R Style Guide

Patrick Schratz edited this page Feb 11, 2018 · 34 revisions

Version 1.0, written by Bernd Bischl

Every programmer knows that code is read more than it is written. Not having a consistent coding style is error prone and annoying for the reader. It also looks very unprofessional. Of course, such a style convention should have been defined on the language level many, many years ago by R core. Unfortunately, it was not. So now every group or single package developer has their own style - or none at all.

Good alternatives are

FIXMEs:

  • check the above sites for nice stuff I forgot

In some cases I have shamelessly stolen from the guides above, and in quite a few places this guide deviates.

A few side remarks: Nobody really enjoys reading such standard definitions and I can easily imagine more enjoyable things than writing them. And while for some of the following rules arguments could (maybe) be made, why they are good rules, others are based on a subjective feeling of aesthetics or simply what I became accustomed to over the years. If people start arguing religiously over where to place curly braces, better go away and talk to less weird guys. In summary, I am not proposing that this is the best set of rules that exists. What is important is that you follow SOME style set. What follows is our personal choice. Stick to it if you collaborate with us on our packages.

  1. Language: Code and documentation is always written in English, never in German, French or whatever. The same holds for file and directory names.

  2. File names: File names must always be meaningful. Don't make them too short, don't make them too long. No special characters or whitespaces occur in them. Stick to [a-z], [A-Z], [0-9] [_,-] nearly always. Source files end in .R not .r or .txt or something else. R Data files end in .RData, not .rdata or .rda or something else.

  3. Line length: Maximum line length is somewhere between 80 and 100 characters.

  4. Tabs: Tabs are converted to spaces and tab width is 2. Configure your editor of choice for this.

  5. Indentation: Always indent one level after opening a curly brace and remove one level after closing one. Indentation width is one tab width, which is 2 spaces. When a line break occurs inside parentheses, do not align the wrapped line with the first character inside the parenthesis, but also indent 2 spaces.

    Good:

    for (i in 1:3) {
      y = y + i
    }
    
    doThatThing(arg1 = "a_long_string_is_passed", arg2 = "a_long_string_is_passed_here_too",
      arg3 = "another_long_string")
    

    Bad:

    for (i in 1:3) {
    y = y + i
    print("foo")
    }
    
    doThatThing(arg1 = "a_long_string_is_passed", arg2 = "a_long_string_is_passed_here_too",
                arg3 = "another_long_string")
    
  6. Strings: Define strings with double quotes, so "hello" instead of 'hello'.

  7. Always use TRUE and FALSE instead of T and F.

  8. Always write integers as 1L instead of 1: Because 1 actually means 1.0, a numeric, in R.

  9. Write 1:3 instead of c(1:3): 1:3 is already an (integer) vector.

  10. Assignment operator: Use = instead of <- for assignments.

Good:

a = 3

Bad:

a <- 3

Other people have the same preference and you can read arguments in favor and against it.

  1. Spacing around operators and commas: Place one space on each side of most binary operators (=, ==, +, -, etc). This also applies when passing named arguments to functions, e.g., f(foo = 5). No space before a comma, one space after a comma.

    Good:

    a = 3
    
    f(a, b)
    
    f(a + b)
    
    f(arg = "value")
    
    if (a == b) {
      ...
    }

    Bad:

    a=3
    
    a =3
    
    f(a,b)
    
    f(a , b)
    
    f(a+b)
    
    if (a==b) {
      ...
    }

    Some exceptions which are better readable without spaces:

    1:n
    
    2^k
    
    x[i+1, j-1]
    
    1:(i+1)
  2. Curly braces: The opening curly brace goes on the same line as the respective syntactic element it belongs to and never on its own line. Always have one space before the opening brace {. The closing curly brace } goes on its own line, except if it occurs before an else statement, see the following point on if statements.

    Good:

    for (i in 1:3) {
      y = y + i
    }

    Bad:

    for (i in 1:3) {y = y + i}
    
    for (i in 1:3){
      y = y + i
    }
    
    for (i in 1:3)
    {
      y = y + i
    }
  3. if, for, while statements: Put a single space in between if, while, repeat and its following, opening parenthesis (. Do not write if (ok == TRUE) if ok already is a boolean value, write if (ok). If the body of the statement consists of only one line, the language allows us to omit the curly braces. This can be a good thing if it keeps the code together (less scrolling is better reading and understanding), but you should only use this when the code structure is very simple, not with, e.g., complicated, nested if statements. If you use it, always put the single line body on a separate line and indent it. If in doubt, always use the braces. If you use curly braces with else, the curly brace before the else goes on the same line as the else.

