In rorynolan/exampletestr: Help for Writing Unit Tests Based on Function Examples

knitr::opts_chunk$set(echo = TRUE, comment = "#>")

Let us take a look at how many packages on CRAN employ unit tests by checking whether their source code contains a non-empty tests/ directory (using \url{https://rdrr.io}). We examine first all packages and then the subset that were authored or updated in 2016 or 2017. We will also compare the popularity of the testing frameworks testthat, RUnit, svUnit svUnit and unitizer [code adapted from mango].

required_pkgs <- c("tidyverse", "stringr", "lubridate", "rvest", "RCurl")
invisible(lapply(required_pkgs, library, character.only = TRUE))
check_for_tests <- function(pkg_name) {
  message(pkg_name)
  rdrr_address <- paste0("https://rdrr.io/cran/", pkg_name)
  if (url.exists(rdrr_address)) {
    if (inherits(try(read_html(rdrr_address), silent = TRUE), "try-error")) {
      NA  # Allow for error reading rdrr.io page
    } else {
      rdrr_page_text <- rdrr_address %>% read_html %>% html_text
    trimmed_lines <- str_split(rdrr_page_text, "\n") %>% unlist %>% str_trim()
    trimmed_lines %>% str_detect("^tests/.+") %>% any
    }
  } else {
    NA  # allow for not found on rdrr.io
  }
}

download.file("http://cran.R-project.org/web/packages/packages.rds",
              "packages.rds", mode = "wb")
cran_packages_info <- readRDS("packages.rds")
cran_packages <- cran_packages_info[, "Package"]
post2015 <- ymd(cran_packages_info[, "Published"]) >= ymd("2016-01-01")
new_packages <- cran_packages[post2015]

tested_check <- cran_packages %>%
  map_lgl(check_for_tests) %>%
  set_names(cran_packages) %>%
  na.omit  # allow for not found on rdrr
n_packages <- length(tested_check)
n_tested_packages <- sum(tested_check)
print(c(n_packages, n_tested_packages))
new_packages <- intersect(new_packages,
                          names(tested_check))  # allow for not found on rdrr
n_new_packages <- length(new_packages)
n_tested_new_packages <- sum(tested_check[new_packages])
print(c(n_new_packages, n_tested_new_packages))
unit_test_packages <- c("testthat", "RUnit", "svUnit", "unitizer")
reverse_deps <- tools:::package_dependencies(packages = unit_test_packages,
                          cran_packages_info, recursive=FALSE, reverse=TRUE,
                          which = c("Depends","Imports","LinkingTo", "Suggests"))
framework_use <- map_int(reverse_deps, length)
print(framework_use)

So we see that (at the time of writing) of the r n_packages packages on CRAN, r round(n_tested_packages / n_packages * 100)% (r n_tested_packages) are unit tested. Also, of the r n_new_packages packages authored or updated since 1st January 2016, r round(n_tested_new_packages / n_new_packages * 100)% (r n_tested_new_packages) are unit tested. We also see that r names(framework_use)[which.max(framework_use)] is the most popular testing framework, preferred in r round(max(framework_use) / sum(framework_use) * 100)% of cases.