Breaking changes

• All linters are now function factories (i.e., functions that return functions) for consistency. Previously, only
  linters with customizable parameters were factories (#245, @fangly, @AshesITR, and @MichaelChirico).

  This means that usage such as lint("file.R", seq_linter) should be updated to lint("file.R", seq_linter()), and
  the following update for custom linters:

  my_custom_linter <- function(source_expression) { ... }
  
  # becomes
  my_custom_linter <- function() Linter(function(source_expression) { ... })

• Exclusions specified in the .lintr file are now relative to the location of that file
  and support excluding entire directories (#158, #438, @AshesITR).

• Removed long-deprecated linters (they've been marked as deprecated since v1.0.1 in 2017):

  • absolute_paths_linter()
  • camel_case_linter()
  • multiple_dots_linter()
  • snake_case_linter()
  • trailing_semicolons_linter()

• Removed return() from all_undesirable_functions because early returns (which often improve
  readability and reduce code complexity) require explicit use of return(). Follow #1100 for
  an upcoming return_linter() to lint unnecessary return() statements (#1146, @AshesITR).
  Note that you can replicate old behavior by supplying return as a custom undesirable function:
  undesirable_function_linter(c(all_undesirable_functions, list(return = NA)))

Deprecations

• Lints are now marked with the name of the linter that caused them instead of the name of their implementation
  function. Deprecated the obsolete linter argument of Lint() (#664, #673, #746, @AshesITR). Downstream custom
  linters should follow suit.
• Renamed semicolon_terminator_linter() to semicolon_linter() for better consistency.
  semicolon_terminator_linter() survives but is marked for deprecation. The new linter also has a new signature,
  taking arguments allow_compound and allow_trailing to replace the old single argument semicolon, again for
  signature consistency with other linters.
• The following linters were subsumed into brace_linter() and are now deprecated; see the item on brace_linter()
  below:
  • closed_curly_linter()
  • open_curly_linter()
  • paren_brace_linter()
• The ... argument for lint(), lint_dir(), and lint_package() has been promoted to an earlier position to
  better match the Tidyverse design principle of
  data->descriptor->details. This change enables passing objects to ... without needing to specify non-required
  arguments, e.g. lint_dir("/path/to/dir", linter()) now works without the need to specify relative_path.
  This affects some code that uses positional arguments (#935, @MichaelChirico).
  • For lint(), ... is now the 3rd argument, where earlier this was cache.
  • For lint_dir() and lint_package(), ... is now the 2nd argument, where earlier this was relative_path.
• Deprecated argument source_file to exported functions with_id() and ids_with_token(). It has been renamed to
  source_expression to better reflect that this argument is typically the output of get_source_expressions().
  For now, the old argument source_file can still be used (with warning). The now-private functional versions of many
  linters also underwent the same renaming (source_file -> source_expression). This has no direct effect on
  packages importing lintr, but is mentioned in case custom linters imitating lintr style had also adopted the
  source_file naming and want to adapt to keep in sync.
• Deprecated with_defaults() in favor of linters_with_defaults(), and add modify_defaults() which is intended to
  be used more generally to modify (i.e., extend, trim, and/or update) a list of defaults. Note that the argument
  corresponding to with_defaults()'s default= is called defaults= (i.e., pluralized) in both of these, and that
  usage like with_defaults(default = NULL, ...) should be converted to linters_with_defaults(defaults = list(), ...)
  (#1029, #1336, #1361, @AshesITR and @MichaelChirico).
• Deprecated the find_line() and find_column() helpers from the item-level expressions returned with
  get_source_expressions(). These helpers were typically associated with regex-based logic for building linters,
  which is rarely needed and prone to false positives; now that lintr almost exclusively uses XPath-based
  logic for linters, these are no longer necessary (#1373, @MichaelChirico).

