tests/testthat/test-package-topic.R

In tidymedia: Media File Preprocessing and Metadata for the 'tidyverse'

# Guards the package's landing topic (M55): `?tidymedia` resolves, and the topic
# stays listed in the installed help index.
#
# Both halves are easy to lose by accident. Dropping the roxygen block above the
# `"_PACKAGE"` sentinel makes roxygen skip it entirely and document() deletes
# tidymedia-package.Rd; adding back the `@keywords internal` line that
# usethis::use_package_doc() generates keeps the .Rd but drops the topic from
# the installed INDEX, so help(package = "tidymedia") stops listing it. Neither
# is visible in any other test.
#
# rd_sources() comes from helper-rd.R, which reads the source tree's man/ under
# devtools::test() and the parsed Rd database under R CMD check -- so this guard
# runs in both shapes rather than skipping in CI (LESSONS M51).

test_that("the package has a landing topic reachable as `?tidymedia`", {
  rd <- rd_sources()
  skip_if(is.null(rd), "no Rd source available")

  topic <- rd[names(rd) %in% c("tidymedia-package.Rd", "tidymedia-package")]
  expect_length(topic, 1L)

  # `?tidymedia` works through the alias, not the file name.
  expect_match(topic[[1]], "\\alias{tidymedia}", fixed = TRUE)
})

test_that("the landing topic is not keyworded internal", {
  rd <- rd_sources()
  skip_if(is.null(rd), "no Rd source available")

  topic <- rd[names(rd) %in% c("tidymedia-package.Rd", "tidymedia-package")]
  skip_if(length(topic) != 1L, "no landing topic")

  # \keyword{internal} is what excludes a topic from the installed INDEX, and so
  # from help(package = "tidymedia"). Measured at M55: with the keyword the
  # index carried 104 entries and no tidymedia-package row; without it, the row
  # is there.
  expect_false(grepl("\\keyword{internal}", topic[[1]], fixed = TRUE))
})

test_that("the vignette navigation paragraph is outside every section", {
  # M087 AC5. The paragraph pointing readers at the four vignettes is the
  # landing topic's closing navigation, not part of any one subject; it opened
  # inside \section{Session options} until M087 moved it. \section{} blocks are
  # top-level and come after \details{}, so appearing before the first one is
  # what "outside every section" means in a generated Rd.
  rd <- rd_sources()
  skip_if(is.null(rd), "no Rd source available")
  topic <- rd[names(rd) %in% c("tidymedia-package.Rd", "tidymedia-package")]
  skip_if(length(topic) != 1L, "no landing topic")
  txt <- topic[[1]]

  needle <- "for the guided tour"
  expect_match(txt, needle, fixed = TRUE)
  first_section <- regexpr("\\section{", txt, fixed = TRUE)
  # The topic really has sections: without one, "before the first section" is
  # vacuously true and this guard would pass on an Rd it cannot judge.
  expect_gt(first_section, 0L)
  expect_lt(regexpr(needle, txt, fixed = TRUE), first_section)
})

test_that("the landing topic qualifies the loudnorm pass's exit class", {
  # M087 AC4, via the review's F1. `?tidymedia` listed the loudnorm analysis
  # pass among the aborts "all classed tidymedia_ffmpeg_exit" with no
  # qualifier, while an executed call at the zero-exit site
  # (R/loudnorm_two_pass.R:112 -- the pass exited zero and printed no parseable
  # block) raises tidymedia_loudnorm_no_measurement alone. The defect is
  # over-attribution, not omission, so the pairing test over observed classes
  # cannot see it: the class named is one that site never raises.
  rd <- rd_sources()
  skip_if(is.null(rd), "no Rd source available")
  topic <- rd[names(rd) %in% c("tidymedia-package.Rd", "tidymedia-package")]
  skip_if(length(topic) != 1L, "no landing topic")
  txt <- topic[[1]]

  # The one place the topic attributes the exit class to the analysis pass:
  # the class's own \item, which since M127 opens with the class name. The
  # item must name the analysis pass and qualify it with the non-zero exit.
  items <- regmatches(txt, gregexpr("\\\\item [^\n]*(\n(?!\\\\item|\\})[^\n]*)*",
                                    txt, perl = TRUE))[[1]]
  exit_item <- items[startsWith(items, "\\item \\code{tidymedia_ffmpeg_exit}")]
  expect_length(exit_item, 1L)
  expect_match(exit_item, "loudnorm} analysis pass", fixed = TRUE)
  expect_match(exit_item, "exited non-zero", fixed = TRUE)
  # And the qualifier really scopes the pass: the two sit in one sentence, so
  # no full stop falls between them. Comparing positions could not fail,
  # because "exited non-zero" opens the item (M127 review O11).
  from <- regexpr("exited non-zero", exit_item, fixed = TRUE)
  to <- regexpr("loudnorm} analysis pass", exit_item, fixed = TRUE)
  expect_lt(from, to)
  expect_no_match(substr(exit_item, from, to), "\\.\\s")
})