Reprex do’s and don’ts

Main requirements

Use the smallest, simplest, most built-in data possible.

• Think: iris or mtcars. Bore me.
• If you must make some objects, minimize their size and complexity.
• Many of the functions and packages you already use to import data from delimited files also offer a way to create a small data frame “inline”:
  • read.table() and friends have a text argument. Example: read.csv(text = "a,b\n1,2\n3,4").

  • tibble::tribble() lets you use a natural and readable layout. Example:

      tibble::tribble(
        ~ a, ~ b,
          1,   2,
          3,   4
      )
      #> # A tibble: 2 x 2
      #>       a     b
      #>   <dbl> <dbl>
      #> 1     1     2
      #> 2     3     4

• Get just a bit of something with head() or by indexing with the result of sample(). If anything is random, consider using set.seed() to make it repeatable.
• dput() is a good way to get the code to create an object you have lying around, if you simply cannot make do with built-in or simulated data. Copy and paste the result of this into your reprex.
• Look at official examples and try to write in that style. Consider adapting one.

Include commands on a strict “need to run” basis.

• Ruthlessly strip out anything unrelated to the specific matter at hand.
• Include every single command that is required, e.g. loading specific packages via library(foo).

Consider including so-called “session info”, i.e. your OS and versions of R and add-on packages, if it’s conceivable that it matters.

• Use reprex(..., si = TRUE) for this.

Whitespace rationing is not in effect.

• Use good coding style.
• Use reprex(..., style = TRUE) to request automated styling of your code.

Pack it in, pack it out, and don’t take liberties with other people’s computers. You are asking people to run this code!

• Don’t start with rm(list = ls()). It is anti-social to clobber other people’s workspaces.
• Don’t start with setwd("C:\Users\jenny\path\that\only\I\have"), because it won’t work on anyone else’s computer.
• Don’t mask built-in functions, i.e. don’t define a new function named c or mean.

• If you change options, store original values at the start, do your thing, then restore them:

  opar <- par(pch = 19)
  <blah blah blah>
  par(opar)

• If you create files, delete them when you’re done:

  write(x, "foo.txt")
  <blah blah blah>
  file.remove("foo.txt")

• Don’t delete files or objects that you didn’t create in the first place.

• Take advantage of R’s built-in ability to create temporary files and directories. Read up on tempfile() and tempdir().

This seems like a lot of work!

Yes, creating a great reprex requires work. You are asking other people to do work too. It’s a partnership.

80% of the time you will solve your own problem in the course of writing an excellent reprex. YMMV.

The remaining 20% of the time, you will create a reprex that is more likely to elicit the desired behavior in others.

Further reading:

How to make a great R reproducible example? thread on StackOverflow

How to write a reproducible example from Hadley Wickham’s Advanced R book

Package philosophy

The reprex code:

• Must run and, therefore, should be run by the person posting. No faking it.

• Should be easy for others to digest, so they don’t necessarily have to run it. You are encouraged to include selected bits of output. :scream:

• Should be easy for others to copy + paste + run, if and only if they so choose. Don’t let inclusion of output break executability.

Accomplished like so:

• Use rmarkdown::render() to run the code and capture output that you would normally see on your screen. This is done in a separate R process, via callr, to guarantee it is self-contained.

• Use chunk option comment = "#>" to include the output while retaining executability.

Other work

If I had known about formatR::tidy_eval(), I probably would never had made reprex! But alas I did not. AFAICT here are the main differences:

• reprex() accepts an expression as primary input, in addition to code on the clipboard, in a character vector, or in a file.
• reprex() runs the reprex in a separate R process, via callr. tidy_eval() uses the existing R process and offers an envir argument.
• reprex() writes the code to a .R file and calls rmarkdown::render(). tidy_eval() runs the code line-by-line via capture.output(eval(..., envir = envir)).
• reprex() uploads figures to imgur and inserts the necessary link.