Other changes to defaults

Updates to default_linters

• New brace_linter() which combines several curly brace related linters, deprecating the following predecessors
  (#1041, @AshesITR):
  • closed_curly_linter(); both now also allow }] in addition to }) and }, as exceptions, i.e., } doesn't
    need to be on its own line if paired with a closing square bracket, a closing parenthesis, or a comma. Also
    improved lint metadata so that source markers land at the closing brace instead of the closing parenthesis to
    improve the experience of fixing the lint (#583, @AshesITR).
  • open_curly_linter(); both also no longer lint unnecessary trailing whitespace (use trailing_whitespace_linter()
    for this) and also allow (, ,, and %>% on preceding lines as exceptions, i.e., { can be alone on a line if
    the previous line is terminated with an opening parenthesis, a comma, or a pipe (%>%) (#487, #1028, @AshesITR).
  • paren_brace_linter(); brace_linter() also lints if/else and repeat with missing whitespace.
  • brace_linter() also newly enforces the following rules surrounding curly braces (originally Google linters, see
    below):
    • Require else to come on the same line as the preceding }, if present (#884, @MichaelChirico).
    • Require functions spanning multiple lines to use curly braces (#987, @MichaelChirico).
    • Require balanced usage of {} in if/else conditions, i.e., if the if branch uses braces,
      then so must the else branch, and vice versa (#983, @MichaelChirico).
• New paren_body_linter() checks that there is a space between a right parenthesis and a body expression (#809,
  @kpagacz).
• Added semicolon_linter() as a default because it enforces a tidyverse style guide rule (#683, @AshesITR).
• assignment_linter() (#915, @MichaelChirico):
  • Right assignments are now linted by default (-> and ->>).
  • New argument allow_cascading_assign (TRUE by default) toggles whether to lint <<- and ->>.
  • New argument allow_right_assign (FALSE by default) toggles whether to lint -> and ->>.
• commented_code_linter(): use the parse tree to find comments, eliminating some false positives (#451, @AshesITR).
• equals_na_linter() (#545, @MichaelChirico):
  • Extended to lint x != NA (before, only == was caught) and NA == x (before, only NA on RHS was caught).
  • Extended to skip usages in comments like is.na(x) # use is.na(x), not x == NA.
• function_left_parentheses_linter(): improved location information (#1266, #1267, @AshesITR).
• infix_spaces_linter():
  • Added argument allow_multiple_spaces (TRUE by default) which toggles
    whether to generate a lint for operators used with multiple spaces, e.g. x + 2.
    The default setting allows extra spacing to be used to increase
    line-to-line alignment (#940, @f-ritter and @MichaelChirico).
  • Extended so that usages like a~b and function(a=1) { ... } are linted (#930, #michaelchirico).
  • Added argument exclude_operators to disable lints on selected infix operators.
    By default, all "low-precedence" operators throw lints; see ?infix_spaces_linter for an enumeration of these.
    (#914, @MichaelChirico).
  • Add an exception for / usage in box::use() declarations (#1087, @klmr).
• line_length_linter(): place the source marker at the margin of the affected line to improve user experience
  during de-linting -- just press Return (#735, @AshesITR).*
• no_tab_linter(): use more reliable matching (e.g., excluding matches found in comments; #441, @russHyde).
• object_length_linter(): correctly detect generics and only count the implementation class towards the length.
  This prevents false positive lints in the case of long generic names, e.g.
  very_very_very_long_generic_name.short_class no longer produces a lint (#871, @AshesITR).
• object_name_linter():
  • Improved generic detection -- in user-defined method my_method.upstream.class,
    upstream.class no longer throws a lint because the generic (my_method)
    properly uses snake_case (#737, @AshesITR).
  • Exclude special R namespace hook functions such as .onLoad() (#500, #614, @AshesITR and @MichaelChirico).
  • Correctly detect imported functions when linting packages (#642, @AshesITR).
  • Correctly detect assignment generics like names<-.class_name (#843, @jonkeane).
  • Added new styles "symbols" and "SNAKE_CASE" (#494, #495, #615, #670, @MichaelChirico and @AshesITR).
    • "symbols" is a new default style which won't lint all-symbol object names. In particular, that means
      operator names like %+% are allowed.
  • No longer lints names used in $ extractions (#582, @AshesITR).
• object_usage_linter():
  • Detect global variables if there are top-level dollar-assignments (#666, @AshesITR).
  • Report usage warnings spanning multiple lines (#507, @AshesITR).
  • Detect usages inside glue::glue() constructs (#942, @AshesITR).
  • Extended to include functions assigned with = instead of <- (#1081, @MichaelChirico).
  • Detect functions exported by packages that are explicitly attached using library() or
    require() calls (#1127, @AshesITR).
  • Improved location information in some cases where the previous regex-based approach didn't work, e.g. unicode
    characters in variable names (#1285, @AshesITR).
  • Correctly detect functions declared within assign() and setMethod() (#1322, @AshesITR).
• spaces_inside_linter(): ignore spaces preceding trailing comments (#636, @MichaelChirico).
• T_and_F_symbol_linter():
  • Added as a default because it enforces a tidyverse style guide rule (#517, @AshesITR).
  • No longer lint occurrences of T and F when used for subsetting, and give a better
    message when used as variable names (#657, @AshesITR).
• trailing_blank_lines_linter():
  • Extended to lint files without a terminal newline (#675, @AshesITR).
  • Also, running lint() on a file without a terminal newline no longer throws a warning().
• trailing_whitespace_linter():
  • Extended to also lint completely blank lines by default (#1044, @AshesITR).
  • Added argument allow_empty_lines (FALSE by default) to toggle this behavior.
  • Improved so that trailing whitespace inside string literals does not trigger a lint (#1045, @AshesITR).
  • Added argument allow_in_strings (TRUE by default) to toggle this behavior.
• undesirable_function_linter():
  • Added new functions to default_undesirable_functions related to debugging (#876, @MichaelChirico):
    • browser()
    • debug()
    • debugcall()
    • debugonce()
    • trace()
    • untrace()
  • No longer lints library() and require() calls attaching a package with an undesired name,
    e.g. library(foo) (#814, @kpagacz and @MichaelChirico).
  • No longer lints undesirable symbols if they are used as names in $ extractions (#1050, @AshesITR).
  • Added more explanation why certain functions might be undesirable and what alternatives to use;
    ditto for undesirable_operator_linter() (#1133, #1146, #1159, @AshesITR).

