lintr Files

Breaking changes

• infix_spaces_linter() distinguishes <-, :=, <<- and ->, ->>, i.e. infix_spaces_linter(exclude_operators = "->") will no longer exclude ->> (#2115, @MichaelChirico). This change is breaking for users relying on manually-supplied exclude_operators containing "<-" to also exclude := and <<-. The fix is to manually supply ":=" and "<<-" as well. We don't expect this change to affect many users, the fix is simple, and the new behavior is much more transparent, so we are including this breakage in a minor release.
• Removed find_line() and find_column() entries from get_source_expressions() expression-level objects. These have been marked deprecated since version 3.0.0. No users were found on GitHub.
• There is experimental support for writing config in plain R scripts (as opposed to DCF files; [#1210], @MichaelChirico). The script is run in a new environment and variables matching settings (?default_settings) are copied over. In particular, this removes the need to write R code in a DCF-friendly way, and allows normal R syntax highlighting in the saved file. We may eventually deprecate the DCF approach in favor of this one; user feedback is welcome on strong preferences for either approach, or for a different approach like YAML. Generally you should be able to convert your existing .lintr file to an equivalent R config by replacing the : key-value separators with assignments (<-). By default, such a config is searched for in a file named '.lintr.R'. This is a mildly breaking change if you happened to be keeping a file '.lintr.R' around since that file is given precedence over '.lintr'.
• We also validate config files up-front make it clearer when invalid configs are present (#2195, @MichaelChirico). There is a warning for "invalid" settings, i.e., settings not part of ?default_settings. We think this is more likely to affect users declaring settings in R, since any variable defined in the config that's not a setting must be removed to make it clearer which variables are settings vs. ancillary.

Bug fixes

• sprintf_linter() doesn't error in cases where whitespace in ... arguments is significant, e.g. sprintf("%s", if (A) "" else y), which won't parse if whitespace is removed (#2131, @MichaelChirico).

Changes to default linters

• assignment_linter() lints the {magrittr} assignment pipe %<>% (#2008, @MichaelChirico). This can be deactivated by setting the new argument allow_pipe_assign to TRUE.
• object_usage_linter():
• assumes glue() is glue::glue() when interpret_glue=TRUE (#2032, @MichaelChirico).
• finds function usages, including infix usage, inside glue() calls to avoid false positives for "unused objects" (#2029 and [#2069], @MichaelChirico).
• object_name_linter() no longer attempts to lint strings in function calls on the LHS of assignments (#1466, @MichaelChirico).
• infix_spaces_linter() allows finer control for linting = in different scenarios using parse tags EQ_ASSIGN, EQ_SUB, and EQ_FORMALS (#1977, @MichaelChirico).
• equals_na_linter() checks for x %in% NA, which is a more convoluted form of is.na(x) (#2088, @MichaelChirico).

New and improved features

• New exclusion sentinel # nolint next to signify the next line should skip linting (#1791, @MichaelChirico). The usual rules apply for excluding specific linters, e.g. # nolint next: assignment_linter.. The exact string used to match a subsequent-line exclusion is controlled by the exclude_next config entry or R option "lintr.exclude_next".
• New xp_call_name() helper to facilitate writing custom linters (#2023, @MichaelChirico). This helper converts a matched XPath to the R function to which it corresponds. This is useful for including the "offending" function in the lint's message.
• New make_linter_from_xpath() to facilitate making simple linters directly from a single XPath (#2064, @MichaelChirico). This is especially helpful for making on-the-fly/exploratory linters, but also extends to any case where the linter can be fully defined from a static lint message and single XPath.
• Toggle lint progress indicators with argument show_progress to lint_dir() and lint_package() (#972, @MichaelChirico). The default is still to show progress in interactive() sessions. Progress is also now shown with a "proper" progress bar (utils::txtProgressBar()), which in particular solves the issue of progress . spilling well past the width of the screen in large directories.
• lint(), lint_dir(), and lint_package() fail more gracefully when the user mis-spells an argument name (#2134, @MichaelChirico).
• Quarto files (.qmd) are included by lint_dir() by default (#2150, @dave-lovell).

New linters

• library_call_linter() can detect if all library/require calls are not at the top of your script (#2027, [#2043], [#2163], and [#2170], @nicholas-masel and @MichaelChirico).
• keyword_quote_linter() for finding unnecessary or discouraged quoting of symbols in assignment, function arguments, or extraction (part of [#884], @MichaelChirico). Quoting is unnecessary when the target is a valid R name, e.g. c("a" = 1) can be c(a = 1). The same goes to assignment ("a" <- 1) and extraction (x$"a"). Where quoting is necessary, the linter encourages doing so with backticks (e.g. x$`a b` instead of x$"a b").
• length_levels_linter() for using the specific function nlevels() instead of checking length(levels(x)) (part of [#884], @MichaelChirico).
• scalar_in_linter() for discouraging %in% when the right-hand side is a scalar, e.g. x %in% 1 (part of [#884], @MichaelChirico).
• if_not_else_linter() for encouraging if statements to be structured as if (A) x else y instead of if (!A) y else x (part of [#884], @MichaelChirico).
• repeat_linter() for encouraging repeat for infinite loops instead of while (TRUE) (#2106, @MEO265).
• length_test_linter() detects the common mistake length(x == 0) which is meant to be length(x) == 0 (#1991, @MichaelChirico).

