This project follows the contributing recommendations outlined by saamwerk. In particular, issues labelled with Status: Postponed are closed even if they are not resolved.

Contributing code

  • Open a PR only when your idea was approved of by a contributor in an issue.
  • Follow guidelines in tidyverse style guide for your code contributions.
  • Make sure your commit pass the pre-commit hooks in this repo. See the precommit on how to install the pre-commit framework and the R package on your system and then run precommit::use_precommit() to make sure the hooks are activated in your local styler clone. If you skip a hook, describe why in the PR.

How to dive in and understanding the source code

Read the vignettes. If you are done, come back here.



style_text("call(1, 2 + 1)")

Go broad before you go deep. Before going into the very deep layers of function calls of style_text(), try to understand that style_text() consists of a few function calls only. Go into each of them and try to understand one layer deep. That is, try to understand what make_transformer() does by reading the names of the functions that get called, the name of the objects that are created by assigning the output of these function calls. Before looking into a functions source code, look at the documentation for that function. All internal important functions are documented and documentation is available also for unexported objects via ? (if you did devtools::load_all()). Then, go into parse_transform_serialize(), and so on.

To understand the most fundamental operation in styler, the manipulation of the columns related to spacing and line break information, pick a rule from R/rules-*.R (e.g. R/rules-spacing), add a break point to a rule, and style a string where you think this rule will be active. Then, see what happens and how this rule is applied on each level of nesting.

Static code analysis

There are multiple packages that can be used to analyze a code base:

  • gitsum: Parses and summarises git repository history.
  • parsesum: Analyses source code through parsing.

Check out the links above to see how the tools listed could help you understanding styler.

Project setup

  • The package is developed with {devtools} suite, which includes {roxgen2} for documentation, {testthat} for unit testing, {pkgdown} for HTML documentation.
  • Continuous integration uses github-actions.
  • A key development principle of styler is to separate infrastructure from style guide. Hence, whenever possible, transformer functions should be adapted, instead of changing the infrastructure for a specific style guide.
  • {styler} was created in 2017 by Kirill Müller. It was then turned from a proof-of-concept into a ready-for-production tool as part of GSOC 2017 with Kirill Müller and Yihui Xie as mentors and Lorenz Walthert as student.

File Structure

The source code is organized as follows:

File Description
addins.R ui and helpers for the Addins of styler.
communicate.R function to communicate to the user via the console.
compat-dplyr.R compatibility functions. Since styler does not depend on dplyr, we define the dplyr functions ourself.
compat-tidyr.R compatibility functions. Since styler does not depend on tidy, we define the tidyr functions ourself.
expr-is.R Functions to check whether an expression matches a predicate (e.g. whether it is a function call, a curly brace expression etc.).
indent.R Computation of whether indention is needed (needs_indention()), if so which indices are indented and how indention is it is triggered.
initialize.R initializer called with the visitor at each nest.
nest.R converting from a text representation into a flat and then into a nested parse table representation.
nested-to-tree.R utilities to create a tree representation from text (after text was converted into a nested parse table).
parse.R parse text into parse table, minor token manipulation, verification of parsed objects.
reindent.R Deals with token-dependent indention and re-indention, opposed to indent.R where all indention is token independent (i.e. a brace just adds one level of indention, whereas in function declaration headers (if mutli-line), indention depends on token position of “function”).
relevel.R Reorganizing the nested parse table, namely relocates expressions on both sides of “%>%” to the same nest.
rules-line-break.R, rules-other.R, rules-replacement.R, rules-spacing.R transformer rules
serialize.R converts flattened parse table into text representation. Complement operation to the functions in nest.R
set-assert-args.R Assertion and setting of arguments.
style-guides.R How to create style guide objects from transformers.
styler.R General package information.
testing.R function used for testing.
token-create.R Utilities for creating tokens, mostly to insert braces around mutli-line if statements.
token-define.R Defines which tokens belong to which group.
transform-code.R, transform-files.R Transformation of code for APIs that manipulate files (e.g. style_file()).
ui.R User interaces. Top-level functions for styling.
unindent.R Certain tokens cause unindention, e.g. closing braces.
utils.R low-level general purpose utilities.
vertical.R S3 class for pretty printing of styled code.
visit.R Functions that apply functions to each level of nesting, either inside out or outside in.
zzz.R backport imports.