Other noteworthy changes

• cyclocomp_linter(): set the default complexity_limit to 15. This brings the default into sync with what
  is enforced via default_linters (#693, @AshesITR).
• lint_package() now lints files in the demo directory by default (#703, @dmurdoch).
• Moved the default lintr cache directory from ~/.R/lintr_cache (which was a violation of
  CRAN policy) to R_user_dir("lintr", "cache"). Note that 3.0.0 is a major version update and invalidates
  the old cache anyway, so it can be safely deleted (#1062, @AshesITR).

New and improved features

New linters

• backport_linter() for detecting mismatched R version dependencies (#506, #1316, #1318, #1319, @MichaelChirico and
  @AshesITR).
• duplicate_argument_linter() similarly checks that there are no duplicate arguments supplied to function calls (#850,
  @renkun-ken).
• missing_argument_linter() to check for empty (missing) arguments in function calls (#563, #1152, @renkun-ken and
  @AshesITR).
• missing_package_linter() to check if packages in calls to library() and friends
  are missing (#536, #1037, @renkun-ken and @MichaelChirico).
• namespace_linter() to check for common mistakes in pkg::symbol usages (#548, @renkun-ken).
• package_hooks_linter() to run a series of checks also done by R CMD check on the .onLoad(), .onAttach(),
  .Last.lib() and .onDetach() hooks (#882, @MichaelChirico).
• pipe_call_linter() to enforce that all steps of magrittr pipelines use explicit calls instead of symbols,
  e.g. x %>% mean() instead of x %>% mean (#801, @MichaelChirico).
• sprintf_linter() to check for common mistakes in sprintf() usage (#544, #624, @renkun-ken and @AshesITR).
• unused_import_linter() to detect unnecessary library() calls in R scripts (#239, @jimhester, @AshesITR).

