Where the Checks Come From

Every checktor check rests on a source, and the source sets its severity. This page maps each check to its authority, links the exact section wherever one exists, and gives the severity that authority earns. For the full argument and the exemptions, read the check’s help page, such as ?lab_option_changes.

Read the severity

Severity tells you what kind of authority stands behind a check.

• policy is a rule CRAN enforces. Breaking one can get a package rejected, and a policy finding counts against overall health. Most of these rules are written in the CRAN Repository Policy or Writing R Extensions, the two documents CRAN treats as binding. Some mirror a CRAN incoming check, one of the NOTEs that R CMD check --as-cran raises against a submission. For a few long-standing reviewer requirements that neither manual states, the CRAN Cookbook is the written record. Restoring options(), par() and the working directory is the clearest case. It is among the most common reasons a package is sent back, and neither manual mentions it.
• robustness is a real defect that CRAN will still accept, such as a detectCores() that can return NA. It counts toward overall health too, because the defect is real whether or not CRAN objects to it.
• opinion is a convention that experienced maintainers and reviewers tend to ask for, with nothing enforcing it. checktor reports it but leaves it out of overall health, which by default counts policy and robustness alone.

The Source column in each table below gives the authority. Many conventions are written down in the CRAN Cookbook, the R Contributor guide to the problems that get packages sent back. Where it has a recipe, the column links to it. For checks with no citable source, the column reads “No formal rule”, “No rule” or “Convention only”.

Which checks run

Most checks run every time. Some depend on the machine. When one of those cannot run, checktor() reports it as skipped rather than passed, so good overall health never rests on a check that did not happen. The skipped column of tidy() marks it, and metadata$skipped_checks lists it. Checks are also skipped when their input cannot be read. If R cannot read DESCRIPTION, every check that reads its fields is skipped, and description_file reports the problem as a policy finding. If Authors@R cannot be read, identifier_format is skipped.

Others run only when you call them. Nothing asked for them, so they are not skipped, and metadata$on_request_checks lists them so you can find them.

Every other check runs always, and new checks join the run unless their row in the registry says otherwise. No check goes missing because someone forgot to switch it on.

Code checks

DESCRIPTION checks

Documentation checks

Example, vignette and demo checks

These checks cover the code outside R/ that CRAN reads, where several of the most common rejections land. Maintainers have received each rule below, verbatim, in a reply from CRAN.

General checks

CRAN policy checks

Leave two recipes to people

checktor has a check for every CRAN Cookbook recipe that describes a pattern in your sources. The two recipes without one describe things that no reading of the sources can settle.

Overall Checktime concerns a measurement. The NOTE reads “Overall checktime 20 min > 10 min”, and nothing in your sources tells how long they will take to run. R CMD check reports the figure, and the fix is fewer or smaller examples, vignettes and tests.

Communicating with CRAN is advice on writing to CRAN, including copying [email protected] and explaining yourself in cran-comments.md. Its one checkable part, whether that file exists, is cran_comments_file.

Resolve a disagreement

Severity records where a check’s authority sits, not how strongly checktor feels about it. Several checks are opinion because no authority backs them in either direction. Of those, title_starts_with_article, description_function_quotes and format_names stay callable but are left out of the default run for that reason. cph_role stays out because its own source says most authors need no cph role. cran_comments_file stays out because it asks about a submission workflow rather than the package.

If a finding contradicts your reading of the policy, look at its severity first. Opinion findings open a conversation and require nothing. To turn a check off for one package, list it in Config/checktor/disable in that package’s DESCRIPTION. To turn it off for every package, set options(checktor.disable = ...), for example in .Rprofile. Either way, the check does not run at all.