v3.3.0-1

Deprecations & breaking changes

• The default for pipe_consistency_linter() is changed from "auto" (require one pipe style, either magrittr or native) to "|>" (R native pipe required) to coincide with the same change in the Tidyverse Style Guide (#2707, @MichaelChirico).
• lint() no longer picks up settings automatically in ad hoc invocations like lint("text\n") or lint(text = "str"). You should set parse_settings=TRUE to force settings to be read. Emacs ESS users may need to update to a recent version, e.g. ESS>20251003.
• Arguments allow_cascading_assign=, allow_right_assign=, and allow_pipe_assign= to assignment_linter() are now defunct.
• Six linters marked as deprecated with warning in the previous release are now fully deprecated: consecutive_stopifnot_linter(), extraction_operator_linter(), no_tab_linter(), single_quotes_linter(), unnecessary_nested_if_linter(), and unneeded_concatenation_linter(). They will be removed in the next release.
• As previously announced, the following fully-deprecated items are now removed from the package:
  • source_file= argument to ids_with_token() and with_id().
  • Passing linters by name or as non-"linter"-classed functions.
  • linter= argument of Lint().
  • with_defaults().
  • Linters closed_curly_linter(), open_curly_linter(), paren_brace_linter(), and semicolon_terminator_linter().
• Argument interpret_glue to object_usage_linter() is deprecated in favor of the more general interpret_extensions, in which "glue" is present by default (#1472, @MichaelChirico). See the description below under 'New and improved features'.
• Lint(), and thus all linters, require that the returned object's message attribute is consistently a simple character string (and not, for example, an object of class "glue"; #2740, @MichaelChirico). In general it is good to avoid slower string builders like glue() inside a loop (a linter might be run on every expression in your pakcage). Classed character strings return a warning in this release, which will be upgraded to an error subsequently.

Bug fixes

• Files with encoding inferred from settings read more robustly under lint(parse_settings = TRUE) (#2803, @MichaelChirico). Thanks also to @bastistician for detecting a regression caused by the initial change for users of Emacs (#2847).
• assignment_linter() no longer errors if "%<>%" is an allowed operator (#2850, @AshesITR).
• expect_lint() conforms to {testthat} v3.3.0+ rules for custom expectations, namely that they produce either exactly one success or exactly one failure (#2937, @hadley).

Changes to default linters

• pipe_consistency_linter(), with its new default to enforce the native pipe |>, is now a default linter, since it corresponds directly to a rule in the Tidyverse Style Guide (#2707, @MichaelChirico).

New and improved features

New linters

• all_equal_linter() warns about incorrect use of all.equal() in if clauses or preceded by ! (#2885, @Bisaloo). Such usages should wrap all.equal() with isTRUE(), for example.
• download_file_linter() encourages the use of mode = "wb" (or mode = "ab") when using download.file(), rather than mode = "w" or mode = "a", as the latter can produce broken files in Windows (#2882, @Bisaloo).
• list2df_linter() encourages the use of the list2DF() function, or the data.frame() function when recycling is required, over the slower and less readable do.call(cbind.data.frame, ) alternative (#2834, @Bisaloo).
• coalesce_linter() encourages the use of the infix operator x %||% y, which is equivalent to if (is.null(x)) y else x (#2246, @MichaelChirico). While this has long been used in many tidyverse packages (it was added to {ggplot2} in 2008), it became part of every R installation from R 4.4.0. Thanks also to @emmanuel-ferdman for fixing a false positive before release.

Linter improvements

• brace_linter() has a new argument function_bodies (default "multi_line") which controls when to require function bodies to be wrapped in curly braces, with the options "always", "multi_line" (only require curly braces when a function body spans multiple lines), "not_inline" (only require curly braces when a function body starts on a new line) and "never" (#1807, #2240, @salim-b).
• seq_linter():
  • recommends using seq_along(x) instead of seq_len(length(x)) (#2577, @MichaelChirico).
  • recommends using sequence() instead of unlist(lapply(ints, seq)) (#2618, @Bisaloo).
• undesirable_operator_linter():
  • Lints operators in prefix form, e.g. `%%`(x, 2) (#1910, @MichaelChirico). Disable this by setting call_is_undesirable=FALSE.
  • Accepts unnamed entries, treating them as undesirable operators, e.g. undesirable_operator_linter("%%") (#2536, @MichaelChirico).
• undesirable_function_linter() accepts unnamed entries, treating them as undesirable functions, e.g. undesirable_function_linter("sum") (#2536, @MichaelChirico).
• indentation_linter() handles un-braced for loops correctly (#2564, @MichaelChirico).
• Setting exclusions supports globs like knitr* to exclude files/directories with a pattern (#1554, @MichaelChirico).
• object_name_linter() and object_length_linter() apply to objects assigned with assign() or generics created with setGeneric() (#1665, @MichaelChirico).
• object_usage_linter() gains argument interpret_extensions to govern which false positive-prone common syntaxes should be checked for used objects (#1472, @MichaelChirico). Currently "glue" (renamed from earlier argument interpret_glue) and "rlang" are supported. The latter newly covers usage of the .env pronoun like .env$key, where key was previously missed as being a used variable.
• boolean_arithmetic_linter() finds many more cases like sum(x | y) == 0 where the total of a known-logical vector is compared to 0 (#1580, @MichaelChirico).
• any_duplicated_linter() is extended to recognize some usages from {dplyr} and {data.table} that could be replaced by anyDuplicated(), e.g. n_distinct(col) == n() or uniqueN(col) == .N (#2482, @MichaelChirico).
• fixed_regex_linter() recognizes usage of the new (R 4.5.0) grepv() wrapper of grep(); regex_subset_linter() also recommends grepv() alternatives (#2855, @MichaelChirico).
• object_usage_linter() lints missing packages that may cause false positives (#2872, @AshesITR)
• sprintf_linter() lints sprintf() and gettextf() calls when a constant string is passed to fmt (#2894, @Bisaloo).
• length_test_linter() is extended to check incorrect usage of nrow(), ncol(), NROW(), NCOL() (#2933, @mcol).
• implicit_assignment_linter() gains argument allow_paren_print to disable lints for the use of ( for auto-printing (#2962, @TimTaylor).
• line_length_linter() has a new argument ignore_string_bodies (defaulting to FALSE) which governs whether the contents of multi-line string bodies should be linted (#856, @MichaelChirico). We think the biggest use case for this is writing SQL in R strings, especially in cases where the recommended string width for SQL & R differ.
• package_hooks_linter() now validates .onUnload() hook signatures, requiring exactly one argument starting with 'lib' (#2940, @emmanuel-ferdman).