Google linters

Google is a heavy user of lintr internally, and has developed a large set of linters improving code consistency
and correcting common R usage mistakes. This release includes many of these linters that are
of general interest to the broader R community. More will be included in future releases. See, e.g.
#884, #979, #998, #1011, #1016, #1036, #1051, #1066, and #1067; special thanks to @MichaelChirico and @michaelquinn32.

• any_duplicated_linter() Require usage of anyDuplicated(x) > 0L over any(duplicated(x)) and similar.
• any_is_na_linter() Require usage of anyNA(x) over any(is.na(x)).
• class_equals_linter() Prevent comparing class(x) with ==, !=, or %in%, where inherits() is typically
  preferred.
• condition_message_linter() Prevent condition messages from being constructed like stop(paste(...))
  (where just stop(...) is preferable).
• conjunct_test_linter() Require usage of expect_true(x); expect_true(y) over expect_true(x && y) and similar.
• consecutive_stopifnot_linter() Require consecutive calls to stopifnot() to be unified into one.
• expect_comparison_linter() Require usage of expect_gt(x, y) over expect_true(x > y) and similar.
• expect_identical_linter() Require usage of expect_identical() by default, and expect_equal() only by exception.
• expect_length_linter() Require usage of expect_length(x, n) over expect_equal(length(x), n) and similar.
• expect_named_linter() Require usage of expect_named(x, n) over expect_equal(names(x), n) and similar.
• expect_not_linter() Require usage of expect_false(x) over expect_true(!x), and vice versa.
• expect_null_linter() Require usage of expect_null(x) over expect_equal(x, NULL) and similar.
• expect_s3_class_linter() Require usage of expect_s3_class(x, k) over expect_equal(class(x), k) and similar.
• expect_s4_class_linter() Require usage of expect_s4_class(x, k) over expect_true(methods::is(x, k)).
• expect_true_false_linter() Require usage of expect_true(x) over expect_equal(x, TRUE) and similar.
• expect_type_linter() Require usage of expect_type(x, t) over expect_equal(typeof(x), t) and similar.
• fixed_regex_linter() Require fixed = TRUE or stringr::fixed() for regular expressions that can be
  expressed statically, e.g. strsplit(x, "[.]") can be strsplit(x, ".", fixed = TRUE).
  • Added parameter allow_grepl (default FALSE) to toggle whether grepl() usages should be linted.
    These might be treated separately because grepl("^x", NA) is FALSE; the startsWith() equivalent to
    get FALSE for missing input is clunkier, but more explicit: !is.na(x) & startsWith(x, string) (#1376, @MichaelChirico).
• ifelse_censor_linter() Require usage of pmax() / pmin() where appropriate, e.g. ifelse(x > y, x, y) is
  pmax(x, y).
• inner_combine_linter() Require inputs to known-vectorized functions to be combined first rather than later,
  e.g. as.Date(c(x, y)) over c(as.Date(x), as.Date(y)).
• literal_coercion_linter() Require using correctly-typed literals instead of direct coercion, e.g. 1L instead of
  as.numeric(1).
• nested_ifelse_linter() Prevent nested calls to ifelse() like ifelse(A, x, ifelse(B, y, z)), and similar.
• numeric_leading_zero_linter() Require a leading 0 in fractional numeric constants, e.g. 0.1 instead of .1.
• outer_negation_linter() Require usage of !any(x) over all(!x) and !all(x) over any(!x).
• paste_linter() lint for common mis-use of paste() and paste0():
  • paste0() encouraged instead of paste(sep = "").
  • toString() or glue::glue_collapse() encouraged instead of paste(x, collapse = ", ").
  • Lint sep= passed to paste0() -- typically a mistake.
• redundant_ifelse_linter() Prevent usage like ifelse(A & B, TRUE, FALSE) or ifelse(C, 0, 1)
  (the latter is as.numeric(!C)).
• regex_subset_linter() Require usage of grep(ptn, x, value = TRUE) over x[grep(ptn, x)] and similar.
• string_boundary_linter() Require usage of startsWith(x, ptn) over grepl("^ptn", x) or substr(x, 1, 3) == ptn
  and similar.
• strings_as_factors_linter() Check for code designed to work before and after the stringsAsFactors = FALSE default
  change in R 4.0 by examining code for data.frame() usages susceptible to assumptions about the default value
  of stringsAsFactors=.
• system_file_linter() Prevent usage like file.path(system.file("A", package = "pkg"), "B") where simply
  system.file("A", "B", package = "pkg") is more concise and readable.
• unreachable_code_linter() Prevent code after return() and stop() statements that will never be reached
  (extended for #1051 thanks to early user testing, thanks @bersbersbers!).
• vector_logic_linter() Require use of scalar logical operators (&& and ||) inside if() conditions and similar.
• yoda_test_linter() Require usage of expect_identical(x, 1L) over expect_equal(1L, x) and similar.

