked with the -m flag. See the output of julia -m Runic --help below for usage details.

The following snippet can be added to your shell startup file so that the CLI can be invoked a bit more ergonomically. This assumes Runic is installed in the @runic shared project as suggested in the Installation section above. Adjust the --project flag if you installed Runic elsewhere.

alias runic="julia --project=@runic -m Runic"

julia -e 'using Runic; exit(Runic.main(ARGS))' -- <args>

alias runic="julia --project=@runic -e 'using Runic; exit(Runic.main(ARGS))' --"

$ julia -m Runic --help
NAME
       Runic.main - format Julia source code

SYNOPSIS
       julia -m Runic [<options>] <path>...

DESCRIPTION
       `Runic.main` (typically invoked as `julia -m Runic`) formats Julia source
       code using the Runic.jl formatter.

OPTIONS
       <path>...
           Input path(s) (files and/or directories) to process. For directories,
           all files (recursively) with the '*.jl' suffix are used as input files.
           If no path is given, or if path is `-`, input is read from stdin.

       -c, --check
           Do not write output and exit with a non-zero code if the input is not
           formatted correctly.

       -d, --diff
           Print the diff between the input and formatted output to stderr.
           Requires `git` to be installed.

       --help
           Print this message.

       -i, --inplace
           Format files in place.

       -o <file>, --output=<file>
           File to write formatted output to. If no output is given, or if the file
           is `-`, output is written to stdout.

In addition to the CLI there is also the two function Runic.format_file and Runic.format_string. See their respective docstrings for details.

Runic can be used as a formatter in Neovim using conform.nvim. Refer to the conform.nvim repository for installation and setup instructions.

Runic is not (yet) available directly in conform so the following configuration needs to be passed to the setup function. This assumes Runic is installed in the @runic shared project as suggested in the Installation section above. Adjust the --project flag if you installed Runic elsewhere.

require("conform").setup({
    formatters = {
        runic = {
            command = "julia",
            args = {"--project=@runic", "-e", "using Runic; exit(Runic.main(ARGS))"},
        },
    },
    formatters_by_ft = {
        julia = {"runic"},
    },
    default_format_opts = {
        -- Increase the timeout in case Runic needs to precompile
        -- (e.g. after upgrading Julia and/or Runic).
        timeout_ms = 10000,
    },
})

Note that conform (and thus Runic) can be used as formatexpr for the gq command. This is enabled by adding the following to your configuration:

vim.o.formatexpr = "v:lua.require('conform').formatexpr()"

Runic can be used as a formatter in VS Code using the extension Custom Local Formatters.

After installing the extension you can configure Runic as a local formatter by adding the following entry to your settings.json:

"customLocalFormatters.formatters": [
    {
      "command": "julia --project=@runic -e 'using Runic; exit(Runic.main(ARGS))'",
      "languages": ["julia"]
    }
]

Using the "Format Document" VS Code command will now format the file using Runic. Note that the first time you execute the command you will be prompted to select a formatter since the Julia language extension also comes with a formatter.

Runic has a check-mode that verifies whether files are correctly formatted or not. This mode is enabled with the --check flag. In check mode Runic will exit with a non-zero code if any of the input files are incorrectly formatted. As an example, the following invocation can be used:

git ls-files -z -- '*.jl' | xargs -0 --no-run-if-empty julia --project=@runic -m Runic --check --diff

This will run Runic's check mode (--check) on all .jl files in the repository and print the diff (--diff) if the files are not formatted correctly. If any file is incorrectly formatted the exit code will be non-zero.

You can use fredrikekre/runic-action to run Runic on Github Actions:

name: Runic formatting
on:
  push:
    branches:
      - 'master'
      - 'release-'
    tags:
      - '*'
  pull_request:
jobs:
  runic:
    name: Runic
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      # - uses: julia-actions/setup-julia@v2
      #   with:
      #     version: '1'
      # - uses: julia-actions/cache@v2
      - uses: fredrikekre/runic-action@v1
        with:
          version: '1'

See fredrikekre/runic-action for details.

