Adding or modifying checks
In pkgcheck: Package Checks for 'rOpenSci'

knitr::opts_chunk$set (
    collapse = TRUE,
    width = 120,
    fig.retina = 2,
    fig.path = "README-"
)

This vignette describes how to modify or add new checks to the existing suite of checks implemented by pkgcheck. Each of the internal checks is defined in a separate file in the R directory of this package with the prefix of check_ (or checks_ for files which define multiple, related checks). Each check requires two main functions:

::: {.alert .alert-info} - One defining the check itself, which must have a prefix pkgchk_, followed by the name of the check; and - One defining summary and print methods based on the result of the first function, which must have a prefix output_pkgchk_. :::

The structure of these two function are described in the following two sections.

Both of these functions must accept a single input parameter of a pkgcheck object, by convention named checks. This object is a list of four main items:

pkg which summarises data extracted from pkgstats::pkgstats(), and includes essential information on the package being checked.
info which contains information used in checks, including info$git detailing git repository information, info$pkgstats containing a summary of a few statistics generated from pkgstats::pkgstats(), along with statistical comparisons against distributions from all current CRAN packages, an info$network_file specifying a local directory to a vis.js visualisation of the function call network of the package, and an info$badges item containing information from GitHub workflows and associated badges, where available.
checks which contains a list of all objects returned from all pkgchk_...() functions, which are used as input to output_pkgchk_...() functions.
meta containing a named character vector of versions of the core packages used in pkgcheck.

pkgcheck objects generally also include a fifth item, goodpractice, containing the results of goodpractice checks. The checks item passed to each pkgchk_...() function contains all information on the package, info, meta, and (optionally) goodpractice items. Checks may use any of this information, or even add additional information as demonstrated below. The checks$checks list represents the output of check functions, and may not be used in any way within pkgchk_...() functions.

1. The check function

An example is the check for whether a package has a citation, defined in R/check-has-citation.R:

knitr::read_chunk ("../R/check-has-citation.R")
knitr::read_chunk ("../R/check-scrap.R")

This check is particularly simple, because a "CITATION" file must have exactly that name, and must be in the inst sub-directory. This function returns a simple logical of TRUE if the expected "CITATION" file is present, otherwise it returns FALSE. This function, and all functions beginning with the prefix pkgchk_, will be automatically called by the main pkgcheck() function, and the value stored in checks$checks$has_citation. The name of the item within the checks$checks list is the name of the function with the pkgchk_ prefix removed.

A more complicated example is the function to check whether a package contains files which should not be there -- internally called "scrap" files. The check function itself, defined in R/check-scrap.R, checks for the presence of files matching an internally-defined list including files used to locally cache folder thumbnails such as ".DS_Store" or "Thumbs.db". The function returns a character vector of the names of any "scrap" files which can be used by the print method to provide details of files which should be removed. This illustrates the first general principle of these check functions; that,

::: {.alert .alert-info} - Any information needed when summarising or printing the check result should be returned from the main check function. :::

A second important principle is that,

::: {.alert .alert-info} - Check functions should never return NULL, rather should always return an empty vector (such as integer(0)). :::

The following section describes the output_pkgchk_... functions which convert these return values to summary and print output.

2. The output function

All output_pkgchk_...() functions must also accept the single input parameter of checks, in which the checks$checks sub-list will already have been populated by calling all pkgchk_...() functions described in the previous section. The pkgchk_has_citation() function will create an entry of checks$checks$has_citation which contains the binary flag indicating whether or not a "CITATION" file is present. Similarly, the the pkgchk_has_scrap() function will create checks$checks$has_scrap which will contain names of any scrap files present, and a length-zero vector otherwise.

::: {.alert .alert-info} - The pkgchk_ functions must not use any data in checks$checks, as they create this data. - The output_pkgchk_ functions must use the data from checks$checks to construct summary or print output. :::

The output_pkgchk_has_scrap() function looks like this:

The first lines are common to all output_pkgchk_...() functions, and define the generic return object. This object must be a list with the following three items:

check_pass as binary flag indicating whether or not a check was passed;
summary containing text used to generate the summary output; and
print containing information used to generate the print output, itself a list of the following items:
- A msg_pre to display at the start of the print result;
- An object to be printed, such as a vector of values, or a data.frame.
- A msg_post to display at the end of the print result following the object.

summary and print methods may be suppressed by assigning values of "". The above example of pkgcheck_has_citation has print = "", and so no information from this check will appear as output of the print method. The summary field is commented-out in the current version, but left to illustrate here that it has a value that is specified for both TRUE and FALSE values of check_pass, via an ifelse statement. The value is determined by the result of the main pkgchk_has_citation() call, and is converted into a green tick if TRUE, or a red cross if FALSE.