Other features and improvements

• Documentation: Reorganize linter documentation into new tag-based Rd pages (#888, #1015, @AshesITR).
  • Each linter has its own help page.
  • ?linters also links to tag help pages, collecting linters with a similar goal.
  • Each linter can have multiple tags.
  • available_linters(): new function to list available linters and their tags.
    This feature is extensible by package authors providing add-on linters for {lintr}.
  • available_tags(): new function to list available tags.
  • linters_with_tags(): new function to help build a list of linters using tags.
• Encodings: lintr now supports non-system character Encodings. The correct the correct encoding
  is auto-detected from .Rproj or DESCRIPTION files in your project.
  Override the default in the encoding setting of lintr (#752, #782, @AshesITR).
• Jenkins CI: Support for writing comments to GitHub repo when running in Jenkins CI (#488, @fdlk).
• Performance: Optimized performance-critical functions in lintr, such as get_source_expressions() resulting in
  about 2x speedup in our test suite and even more for complex files (#1169, #1197, #1200, #1201, #1214, @MichaelChirico
  and @AshesITR). Average lint_package() execution time is down about 30% and the median package sees about 40%
  improvement.
• Raw strings: Several linters tightened internal logic to allow for raw strings like R"( a\string )"
  (#1034, #1285, @MichaelChirico and @AshesITR).
• Selective exclusion syntax: New syntax to exclude only selected linters from certain lines or passages.
  Use # nolint: linter_name, linter2_name. or # nolint start: linter_name, linter2_name.
  in source files or named lists of line numbers in .lintr. Note the terminal . is required.
  Also allows for partial matching as long as the supplied prefix is unique, e.g.
  # nolint: infix_spaces. works to exclude infix_spaces_linter (#605, #872, @AshesITR).
  • Added the linter name to lintrs output to facilitate discovery of the correct name (#1357, @AshesITR).
• Improved S3 generic detection for non-standard S3 generics where UseMethod() is called after several
  preceding expressions (#846, @jonkeane).
• extraction_operator_linter(): no longer lint x[NULL] (#1273, @AshesITR).
• is_lint_level(): new exported helper for readably explaining which type of expression is required for a custom
  linter. Some linters are written to require the full file's parse tree (for example, single_quotes_linter()).
  Others only need single expressions, which is more cache-friendly (most linters are written this way to leverage
  caching) (#921, @MichaelChirico).
• lint_dir() excludes the renv and packrat directories by default (#697, @AshesITR).
• lint(): new optional argument text for supplying a line or lines directly, e.g. if the file is already
  in memory or linting is being done ad hoc (#503, @renkun-ken).
• seq_linter(): improve lint message to be clearer about the reason for linting (#522, @MichaelChirico).
• unneeded_concatenation_linter():
  • Correctly considers arguments in pipelines (%>% or |>; #573, #1270, @michaelquinn32 and @AshesITR).
  • New argument allow_single_expression, default TRUE, toggling whether c(x) should be linted, i.e.,
    a call to c() with only one entry which is not a constant. In some such cases, c() can simply be dropped,
    e.g. c(a:b); in others, the parentheses are still needed, e.g. -c(a:b) should be -(a:b);
    and in still others, c() is used for the side-effect of stripping attributes, e.g.
    c(factor(letters)) or c(matrix(1:10, 5, 2)). In this last case, c() can (and should) in most cases
    be replaced by as.vector() or as.integer() for readability. In fact, we suspect it is always
    preferable to do so, and may change the default to allow_single_expression = FALSE in the future. Please
    report your use case if as.vector() does not suit your needs (#1344, @MichaelChirico).
• use_lintr(): new exported helper for creating a minimal .lintr configuration (#902, @AshesITR).
• xml_nodes_to_lints(): new exported helper for converting xml_node objects obtained using linter logic
  expressed in XPath into Lint objects (#1124, #1216, #1234, @MichaelChirico and @AshesITR).

