Debug Code Using Empty RStudio Jobs

Running jobs in clean sessions is a great to test code in isolation without having to launch new RStudio applications.

Case: forgot to rename a variable

Here’s an embarrassing error I commit too often: Say I want to write a function that computes the Mean Squared Error (MSE) of a numerical vector. First, I fiddle around with it in the console, ending up with:

tmp = rnorm(1000, sd = 2)
mean((tmp - mean(tmp))^2)

## [1] 4.016175

It works! Then I wrap the code in a function and try it out:

mse = function(x) {
  mean((tmp - mean(tmp))^2)
}

mse(rnorm(1000, sd = 1))

## [1] 4.016175

Huh!? It runs but the MSE should be around 1 here, not 4.

Solution: fast debugging with job::empty()

You can select the code and click the addin “Run selection as job in empty session”. Or you can do the exact same action using code via job::empty():

job::empty({
  mse = function(x) {
    mean((tmp - mean(tmp))^2)
  }

  mse(rnorm(1000, sd = 1))
})

In the job, you’ll see the error message:

#> Error in mean((tmp - mean(tmp))^2) : object 'tmp' not found 

Aha, by running the code in a separate session, we learned that it only seemed to work because I had the tmp variable lingering in my environment. Errors like this become less trivial as the code (and your environment) grows more complex.

Selective import

In the example above, you can, of course, import selectively too. These two jobs are identical, but I prefer the first.

# Add to empty
job::empty({
  mse(rnorm(1000, sd = 1))
}, import = c(mse))

# Avoid from full
job::job({
  mse(rnorm(1000, sd = 1))
}, import = c(mse), packages = NULL, opts = NULL)

testthat unit tests

testthat unit tests have access to your current environment so it is prone to the kind of errors introduced above. job::empty() to the rescue:

job::empty({devtools::test()})
job::empty({devtools::check()})  # while we're at it

This also frees up your console in the meantime.