Testthat fails when setting up rms by calling datadist() + options()

Question

I'm trying to do some unit testing using the testthat package but I can't seem to get it to work properly together with the rms package. The following example:

library(rms)
set.seed(10)
ds <- data.frame(
  ftime = rexp(200),
  fstatus = sample(0:1,200,replace=TRUE),
  x1 = runif(200),
  x2 = runif(200),
  x3 = factor(sample(LETTERS[1:3], size=200, replace=TRUE)))


ddist <- datadist(ds)
options(datadist="ddist")

s <- Surv(ds$ftime, ds$fstatus == 1)
fit <- cph(s ~ x1 + x2 + x3, data=ds)

returns this error:

Error in Design(eval.parent(m)) : dataset ddist not found for options(datadist=)

This even though print(ddist) works and the options("datadist") returns the proper variable. Does testthat have a different variable scope that causes errors?

Update

I run the testthat by a R console started in the my package dir (Eclipse StatET):

library(testthat)
test_dir("inst/tests")
q()

The same error occurs with the R CMD check --as-cran

Answer 1

While @agstudy's suggestion is correct I've figured out a simple workaround for the bug by using the <<- operator that assigns the variable to the global environment, here's a test-file that works:

set.seed(10)
n <- 11
ds <- data.frame(
  y = rnorm(n),
  x1 = factor(sample(c("a", "aa", "aaa"), size = n, replace = TRUE)))

suppressMessages(library(rms))
dd <<- datadist(ds)
options(datadist = "dd")

context("rms")
test_that("test", {
  fit <- ols(y ~ x1, data=ds)
  s <- summary(fit)
  expect_true(inherits(s, "summary.rms"))
})

This works also if you happen to do the assignment within the test_that:

context("rms")
test_that("test", {
  set.seed(10)
  n <- 11
  ds <- data.frame(
    y = rnorm(n),
    x1 = factor(sample(c("a", "aa", "aaa"), size = n, replace = TRUE)))

  suppressMessages(library(rms))
  dd <<- datadist(ds)
  options(datadist = "dd")

  fit <- ols(y ~ x1, data=ds)
  s <- summary(fit)
  expect_true(inherits(s, "summary.rms"))
})

This is also equivalent to the following code (perhaps easier to understand):

env <- globalenv() # Grab the global environment
env$dd <- datadist(ds) # Assign the datadist to it

If you want to learn more about how environments work I can recommend Hadley's excellent Advanced R coverage of the topic. I found this explaining many of the issues that I was running into.

Answer 2

yes It is a scope problem as suggested by the error.

A possible work around is to define your ds where you call test_dir

for example You create file, runtest.R like this

library(rms)
set.seed(10)
ds <- data.frame(
  ftime = rexp(200),
  fstatus = sample(0:1,200,replace=TRUE),
  x1 = runif(200),
  x2 = runif(200),
  x3 = factor(sample(LETTERS[1:3], size=200, replace=TRUE)))
ddist <- datadist(ds)
options(datadist="ddist")
library(testthat)
test_dir("inst/tests")