Bug fixes

• RStudio: Source markers are cleared when there are no lints (#520, @AshesITR).
• Error message for mismatched starts and ends of exclusion ranges is now more helpful.
  (#571, #860, @AshesITR and @danielinteractive).
• Improved location information for R parse errors (#894, #892, @renkun-ken and @AshesITR).
• get_source_expressions():
  • Fix possible error on invalid XML produced by xmlparsedata::xml_parse_data() (#559, @renkun-ken).
  • Fix handling zero-length variable name error (#566, @renkun-ken).
  • Malformed Rmd files now cause a lint instead of an error (#571, @AshesITR).
  • No longer fails if getParseData() returns a truncated (invalid) Unicode character as parsed text (#815,
    @leogama).
  • Fixes the text value for STR_CONST nodes involving 1- or 2-width octal escapes
    (e.g. "\1") to account for an R parser bug (https://bugs.r-project.org/show_bug.cgi?id=18323; #1056,
    @MichaelChirico).
  • Handle Rmd inputs containing unevaluated code blocks with named format specifiers (#472, @russHyde).
• line_length_linter(): fix a bug causing duplicate lints for lines containing multiple expressions (#681, @AshesITR).
• lint_package():
  • Warns and returns NULL if no package is found (instead of giving a peculiar error message; #776,
    @MichaelChirico).
  • Stricter about what is considered to be a package -- folders named DESCRIPTION are ignored (#702,
    @MichaelChirico).
• linters_with_defaults() (formerly with_defaults()):
  • No longer duplicates the lintr_function class when it is already present (#511, @AshesITR).
  • Warns if a named argument is NULL but its name is not in defaults (#1049, @AshesITR).
• linters_with_defaults() handles automatic naming of very long arguments correctly (#774, @MichaelChirico).
• save_cache() will now recursively create the cache directory; this avoids errors that could arise if any parent
  directories do not exist (#60, @dankessler).
• spaces_left_parentheses_linter(): fix a bug causing warnings like "In parent == parent[before_operator_idx]
  longer object length is not a multiple of shorter object length" in nested expressions (#654, @AshesITR).

Internals

• Added a new, more restrictive test workflow - test-package - that fails on warnings emitted by tests
  (#1263, #1272, @AshesITR).
• Added a secondary, more restrictive lint workflow - lint-changed-files - for newly written / modified code
  (#641, @dragosmg).
• Several optional Imported packages have become Suggested dependencies: httr, testthat, and rstudioapi.
  This should allow snappier CI builds for usages not relying on some more "peripheral" features of the package.
• Special thanks to @bersbersbers for early testing on the 3.0.0 changes.
• Switched CI from Travis to GitHub Actions, using the full tidyverse recommended R CMD check. Code coverage and
  linting are implemented using separate GitHub Actions workflows (#572, @dragosmg).
• Updated R CMD GitHub Actions workflow to check for R 3.6 on Ubuntu, instead of R 3.3, and for R 4.0 on Windows,
  instead of R 3.6 (#803, @ dragosmg).
• lintr now uses the 3rd edition of testthat (@MichaelChirico, @AshesITR, #910, #967).