Plotting Test Paths
In covtracer: Contextualizing Tests

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

options(width = 120L)

```{css, echo=FALSE} img { border: none; margin: 0; }

## Setup

As with all covtracer analysis, we need to start by collecting coverage traces
of a package. Below is an example where a package is installed with the
necessary flags such that the coverage traces can be collected.

```r
library(covtracer)

library(withr)
library(covr)

init_libs <- .libPaths()
dir.create(lib <- tempfile("covtracer_pkgs_"))
.libPaths(c(lib, .libPaths()))

options(keep.source = TRUE, keep.source.pkg = TRUE, covr.record_tests = TRUE)
examplepkg_source_path <- system.file("examplepkg", package = "covtracer")

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

examplepkg_cov <- covr::package_coverage(examplepkg_source_path)
examplepkg_ns <- getNamespace("examplepkg")

ttdf <- covtracer::test_trace_df(examplepkg_cov, aggregate_by = NULL)

As well, for this analysis we will use a few supporting packages.

library(dplyr)
library(igraph)

Preparing Graph Data

Before we use the test data, we will clean our incoming data, removing untested records and filtering out untestable objects like S4 class definitions.

ttdf <- ttdf %>%
  filter(!is.na(test_name)) %>%
  filter(is.na(doctype) | !doctype %in% "class") %>%
  select(test_name, alias, is_exported, i) %>%
  arrange(test_name, i) %>%
  mutate(test_id = cumsum(!duplicated(test_name)))

head(ttdf)

Create Edges of Our Test Path

Our test-trace dataframe has an index of test expressions, each linked to the traces that they evaluate, with added order of evaluation, i. To prepare this for visualization, we want to convert this to a dataframe where each record describes a step of this process. Instead of a test linking to a trace with an index, each jump in the test path should link from the calling expression to the evaluated expression.

edges_df <- ttdf %>%
  split(.$test_name) %>%
  lapply(function(sdf) {
    unique(data.frame(
      from = c(sdf$test_name[[1L]], head(sdf$alias, -1L)),
      to = sdf$alias
    ))
  }) %>%
  bind_rows() %>%
  distinct()

head(edges_df)

Likewise, we want to capture some metadata about each vertex. Since a vertex in this context can be either a test or a trace, we have some data that is captured differently for each class of vertex.

test_names <- Filter(Negate(is.na), unique(ttdf$test_name))
obj_names <- Filter(Negate(is.na), unique(ttdf$alias))

n_tests <- length(test_names)
n_objs <- length(obj_names)

vertices_df <- data.frame(
  name = c(test_names, obj_names),
  color = rep(c("cornflowerblue", "darkgoldenrod"), times = c(n_tests, n_objs)),
  label = c(sprintf("Test #%d", seq_along(test_names)), obj_names),
  test_id = c(seq_along(test_names), rep_len(NA, n_objs)),
  is_test = rep(c(TRUE, FALSE), times = c(n_tests, n_objs)),
  is_exported = c(rep_len(NA, n_tests), ttdf$is_exported[match(obj_names, ttdf$alias)])
)

vertices_df <- vertices_df %>%
  mutate(color = ifelse(is_exported, "goldenrod", color))

vertices_df %>%
  select(name, label) %>%
  head()

# for whatever reason... this fixes errors when building vignettes on R-devel
edges_df
vertices_df

Plotting Our Test Paths

Finally, we can plot this network of test executions:

g <- igraph::graph_from_data_frame(edges_df, vertices = vertices_df)

par(mai = rep(0, 4), omi = rep(0, 4L))
plot.igraph(g,
  vertex.size = 8,
  vertex.label = V(g)$label,
  vertex.color = V(g)$color,
  vertex.label.family = "sans",
  vertex.label.color = "black",
  vertex.label.dist = 1,
  vertex.label.degree = -pi / 2,
  vertex.label.cex = 0.8,
  mark.border = NA,
  margin = c(0, 0.2, 0, 0.2)
)

legend(
  "bottomleft",
  inset = c(0.05, 0),
  legend = c("test", "exported function", "unexported function"),
  col = c("cornflowerblue", "goldenrod", "darkgoldenrod"),
  pch = 16,
  bty = "n"
)

Naturally, there are a plethora of wonderful visualization packages available that accept igraph data as input. This graph could just as well be plotted with the visNetwork package, though it is omitted to keep this example analysis minimal.