Lint accuracy fixes: removing false positives

• unnecessary_nesting_linter():
  • Treats function bodies under the shorthand lambda (\()) the same as normal function bodies (#2748, @MichaelChirico).
  • Treats = assignment the same as <- when deciding to combine consecutive if() clauses (#2245, @MichaelChirico).
• string_boundary_linter() omits lints of patterns like \\^ which have an anchor but are not regular expressions (#2636, @MichaelChirico).
• implicit_integer_linter(allow_colon = TRUE) is OK with negative literals, e.g. -1:1 or 1:-1 (#2673, @MichaelChirico).
• missing_argument_linter() allows empty calls like foo() even if there are comments between ( and ) (#2741, @MichaelChirico).
• return_linter() works on functions that happen to use braced expressions in their formals (#2616, @MichaelChirico).
• object_name_linter() and object_length_linter() account for S3 class correctly when the generic is assigned with = (#2507, @MichaelChirico).
• assignment_linter() with operator = "=" does a better job of skipping implicit assignments, which are intended to be governed by implicit_assignment_linter() (#2765, @MichaelChirico).
• implicit_assignment_linter() with allow_scoped=TRUE doesn't lint for if (a <- 1) print(a) (#2913, @MichaelChirico).
• expect_true_false_linter() is pipe-aware, so that 42 |> expect_identical(x, ignore_attr = TRUE) no longer lints (#1520, @MichaelChirico).
• T_and_F_symbol_linter() ignores T and F:
  • When used as symbols in formulas (y ~ T + F), which can represent variables in data not controlled by the author (#2637, @MichaelChirico).
  • If followed by [ or [[ (#2944, @mcol).

Lint accuracy fixes: removing false negatives

• todo_comment_linter() finds comments inside {roxygen2} markup comments (#2447, @MichaelChirico).
• Linters with logic around function declarations consistently include the R 4.0.0 shorthand \() (#2818, continuation of earlier #2190, @MichaelChirico).
  • library_call_linter()
  • terminal_close_linter()
  • unnecessary_lambda_linter()
• More consistency on handling @ extractions to match how similar $ extractions would be linted (#2820, @MichaelChirico).
  • function_left_parentheses_linter()
  • indentation_linter()
  • library_call_linter()
  • missing_argument_linter()
• condition_call_linter() no longer covers cases where the object type in the ellipsis cannot be determined with certainty (#2888, #2890, @Bisaloo). In particular, this fixes the known false positive of custom conditions created via errorCondition() or warningCondition() not being compatible with the call. argument in stop() or warning().

Other improvements

• get_source_expression() captures warnings emitted by the R parser (currently always for mis-specified literal integers like 1.1L) and lint() returns them as lints (#2065, @MichaelChirico).
• expect_lint() has a new argument ignore_order (default FALSE), which, if TRUE, allows the checks= to be provided in arbitary order vs. how lint() produces them (@MichaelChirico).
• New gitlab_output() function to output lints to GitLab format (#2858, @lschneiderbauer).
• New argument include_s4_slots for the xml_find_function_calls() entry in the get_source_expressions() to govern whether calls of the form s4Obj@fun() are included in the result (#2820, @MichaelChirico).
• use_lintr() adds the created .lintr file to the .Rbuildignore if run in a package (#2926, initial work by @MEO265, finalized by @Bisaloo).

Notes

• {lintr} now has an associated paper at the Journal of Open Source Software that you can use to cite the package if you use it in a paper - see citation("lintr") for details.
• expect_lint_free() and other functions that rely on the {testthat} framework now have a consistent error message. (#2585, @F-Noelle).
• unnecessary_nesting_linter() gives a more specific lint message identifying:
  • the unmatched "exit call" that prompts the recommendation to reduce nesting (#2316, @MichaelChirico).
  • the specific if() statement that can be combined with the linted one (#1891, @MichaelChirico).
• The description in ?paste_linter of allow_file_path= has been corrected (#2675, @MichaelChirico). In particular, allow_file_path="never" is the most strict form, allow_file_path="always" is the most lax form.
• comment_token is removed from settings. This was a vestige of the now-defunct support for posting GitHub comments.

New Contributors

• @lschneiderbauer made their first contribution in #2862
• @marberts made their first contribution in #2924
• @emmanuel-ferdman made their first contribution in #2938
• @mcol made their first contribution in #2942
• @TimTaylor made their first contribution in #2962
• @clerousset made their first contribution in #2961

Full Changelog: v3.2.0...v3.3.0-1