Runic can be used together with pre-commit using fredrikekre/runic-pre-commit. After installing pre-commit you can add the following to your .pre-commit-config.yaml to run Runic before each commit:

repos:
  - repo: https://github.com/fredrikekre/runic-pre-commit
    rev: v1.0.0
    hooks:
      - id: runic

See fredrikekre/runic-pre-commit for details.

If you don't want to use pre-commit you can also use a plain git hook. Here is an example hook (.git/hooks/pre-commit):

#!/usr/bin/env bash

# Redirect output to stderr.
exec 1>&2

# Run Runic on added and modified files
git diff-index -z --name-only --diff-filter=AM master | \
    grep -z '\.jl$' | \
    xargs -0 --no-run-if-empty julia --project=@runic -m Runic --check --diff

When setting up Runic formatting for a repository for the first time (or when upgrading to a new version of Runic) the formatting commit will likely result in a large diff with mostly non functional changes such as e.g. whitespace. Since the diff is large it is likely that it will show up and interfere when using git-blame. To ignore commits during git-blame you can i) add them to a file .git-blame-ignore-revs and ii) tell git to use this file as ignore file by running

git config blame.ignoreRevsFile .git-blame-ignore-revs

See the git-blame documentation for details.

For example, such a file may look like this:

# Adding Runic formatting
<commit hash of formatting commit>

# Upgrading Runic from 1.0 to 2.0
<commit hash of formatting commit>

This is a list of things that Runic currently is doing:

• Toggle formatting
• Line width limit
• Newlines in blocks
• Indentation
• Explicit return
• Spaces around operators, assignment, etc
• Spaces around keywords
• Multiline listlike expressions
• Spacing in listlike expressions
• Trailing semicolons
• Literal floating point numbers
• Literal hex and oct numbers
• Parentheses around operator calls in colon
• in instead of ∈ and =
• Braces around right hand side of where
• Whitespace miscellaneous

It is possible to toggle formatting around expressions where you want to disable Runic's formatting. This can be useful in cases where manual formatting increase the readability of the code. For example, manually aligned array literals may look worse when formatted by Runic.

The source comments # runic: off and # runic: on will toggle the formatting off and on, respectively. The comments must be on their own line, they must be on the same level in the syntax tree, and they must come in pairs. An exception to the pairing rule is made at top level where a # runic: off comment will disable formatting for the remainder of the file. This is so that a full file can be excluded from formatting without having to add a # runic: on comment at the end of the file.

For example, the following code will toggle off the formatting for the array literal A:

function foo()
    a = rand(2)
    # runic: off
    A = [
        -1.00   1.41
         3.14  -4.05
    ]
    # runic: on
    return A * a
end

No. Use your Enter key or refactor your code.

The body of blocklike expressions (e.g. if, for, while, function, struct, etc.) always start and end with a newline. Examples:

-if c x end
 if c
     x
 end

-function f(x) x^2 end
 function f(x)
     x^2
 end

An exception is made for empty blocks so that e.g.

struct A end

is allowed.

Consistently four spaces for each indentation level.

Standard code blocks (function, for, while, ...) all increase the indentation level by one until the closing end. Examples:

 function f()
-  for i in 1:2
-    # loop
-  end
-  while rand() < 0.5
-    # loop
-  end
     for i in 1:2
         # loop
     end
     while rand() < 0.5
         # loop
     end
 end

Listlike expressions like e.g. tuples, function calls, array literals, etc. also increase the indentation level by one until the closing token. This only has an effect if the list span multiple lines. Examples:

 x = (
-  a, b, c, d,
-  e, f, g, h,
     a, b, c, d,
     e, f, g, h,
 )

 foo(
-  a, b, c, d,
-  e, f, g, h,
     a, b, c, d,
     e, f, g, h,
 )

 [
-  a, b, c, d,
-  e, f, g, h,
     a, b, c, d,
     e, f, g, h,
 ]