Obtaining contextual information

You may have problems understanding some code because documentation is minimal, some code / functions seem to solve problems you don’t understand or handle cases that seem unreasonable or otherwise incomprehensible. You can resort to the following strategies:

  • Use full-text search to see where functions are defined or called and how different parts of {styler} depend on it.
  • Use $ git blame to see where changes were introduced. Look at the commit message, check changes that were made to the code in the same commit. If you are using the GUI of GitHub, you can easily obtain more contextual information such as the pull request with which a change was introduced. Often, functionality was introduced with testing. So, you can easily see which new tests are related to the new functionality. You can remove the changes in the source code and re-run the tests and see what fails and why.
  • Search Issues and Pull Requests on GitHub with the full text search. Make sure you also search for closed Issues and PRs.

High-level conventions

  • The project follows a highly functional approach. This means that functionality should be capsuled into functions, even if they are only called once. This makes abstraction from the code easier, reduces the number of lines for each function declaration considerably, and makes it easier for people not familiar with the codebase to dive into it.
  • All internal functions (except if they are 100% self-explanatory) are to be documented.
  • New functionality (e.g. in terms of styling rules) needs to be unit tested. If the new functionality changes how code is to be styled, the infrastructure with test_collection() should be used.
  • Cases that are not yet formatted correctly can be labelled with a FIXME.
  • GitHub is the platform where communication about source code happens. We refrain from adding extensive in-line code comments. One can use $ git blame to track when changes were introduced and find the corresponding pull request and associated issues to understand the thought process that lead to a change in the source code. This also implies that issues and / or pull request contain verbose explanation of problems and solutions provided.

Low-level coventions

This project follows the tidyverse style guide. If we refer to specific variables / values etc. in the following sections, you can use RStudio’s full text search to find where remove_line_break_before_round_closing_after_curly() is declared or called.


  • File names only contain alphanumeric characters and dashes.
  • Files are named according to topics / contexts, not according to functions that live in these files.


  • Function names should be verbs. No abbreviations should be used, we don’t care if function names are particularly long. For example, there is a function with the name remove_line_break_before_round_closing_after_curly().
  • Only very low-level functions or functions that don’t fit in any other file go to utils.R.

Control Flow

  • Conditional statements should always evaluate to TRUE or FALSE, i.e. we don’t encourage if (length(x)), but rather if (length(x) > 0L).
  • We avoid loops whenever possible and use functions like purrr::map() and friends when possible and prefer them over R base counterparts like base::lapply().

Boolean Values

Functions that return Boolean values or variables that hold Boolean values are to be prefixed with is or has. For example, is_rmd_file(path) is a function that returns TRUE if path is the path to a .Rmd file and FALSE otherwise.

Vectors with indices

Vectors that hold indices are often suffixed with idx. For example, else_idx indicates for every row in a parse table whether it contains an else token.


The use of closures is discouraged. We prefer to prefill a template function with purrr::partial().


We have a testing framework powered by test_collection(). Essentially, there is an *-in.R file and a *-out.R file. The *-in.R file is the input that is transformed and - if it matches the *-out.R file, the test will pass. You can create an *-in.R file, run devtools::test(f = "[your file]") and an *-out.R file is generated. If the file matches your expectation, you can commit it. Note that files are overwritten and version control should be used to track failed tests. The files are placed in tests/testthat under the category they fit. Please have a look at the documentation for test_collection() and see other unit tests.


Please open an issue if something is unclear so that we can improve the contributing guidelines.