Combining srcref Data
In covtracer: Contextualizing Tests

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

options(width = 120L)

library(covtracer)

There are two key relationships that we will explore. The first is a relationship of srcref objects, and the second is the relationship between namespace object definitions and their associated documentation.

Setup

Before we begin, we'll set up a demo coverage object and package namespace that we can use to showcase these relationships:

library(withr)
library(covr)

withr::with_temp_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,
    INSTALL_opts = c("--with-keep.source", "--install-tests")
  )
  examplepkg_cov <- covr::package_coverage(examplepkg_source_path)
  examplepkg_ns <- getNamespace("examplepkg")
})

Relational `srcref` data

First and foremost, we want to be able to associate srcref objects. These relationships are define the location of code. A srecref describes a region of code where the expression was pulled from, and we can compare these to determine whether a srcref is within, containing or independent of another.

This vignette will gloss over each of these tables. For more details see the Working with srcrefs vignette.

Linking `covr` traces to package object `srcref`s

It's important to note that coverage traces always sit within a package namespace object. Where a namespace object might have a srcref to the full code for a function, coverage traces trace individual expressions within that function.

To associate srcrefs by this relation, we provide a special joining function to combine data.frames by srcref columns.

traces_df <- trace_srcrefs_df(examplepkg_cov)
pkg_ns_df <- pkg_srcrefs_df(examplepkg_ns)

Just looking at these two data.frames, we can use the first trace and package object to illustrate the relationship:

cat("pkg  : ", format(pkg_ns_df$srcref["s3_example_func.list"]), "\n")
cat("trace: ", format(traces_df$srcref[1L]), "\n")

Although still a little arcane, you can see that the package object code contains the coverage trace. The package code spans lines 19-21, whereas the coverage trace lies in line 20. With this information, we can couple each package object with the coverage traces contained within each.

head(join_on_containing_srcrefs(traces_df, pkg_ns_df))

As expected, we can see that this test trace (now with the ".x" suffix) is mapped to the expected corresponding package namespace object.

Linking unit tests to evaluated `covr` traces

Although this relationship doesn't require any fancy srcref joining, we can associate tests and traces by a simple mapping of indices. FOr this, the test_trace_mapping() function is provided which will reshape a covr object (produced using options(covr.record_tests = TRUE)) to create a unified table across all covr traces:

head(test_trace_mapping(examplepkg_cov))

The test and trace columns contain the row indices in the respective test_srcrefs_df() and trace_srcrefs_df() data.frames, allowing for this data to be joined. However, since it is easy for a testing suite to cause the evaluation of an enormous number of traces, this matrix can become extremely long. It is recommended to do some aggregation or subsetting of this matrix before trying to use it to join more data-rich data.

You can also see that the evaluation order is stored (i), as well as the stack depth when it was evaluated (depth). With this added info, you might consider first filtering for only the first trace evaluated by each test, or to count all the times that a line of code was evaluated by each test by aggregating rows.

Relational documentation data

On the other side of the process, we also need to associate package objects with documentation. In many cases, this is trivial, and the name of the exported object can be used directly to find documentation as you have come to expect using ?<object>. This holds for simple functions. However, some objects are aliased to different documentation files or are built at package build time into internal representations, as is with S4 classes, and R6 classes.

To handle these cases, we can use the Rd_df() function to associate any available source code with a documentation file.

# filter for interesting columns for display
cols <- c("file", "alias", "doctype")
Rd_df(examplepkg_source_path)[, cols]

These aliases are also used when we use pkg_srcrefs_df() and can be used to associate srcrefs with .Rd files.

pkg_srcrefs_df(examplepkg_ns)

You'll see that we don't have any srcrefs associated with the "data" and "class" doctype documentation because these objects do not themselves have source code, even if there is source code in that was used to create them at bulid time.

Summary

With these relationships, we can build some really deep understandings of exactly what code a test evaluates and tie that test together with the documented behaviors.