The examples above both result in "hard" indentation levels. Other expressions that span multiple lines result in "soft" indentation levels. The difference between the two is that soft indentation levels don't nest (this is really only applicable to multiline operator call chains).

 using Foo:
-  foo, bar
     foo, bar

 x = a   b  
-  c
     c

 x = a ? b :
-  c
     c

Without soft indentation levels operators chains can result in ugly (but logically correct) indentation levels. For example, the following code:

x = a   b *
        c  
    d

would be "correct". Such a chain looks better the way it is currently formatted:

x = a   b *
    c  
    d

Explicit return statements are ensured in function and macro definitions by adding return in front of the last expression, with some exceptions listed below.

• If the last expression is a for or while loop (which both always evaluate to nothing) return is added after the loop.
• If the last expression is a if or try block the return is only added in case there is no return inside any of the branches.
• If the last expression is a let or begin block the return is only added in case there is no return inside the block.
• If the last expression is a macro call, the return is only added in case there is no return inside the macro.
• No return is added in short form functions (f(...) = ...), short form anonymous functions ((...) -> ...), and do-blocks (f(...) do ...; ...; end).
• If the last expression is a function call, and the function name is (or contains) throw or error, no return is added. This is because it is already obvious that these calls terminate the function and don't return any value.

Note that adding return changes the expression in a way that is visible to macros. Therefore it is, in general, not valid to add return to a function defined inside a macro since it isn't possible to know what the macro will expand to. For this reason this formatting rule is disabled for functions defined inside macros with the exception of some known and safe ones from Base (e.g. @inline, @generated, ...).

For the same reason mentioned above, if the last expression in a function is a macro call it isn't valid to step in and add return inside. Instead the return will be added in front of the macro call like any other expression (unless there is already a return inside of the macro as described above).

Examples:

 function f(n)
-    sum(rand(n))
     return sum(rand(n))
 end

 macro m(args...)
-    :(generate_expr(args...))
     return :(generate_expr(args...))
 end

• If the last expression is a if or try block it might be better to recurse into the branches and add return there. Looking at real code, if a function ends with an if block, it seems about 50/50 whether adding return after the block or adding return inside the branches is the best choice. Quite often return if is not the best but at least Runic's current formatting will force to think about the return value. See issue #52.

Runic formats spaces around infix operators, assignments, comparison chains, and type comparisons (binary <: and >:), and some other operator-like things. If the space is missing it will be inserted, if there are multiple spaces it will be reduced to one. Examples:

-1 2*3
-1     2  *  3
 1   2 * 3
 1   2 * 3

-x=1
-x= 1
-x =1
-x. =1
 x = 1
 x =  1
 x  = 1
 x . = 1
-1<2>3
-1  <  2  >  3
 1 < 2 > 3
 1 < 2 > 3

-T<:Integer
-T  >:  Integer
 T <: Integer
 T >: Integer

-x->x
-a  ?  b  :  c
 x -> x
 a ? b : c

Note that since Runic's rules are applied consistently, no matter the context or surrounding code, the "spaces around assignment" rule also means that there will be spaces in keyword arguments in function definitions and calls. Examples:

-foo(; a=1) = a
-foo(a=1)
 foo(; a = 1) = a
 foo(a = 1)

Exceptions to the rule above are :, .., ^, ::, and unary <: and >:. These are formatted without spaces around them. Examples:

-a : b
 a:b

-a ^ 5
 a^5

-a :: Int
 a::Int

-<: Integer
->:  Integer
 <:Integer
 >:Integer

• Perhaps the rule for some of these should be "at least one space" instead. This could help with alignment issues. Discussed in issue #12.

Consistently use single space around keywords. Examples:

-struct  Foo
 struct Foo

-mutable  struct  Bar
 mutable struct Bar

-function  foo(x::T)  where  {T}
 function foo(x::T) where {T}

Listlike expressions (tuples, function calls/definitions, array literals, etc.) that already span multiple lines are formatted to consistently have a leading and a trailing newline. Trailing commas are enforced for array/tuple literals (where adding another item is common) but optional for function/macro calls/definitions.