Checks for which print information is desired require a non-empty print item. In this case, both summary and print methods are only triggered if (!out$check_pass) -- so only if the check fails. The print method generates the heading specified in out$print$msg_pre, with any vector-valued objects stored in the corresponding obj list item displayed as formatted lists. A package with "scrap" files, "a" and "b", would thus have out$print$obj <- c ("a", "b"), and when printed would look like this:

cli::cli_alert_danger ("Package contains the following unexpected files:")
cli::cli_ul ()
cli::cli_li (c ("a", "b"))
cli::cli_end ()

This formatting is also translated into corresponding markdown and HTML formatting in the checks_to_markdown() function.

The design of these pkgchk_ and output_pkgchk_ functions aims to make the package readily extensible, and we welcome discussions about developing new checks. The primary criterion for new package-internal checks is that they must be of very general applicability, in that they should check for a condition that almost every package should or should not meet.

The package also has a mechanism to easily incorporate more specific, locally-defined checks, as explored in the following section.

3. Creating new checks

3.1 New Local Checks (for package users)

The main pkgcheck() function has an additional parameter, extra_env which specifies,

Additional environments from which to collate checks. Other package names may be appended using c, as in c(.GlobalEnv, "mypkg").

This allows specific checks to be defined locally, and run by passing the name of the environment in which those checks are defined in this parameter. This section illustrates the process using the bundled "tarball" (that is, .tar.gz file) of one version of the pkgstats package included with that package.

f <- system.file ("extdata", "pkgstats_9.9.tar.gz", package = "pkgstats")
path <- pkgstats::extract_tarball (f)
checks <- pkgcheck (path)
summary (checks)

cli::cli_h1 ("pkgstats 9.9")
message ("")
s <- c (
    "- :heavy_check_mark: Package name is available",
    "- :heavy_multiplication_x: does not have a 'contributing' file.",
    "- :heavy_check_mark: uses 'roxygen2'.",
    "- :heavy_check_mark: 'DESCRIPTION' has a URL field.",
    "- :heavy_check_mark: 'DESCRIPTION' has a BugReports field.",
    "- :heavy_multiplication_x: Package has no HTML vignettes",
    "- :heavy_multiplication_x: These functions do not have examples: [pkgstats_from_archive].",
    "- :heavy_check_mark: Package has continuous integration checks.",
    "- :heavy_multiplication_x: Package coverage failed",
    "- :heavy_multiplication_x: R CMD check found 1 error.",
    "- :heavy_check_mark: R CMD check found no warnings."
)

for (i in s) {
    msg <- strsplit (i, "(mark|\\_x):\\s+") [[1]] [2]
    if (grepl ("heavy_check_mark", i)) {
        cli::cli_alert_success (msg)
    } else {
        cli::cli_alert_danger (msg)
    }
}

message ("")
cli::cli_alert_info ("Current status:")
cli::cli_alert_danger ("This package is not ready to be submitted.")

Let's now presume I have a reputation in the R community for all of my packages starting with "aa", to ensure they are always listed first. This section demonstrates how to implement a check that only passes if the first two letters of the package name are "aa". The first step described above is to define the check itself via a function prefixed with pkgchk_. The easiest approach would be for the pkgcheck_ function to directly check the name, and return a logical flag indicating whether or not the same starts with "aa". The resultant summary and print methods can, however, only use the information provided by the initial pkgchk_ function. That means if we want to print the actual name in the result of either of those functions, to show that it indeed does not form the desired patter, we need to return that information. The check function is then simply:

pkgchk_starts_with_aa <- function (checks) {
    checks$pkg$name
}

We then need to define the output functions:

output_pkgchk_starts_with_aa <- function (checks) {

    out <- list (
        check_pass = grepl ("^aa",
            checks$checks$starts_with_aa,
            ignore.case = TRUE
        ),
        summary = "",
        print = ""
    )

    out$summary <- paste0 (
        "Package name [",
        checks$checks$starts_with_aa,
        "] does ",
        ifelse (out$check_pass,
            "",
            "NOT"
        ),
        " start with 'aa'"
    )

    return (out)
}

If we simply define those function in the global workspace of our current R session, calling pkgcheck() again will automatically detect those checks and include them in our output:

cli::cli_h1 ("pkgstats 9.9")
message ("")
s <- c (
    "- :heavy_check_mark: Package name is available",
    "- :heavy_multiplication_x: does not have a 'contributing' file.",
    "- :heavy_check_mark: uses 'roxygen2'.",
    "- :heavy_check_mark: 'DESCRIPTION' has a URL field.",
    "- :heavy_check_mark: 'DESCRIPTION' has a BugReports field.",
    "- :heavy_multiplication_x: Package has no HTML vignettes",
    "- :heavy_multiplication_x: These functions do not have examples: [pkgstats_from_archive].",
    "- :heavy_check_mark: Package has continuous integration checks.",
    "- :heavy_multiplication_x: Package coverage failed",
    "- :heavy_multiplication_x: Package name [pkgstats] does NOT start with 'aa'",
    "- :heavy_multiplication_x: R CMD check found 1 error.",
    "- :heavy_check_mark: R CMD check found no warnings."
)

for (i in s) {
    msg <- strsplit (i, "(mark|\\_x):\\s+") [[1]] [2]
    if (grepl ("heavy_check_mark", i)) {
        cli::cli_alert_success (msg)
    } else {
        cli::cli_alert_danger (msg)
    }
}

message ("")
cli::cli_alert_info ("Current status:")
cli::cli_alert_danger ("This package is not ready to be submitted.")

Customised personal checks can be incorporated by defining them in a local package, loading that into the workspace, and passing the name of the package to the extra_env parameter.

3.2 New `pkgcheck` Checks (for `pkgcheck` developers)

New checks can be added to this package by creating new files in the /R directory prefixed with pkgchk_, and including the two functions described above (a check and an output function). The check name will then need to be included in the order_checks() function in the R/order-check-summaries.R file, which determines the order of checks in the summary output. Checks which are not defined in this ordering, including any defined via extra_env parameters, appear after all of the standard checks, and prior to the R CMD check results which always appear last. This order may only be modified by editing the list in that function.

The print_check_screen() function will then automatically activate the print method of any new checks. This line should be added even if a new check has no print method (as in the starts_with_aa example above), to provide an explicit record of all internally-defined miscellaneous checks.

3.2a Check types

Some checks are defined so that failure results in a 👀 symbol, rather than a default ❌ symbol. This 👀 symbol indicates that the failures may be worth examining further, and yet do not cause the overall check report to fail. This sub-section describes how to define such checks.

All checks include a binary flag, check_pass, defined in their output_pkgchk_... function, like in the example above, which also define their output conditions. If out$summary is defined for check_pass, then that output will by default be prefixed with ✅, while if out$summary is defined for !check_pass, then that output will by default be prefixed with ❌. Any instances of ❌ will cause the whole check suite to fail, including on pkgcheck-action.

The out value returned from all output_pkgchk_... functions is a list that must include check_pass, summary, and print items, like in the example above. Non-default check types can be defined by an optional extra list-item named check_type, specified as a string with conditions for "(pass)"_"(fail)", where a value of "watch" will replace the default ✅ or ❌ symbols with 👀. For example, a check which should issue ✅ on pass yet 👀 on fail would be specified as,

out$check_type <- "pass_watch"

A check which should only issue 👀 on fail and nothing on success would be specified as,

out$check_type <- "none_watch"

A check which should issue 👀 on success and nothing on failure would be specified as,

out$check_type <- "watch_none"

3.2b Testing new checks

Finally, any new checks also need to be included in tests, with most checks having a corresponding file in the tests/testthat directory. The test suite includes helper functions used to create generic pkgcheck objects which are then modified for testing within individual tests. Most checks start with the following lines:

checks <- make_check_data ()
ci_out <- output_pkgchk_my_fn (checks)

Data from any newly-added checks will automatically appear in the result of make_check_data(), and may be tested directly. Most tests nevertheless only need to test the output of the output_pkgchk_ functions. This is generally done by modifying the checks data obtained in the initial call to make_check_data(), and then passing those modified data to the matching output_pkgchk_ function. For example, in test-check-scrap.R, the default value returned from the output function is first tested, and then the checks value is modified by,

For checks which require either goodpractice or srr (software review roclets) for statistical software, initial check values should be constructed with an alternative helper function, make_check_data_srr(), which accepts an additional parameter, goodpractice, which can be specified as FALSE (default) or TRUE to return full goodpractice data. See examples in test-check-covr.R.

checks$checks$has_scrap <- "scrap"

In this way, all possible forms and modes of each output_pkgchk_ function should be extensively tested. Finally, snapshot results need to be updated to reflect any additional tests, as does the test-list-checks.R file, which tests the total number of internally-defined checks as expect_length (ncks, ..). The number tested there also needs to be incremented by one for each new check.