In Genentech/covtracer: Contextualizing Tests

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

options(
  tibble.print_min = 5,
  tibble.print_max = 5,
  width = 120L
)

suppressPackageStartupMessages(library(dplyr))

covtracer

Tools for contextualizing tests, built using covr test traces. This package provides utilities for linking an assortment of test and package information to paint a more complete picture of how a test was performed.

flowchart LR
    tests[Tests] <--> traces[Traced Exprs] <--> code[Package Code] <--> docs[Package Documentation]

Installation

To install, use remotes to install directly from GitHub

Functionality hinges heavily on coverage objects prepared using covr (≥ 3.5.1.9003). To ensure suggested dependency requirements are met, install with dependencies = TRUE (satisfy all dependencies).

# will install covr >= v3.5.1.9003 for examples
remotes::install_github("Genentech/covtracer", dependencies = TRUE)

Motivation

Tests are not black boxes. When it comes to verifying behaviors of code, we can use observations about the code that is executed by a test to build a more complete picture of exactly what the test does. This is a core part of software validation. By combining information about each test, the tested package code and linking that code to package documentation, we can link documented behaviors to their respective tests.

Getting Started

Test traces are connected to evaluated code using covr (≥ 3.5.1.9003). Likewise, a new option flag (covr.record_tests) must be set in order to record tests alongside the coverage traces. Finally, the package to evaluate must be installed with source references in order to map all the components together.

That's a lot to configure, but if you're in a position where this test data is valuable hopefully it's worth the setup.

library(covtracer)

# additional demo packages
library(dplyr)
library(withr)
library(covr)

withr::with_temp_libpaths({
  pkg <- system.file("examplepkg", package = "covtracer")

  install.packages(
    pkg,
    type = "source",
    repos = NULL,
    quiet = TRUE,
    INSTALL_opts = c("--with-keep.source", "--install-tests")
  )

  options(covr.record_tests = TRUE)
  cov <- covr::package_coverage(pkg)

  ttdf <- test_trace_df(cov)
})

There's a lot of info in the resulting data.frame, but we'll focus on just the critical piece, showing which tests evaluate code related to which documented behaviors. Below we show how one might map unit tests to evaluated, documented objects.

Note: Below we ignore documentation for datasets and S4 class constructors. Although these are defined in the package, they don't map to testable lines of code because they are constructed when the package is built.

traceability_matrix <- ttdf %>%
  filter(!doctype %in% c("data", "class")) %>% # ignore objects without testable code
  select(test_name, file) %>%
  filter(!duplicated(.)) %>%
  arrange(file)

traceability_matrix

We can quickly see which functions or methods are entirely untested.

Use Cases

The data.frame returned by test_trace_df contains a ton of information, and we can measure a few dimensions of the quality of tests with some relatively straightforward analysis.

Traceability Matrix

Perhaps the most immediate use case is to map unit tests to documented behaviors.

ttdf %>%
  filter(!doctype %in% c("data", "class")) %>% # ignore objects without testable code
  select(test_name, file) %>%
  filter(!duplicated(.)) %>%
  arrange(file)

Finding Untested Behaviors

Once we can map unit testing to documentation, we can filter down to only documentation that is not covered by any test.

ttdf %>%
  filter(!doctype %in% c("data", "class")) %>% # ignore objects without testable code
  select(test_name, count, alias, file) %>%
  filter(is.na(count)) %>%
  arrange(alias)

Filter For Only Directly Tested Behaviors

Some tests evaluate a broad set of functionality by calling functions that themselves call out to internal package functions. This is often perfectly fine, since the mechanisms of calling those internal functions are limited by the surfaced user-facing functions. Nevertheless, whether a function is called directly is a good indication of the "unit"-ness of a unit test. You may consider only the coverage of directly tested functions.

ttdf %>%
  filter(!doctype %in% c("data", "class")) %>% # ignore objects without testable code
  select(direct, alias) %>%
  group_by(alias) %>%
  summarize(any_direct_tests = any(direct, na.rm = TRUE)) %>%
  arrange(alias)