-(a,
-    b)
 (
     a,
     b,
 )

-foo(a,
-    b)
 foo(
     a,
     b
 )

-[1 2
- 3 4]
 [
     1 2
     3 4
 ]

Note that currently there is no line-length limit employed so expressions that only take up a single line, even if they are long, are not formatted like the above. Thus, only expressions where the original author have "committed" to mulitples lines are affected by this rule.

Listlike expressions (tuples, function calls/definitions, array literals, etc.) use a consistent rule of no space before , and a single space after ,. Trailing commas are enforced for array/tuple literals (where adding another item is common) but optional for function/macro calls/definitions. Leading/trailing spaces are removed. Examples:

-f(a,b)
-(a,b)
-[a,  b]
 f(a, b)
 (a, b)
 [a, b]


-(a,b,)
 (a, b)
 (
     a,
-    b
     b,
 )

-( a, b )
 (a, b)

• Perhaps the rule for some of these should be "at least one space" instead. This could help with alignment issues. Discussed in issue #12.

Trailing semicolons are removed in the body of blocklike expressions. Examples

 function f(x)
-    y = x^2;
-    z = y^2; # z = x^4
-    return z;
     y = x^2
     z = y^2  # z = x^4
     return z
 end

Trailing semicolons at top level and module level are kept since they are sometimes used there for output supression (e.g. Documenter examples or scripts that are copy-pasted/included in the REPL).

Floating point literals are normalized so that they:

• always have a decimal point
• always have a digit before and after the decimal point
• never have leading zeros in the integral and exponent part
• never have trailing zeros in the fractional part
• always use e instead of E for the exponent

Examples:

-1.
-.1
 1.0
 0.1

-01.2
-1.0e01
-0.10
 1.2
 1.0e1
 0.1

-1.2E5
 1.2e5

• Always add the implicit for the exponent part, i.e. 1.0e 1 instead of 1.0e1. Discussed in issue #13.
• Allow multiple trailing zeros in the fractional part, i.e. don't change 1.00 to 1.0. Such trailing zeros are sometimes used to align numbers in literal array expressions. Discussed in issue #14.

Hex literals are padded with zeros to better highlight the resulting type of the literal: UInt8 to 2 characters, UInt16 to 4 characters, UInt32 to 8 characters etc. Examples:

-0x1
-0x123
-0x12345
 0x01
 0x0123
 0x00012345

Add parentheses around operator calls in colon expressions to better highlight the low precedence of :. Examples:

-1   2:3 * 4
-1   2:3
-1:3 * 4
 (1   2):(3 * 4)
 (1   2):3
 1:(3 * 4)

The keyword in is used consistently instead of ∈ and = in for loops. Examples:

-for i = 1:2
 for i in 1:2

-for i ∈ 1:2
 for i in 1:2

Note that ∈ not replaced when used as an operator outside of loop contexts in order to be symmetric with ∉ which doesn't have a direct ASCII equivalent. See #17 for more details.

Braces are consistently used around the right hand side of where expressions. Examples:

-T where T
-T where T <: S where S <: Any
 T where {T}
 T where {T <: S} where {S <: Any}

Trailing spaces are removed in code and comments (but not inside of multiline strings where doing so would change the meaning of the code). Examples:

-1   1 
 1   1

-x = 2 # x is two 
 x = 2 # x is two

Tabs are replaced with spaces. Example:

-function f()
-	return 1
-end
 function f()
     return 1
 end

Extra vertical spacing is trimmed so that there are at maximum two empty lines between expressions. Examples:

-function f()
-     x = 1
-
-
-
-    return x
-end
 function f()
      x = 1
 
 
     return x
 end

Any newlines at the start of a file are removed and if the file ends with more than one newline the extra ones are removed.