Skip to content

Commit

Permalink
reorganize NEWS
Browse files Browse the repository at this point in the history
  • Loading branch information
@MichaelChirico
MichaelChirico committed Oct 3, 2023
1 parent 9c49c6e commit f5b0fe7
Showing 1 changed file with 61 additions and 51 deletions.
112 changes: 61 additions & 51 deletions NEWS.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,11 +4,22 @@

* `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'.

## 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"`.
Expand All @@ -17,17 +28,37 @@
* 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).
* 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'.

### 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).
* `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()` (#2192, @IndrajeetPatil).

### 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).
+ 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("[$]", x)` but not on `grepl("a", x)` since the latter does not use any regex special characters.
+ Ignores non-string inputs to `pattern=` as a keyword argument (#2159, @MichaelChirico).
* Linters with logic around the magrittr pipe `%>%` consistently apply it to the other pipes `%!>%`, `%T>%`, `%<>%` (and possibly `%$%`) where appropriate (#2008, @MichaelChirico).
+ `brace_linter()`
+ `pipe_call_linter()`
+ `pipe_continuation_linter()`
+ `unnecessary_concatenation_linter()`
+ `unnecessary_placeholder_linter()`
* Several linters avoiding false positives in `$` extractions get the same exceptions for `@` extractions, e.g. `S4@T` will no longer throw a `T_and_F_symbol_linter()` hit (#2039, @MichaelChirico).
+ `T_and_F_symbol_linter()`
+ `for_loop_index_linter()`
Expand All @@ -36,6 +67,28 @@
+ `undesirable_function_linter()`
+ `unreachable_code_linter()`
+ `yoda_test_linter()`
* `sprintf_linter()` is pipe-aware, so that `x %>% sprintf(fmt = "%s")` no longer lints (#1943, @MichaelChirico).
* `condition_message_linter()` ignores usages of extracted calls like `env$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).

### Lint accuracy fixes: removing both false negatives and false positives

* 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()`
Expand All @@ -46,49 +99,6 @@
+ `paren_body_linter()`
+ `unnecessary_lambda_linter()`
+ `unreachable_code_linter()`
* `sprintf_linter()` is pipe-aware, so that `x %>% sprintf(fmt = "%s")` no longer lints (#1943, @MichaelChirico).
* `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.
* `equals_na_linter()` checks for `x %in% NA`, which is a more convoluted form of `is.na(x)` (#2088, @MichaelChirico).
* `commas_linter()` gains an option `allow_trailing` (default `FALSE`) to allow trailing commas while indexing. (#2104, @MEO265)
* `unreachable_code_linter()`
+ finds unreachable code even in the presence of a comment or semicolon after `return()` or `stop()` (#2127, @MEO265).
+ checks for code inside `if (FALSE)` and other conditional loops with deterministically false conditions (#1428, @ME0265).
* `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).
+ gains an argument `allow_lazy` (default `FALSE`) that allows optionally skipping lazy assignments like `A && (B <- foo(A))` (#2016, @MichaelChirico).
* `implicit_assignment_linter()` finds assignments in call arguments besides the first one (#2136, @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.
* `condition_message_linter()` ignores usages of extracted calls like `env$stop(paste(a, b))` (#1455, @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).
* `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.
* `unnecessary_lambda_linter()` checks for cases using explicit returns, e.g. `lapply(x, \(xi) return(sum(xi)))` (#1567, @MichaelChirico).
* `default_undesirable_functions` is updated to also include `Sys.unsetenv()` (#2192, @IndrajeetPatil).

### 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).

## Changes to defaults

* `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).

# lintr 3.1.0

Expand Down

0 comments on commit f5b0fe7

Please sign in to comment.