Extensions to existing linters

• fixed_regex_linter() gains an option allow_unescaped (default FALSE) to toggle linting regexes not requiring any escapes or character classes (#1689, @MichaelChirico). Thus fixed_regex_linter(allow_unescaped = TRUE) would lint on grepl("[$]", x) but not on grepl("a", x) since the latter does not use any regex special characters.
• line_length_linter() helpfully includes the line length in the lint message (#2057, @MichaelChirico).
• conjunct_test_linter() also lints usage like dplyr::filter(x, A & B) in favor of using dplyr::filter(x, A, B) (part of [#884]; [#2110] and [#2078], @salim-b and @MichaelChirico). Option allow_filter toggles when this applies. allow_filter = "always" drops such lints entirely, while "not_dplyr" only lints calls explicitly qualified as dplyr::filter(). The default, "never", assumes all unqualified calls to filter() are dplyr::filter().
• sort_linter() checks for code like x == sort(x) which is better served by using the function is.unsorted() (part of [#884], @MichaelChirico).
• paste_linter() gains detection for file paths that are better constructed with file.path(), e.g. paste0(dir, "/", file) would be better as file.path(dir, file) (part of [#884], [#2082], @MichaelChirico). What exactly gets linted here can be fine-tuned with the allow_file_path option ("double_slash" by default, with alternatives "never" and "always"). When "always", these rules are ignored. When "double_slash", paths appearing to construct a URL that have consecutive forward slashes (/) are skipped. When "never", even URLs should be constructed with file.path().
• seq_linter() recommends rev() in the lint message for lints like nrow(x):1 (#1542, @MichaelChirico).
• function_argument_linter() detects usage of missing() for the linted argument (#1546, @MichaelChirico). The simplest fix for function_argument_linter() lints is typically to set that argument to NULL by default, in which case it's usually preferable to update function logic checking missing() to check is.null() instead.
• commas_linter() gains an option allow_trailing (default FALSE) to allow trailing commas while indexing. (#2104, @MEO265)
• unreachable_code_linter()
• checks for code inside if (FALSE) and other conditional loops with deterministically false conditions (#1428, @ME0265).
• checks for unreachable code inside if, else, for, while, and repeat blocks, including combinations with break and next statements. (#2105, @ME0265).
• implicit_assignment_linter() gains an argument allow_lazy (default FALSE) that allows optionally skipping lazy assignments like A && (B <- foo(A)) (#2016, @MichaelChirico).
• unused_import_linter() gains an argument interpret_glue (default TRUE) paralleling that in object_usage_linter() to toggle whether glue::glue() expressions should be inspected for exported object usage (#2042, @MichaelChirico).
• default_undesirable_functions is updated to also include Sys.unsetenv() and structure() (#2192 and [#2228], @IndrajeetPatil and @MichaelChirico).
• Linters with logic around the magrittr pipe %>% consistently apply it to the other pipes %!>%, %T>%, %<>% (and possibly %$%) where appropriate (#2008, @MichaelChirico).
• brace_linter()
• pipe_call_linter()
• pipe_continuation_linter()
• unnecessary_concatenation_linter()
• unnecessary_placeholder_linter()
• Linters with logic around function declarations consistently include the R 4.0.0 shorthand \() (#2190, @MichaelChirico).
• brace_linter()
• function_left_parentheses_linter()
• indentation_linter()
• object_length_linter()
• object_name_linter()
• package_hooks_linter()
• paren_body_linter()
• unnecessary_lambda_linter()
• unreachable_code_linter()

Lint accuracy fixes: removing false positives

• fixed_regex_linter()
• Is pipe-aware, in particular removing false positives around piping into {stringr} functions like x |> str_replace(fixed("a"), "b") (#1811, @MichaelChirico).
• Ignores non-string inputs to pattern= as a keyword argument (#2159, @MichaelChirico).
• Several linters avoiding false positives in $ extractions get the same exceptions for @ extractions, e.g. S4@T will no longer throw a T_and_F_symbol_linter() hit (#2039, @MichaelChirico).
• T_and_F_symbol_linter()
• for_loop_index_linter()
• literal_coercion_linter()
• object_name_linter()
• undesirable_function_linter()
• unreachable_code_linter()
• yoda_test_linter()
• sprintf_linter() is pipe-aware, so that x %>% sprintf(fmt = "%s") no longer lints (#1943, @MichaelChirico).
• condition_message_linter() ignores usages of extracted calls like env$stop(paste(a, b)) (#1455, @MichaelChirico).
• inner_combine_linter() no longer throws on length-1 calls to c() like c(exp(2)) or c(log(3)) (#2017, @MichaelChirico). Such usage is discouraged by unnecessary_concatenation_linter(), but inner_combine_linter() per se does not apply.
• sort_linter() only lints on order() of a single vector, excluding e.g. x[order(x, y)] and x[order(y, x)] (#2156, @MichaelChirico).
• redundant_ifelse_linter() is aware of dplyr::if_else()'s missing= argument, so that if_else(A, TRUE, FALSE, missing = FALSE) doesn't lint, but if_else(A, TRUE, FALSE, NA) does (#1941, @MichaelChirico). Note that dplyr::coalesce() or tidyr::replace_na() may still be preferable.

Lint accuracy fixes: removing false negatives

• unreachable_code_linter() finds unreachable code even in the presence of a comment or semicolon after return() or stop() (#2127, @MEO265).
• implicit_assignment_linter()
• finds assignments in call arguments besides the first one (#2136, @MichaelChirico).
• finds assignments in parenthetical expressions like if (A && (B <- foo(A))) { } (#2138, @MichaelChirico).
• unnecessary_lambda_linter() checks for cases using explicit returns, e.g. lapply(x, \(xi) return(sum(xi))) (#1567, @MichaelChirico).
• thanks to @Bisaloo and @strengejacke for detecting a regression in the original fix (#2231, [#2247]).