    Good:

    if (condition) {
      ...
    }
    
    if (condition) {
      ...
    } else {
      ...
    }
    
    # in rare cases OK
    if (condition)
      x = 1
    
    for (i in 1:10) {
      ...
    }
    
    while (not.done) {
      ...
    }

    Bad:

    if(condition) {
      ...
    }
    
    if (condition) {
      ...
    }
    else {
      ...
    }
    
  4. No extra whitespaces for parenthesis: Do not put whitespaces before or after the parenthesis ( and ) when defining or calling functions.

    Good:

    doIt(1, 4)

    Bad:

    doIt( 1, 4 )
    doIt (1, 4)
  5. Return statement: Try to explicitly use the return statement and do not rely on R's convention to return the value of the last evaluated expression of the called function, especially if your function is longer and you return in multiple places. Deviate if your function is shorter, e.g., for short anonymous functions. The basic distinction is whether you have used an imperative or a functional coding style for the respective function. R allows both and I mix both styles heavily. If your function is more like a procedure, i.e., it has no meaningful return value, return invisible(NULL).

  6. One command per line and semicolon: Put every statement / command in its own line. Do not put a semicolon at the end of a statement. This is R not C.

    Bad:

    x = 1;
    x = 1; y = 2; z = 3
  7. Use whitespace lines to structure your code: Do not put arbitrary empty lines in your code, but instead use them sparsely to structure your code into "blocks of actions" that make sense. Usually, you want to put at least a short comment line before such a block that explains its contents, see next point. This structuring guides the reader and allows him to catch his breath.

  8. Comments: Use a single # (not two ##), then one space, on the same level of indentation as the code you comment, to start a comment line. Usually, you should not put a comment on the same line as the code you comment. Combine meaningful identifier names and well written code, which is as self-documenting as possible, with short, precise lines of comments. Complicated stuff needs lengthier comments. No or too few comments are bad, but too verbose or unnecessary comments are also (less) bad. Usually, it is good style to prefix smaller "blocks of code", e.g., half a page of a for loop, where you "do a certain thing" with 1-2 comment lines that explain what is going to happen now.

  9. Function names: Functions are named in "verb style", written in camel case and the name begins with a lowercase letter. Names have to be meaningful and are important, hence, invest some time to find good, expressive names. Don't make them too short, don't make them too long. In case of doubt, choose the longer, more expressive name, but don't overdo it.

    Good:

    doThatThing()

    Bad:

    doThatThingYouKnowWhatIMeanItIsReallyCool()
    dtt()
    do.that.thing()
    dothatthing()
    do_that_thing()
  10. Variable and function argument names: Use lowercase letters and separate words with a dot. This allows to visually discriminate functions from arguments / variables. (Yes, in R functions are first-class objects. We can live with that and this won't hurt or confuse us in 99.999% of cases. If you did not understand this subtle point, do not worry.) Names have to be meaningful, they are important, hence, invest some time to find good, expressive names. Don't make them too short, don't make them too long. Here is a rule of thumb to decide whether a name should be shorter or longer: Is the variable used in various places of a long and complicated function after it was introduced? Make the name longer and very precise. Is the variable used in a local, very restricted context only and its meaning pretty clear from the context? A shorter name is probably not only OK, but even better.

    Good:

    multiply = function(a, b) {
      ...
    }
    
    writeLinesToFile = function(file.path, lines, show.info = TRUE) {
      ...
    }
    
    for (i in 1:10) {
      vec[i] = 1
    }

    Bad:

    writeLinesToFile = function(filePath, lines, showInfo = TRUE) {
      ...
    }
    
    # name too long, simply "i" would be OK
    for (the.iterator in 1:10) {
      vec[the.iterator] = 1
    }
  11. Documenting and argument checks: Use roxygen2 to document your functions. A good starting point to get familiar with documenting R code is the documentation section of the chapter about R package development from Hadleys book.

    For example take a glance at the documentation of the clipString function in the BBmisc package.

    See how we also document the argument and return types in a formal fashion? Always do that. Yes, R is a dynamically typed language but in more than 90% of my function definitions I have a certain type in mind for my function arguments. This holds true for many R packages, but very often this information is not easily available from the help page. Also mention the default value, even if it is visible in the usage line.

    A scalar integer parameter
    #' @param n (`integer(1)`)\cr
    #'   My argument.
    #'   Default is 1.
    
    An integer vector of arbitrary size
    #' @param x ([integer])\cr
    #'   The vector.
    
    An S3 object of a class you defined in the same package.
    #' @param obj ([MyS3Class}])\cr
    #'   A nice object.
    
    Data type for returned object.
    #' @return (`integer(1)`)\cr
    #'   Returns the length of the vector.
    

    You should also check the validity of the input arguments of your function. This provides useful feedback to client users when they make mistakes and can be even helpful for yourself and make bug detection easier. In order to make your life easier in that regard, BBmisc provides the functions checkArg and checkListElementClass.

    Here is the relevant section from clipString that performs these checks:

    clipString = function(x, len, tail="...") {
      checkArg(x, "character", na.ok=TRUE)
      len = convertInteger(len)
      checkArg(tail, "character", len=1L, na.ok=FALSE)
      checkArg(len, "integer", len=1L, na.ok=FALSE, lower=nchar(tail))
      ...
    }

    Note that these checks come with a certain (small) overhead, so you should maybe not perform them in fast running functions that get called a few thousand times in client code. We are currently writing a new, C-based arg-check system which will not have this overhead problem anymore.

  12. Code distribution in files: Try to put one single function definition into one .R file. Name the file like the function. If you have some very short helper functions you can deviate from this.

  13. Function length and abstraction: Good functions very often cover 1 to 3 screen pages. Of course, some complicated stuff sometimes is longer. If that happens, think about introducing another level of indirection, e.g., more functions or data types. Maybe this is a good time for refactoring? If your function or source file covers 5000 lines of code (Have seen those. Not just once.) you are doing it wrong - and your code will not be maintainable.

  14. R scripts: library / require and source statements. Always put these into a "header" block at the beginning of your scripts. First the library statements, then the source statements. NEVER use absolute file paths in your source statements. Define a project-related working directory (usually either the top level directory or something like a "src" subdirectory of your project) that all paths refer to and use relative paths. If that does not work for you for some reason, define ONE path variable at the beginning of your script that can easily be changed in one place and use it with file.path. Note that this option is still worse than the one described before.

    Good:

    library(MASS)
    library(rpart)
    
    source("foo.R")
    source("bar.R")
    source("lib/helpers.R")

    Somewhat worse alternative:

    source.path = "/home/bischl/myproject/src"
    source(file.path(source.path, "foo.R"))
    source(file.path(source.path, "bar.R"))
    source(file.path(source.path, "lib", "helpers.R"))

    Bad:

    source("C:/Documents and Settings/Joe/Desktop/myproject/foo.R")
    
    x = 1
    doIt(x)
    
    library(rpart)
    
    source("C:/Documents and Settings/Joe/Desktop/myproject/bar.R")
    
    # and so on ad nauseam, we have all seen this

    Some people prefer library("MASS") and require("MASS") to library(MASS), but I don't care and usually use the latter.

  15. Local helper functions defined in a parent function: Can be OK, if the inner function is only used in this context and pretty simple. Otherwise try to avoid.

  16. Nearly never use global variables.

  17. FIXMEs: If you discover something bad or suspicious, and it is not a bug which you enter into the issue tracker, comment the problem and add # FIXME:. Do not use TODO instead. Be precise in the description and err on the side of verbosity, otherwise other people (possibly including yourself) will not understand what you meant when they read this in the future. If you use a proper editor, it will help you searching through these issues later.

  18. Object-oriented programming: Use S3 instead of S4. S4 results in code bloat without real benefits. Yes, S3 sucks too, but less so. No final verdict on reference classes. Whether an OO-style of programming makes sense for your R project cannot be answered in general. If you need to define your own abstract data types and respective operations on them, it likely does make sense to use OO.

  19. Don't repeat yourself: Copy-paste code is always an indicator that something is wrong.

  20. Exceptions to the rules: Intelligent and experienced people stick to their style definition 99% of the time and are able to recognize the 1% of cases where deviations are not only OK, but better. In case of doubt, stick to the law.

  21. The existing code base always has preference: If the style of an already existing project differs from the above, stick to its style.

You can’t perform that action at this time.
You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session.
Press h to open a hovercard with more details.