This is the first CRAN release.
add_new_rule() now errors if the file already exists
(#87).
In all functions that create or modify files (fix(),
setup_flir(), etc.), the argument path has to
be specified. Linting functions keep using the default path
".". This is due to the CRAN policy that a package cannot
modify the user file system (#79).
Most fix_*() functions have a new argument
interactive (FALSE by default). When set to
TRUE, it opens a Shiny app in the Viewer so that the user
can review fixes that would be applied to a file. It is therefore
possible to accept or skip fixes on a file-by-file basis (#76).
In R packages, it is now possible to create a list of rules that will be available to the users of the package. This can be useful to provide automatic fixes for deprecated or superseded functions, for example. More information is available in the vignette “Sharing rules across packages”. Thanks to @maelle and @Bisaloo for the suggestion and early feedback on the implementation (#78, #84).
add_new_rule() now accepts several
names at once (#88).
unreachable_code is deactivated by default. It can
still be activated with the argument linters or in
flir/config.yml after running setup_flir()
(#75).
T_and_F_symbol do not detect anymore the use of
F and T when those are used in the
: operation (#81).
semicolon_linter is no longer used. It is a linter
related to code formatting only and therefore isn’t in the scope of
flir (and the current implementation was buggy). It is also
handled by the Air formatter (#93).
lint_*() and fix_*() now work with
relative paths to YAML files, for example
lint(linters = "my_rule.yml") (#92).testthat (such as
expect_named) are ignored if the files that are parsed
belong to a package that doesn’t have a tests/testthat
folder (for instance if you use tinytest instead).
(#74)Changes were still directly applied to files that are unstaged in Git while there should have been a warning. This is now fixed.
Properly skip changes in the user picks “No” in the menu about modifying unstaged files in Git.
setup_flir() no longer imports all built-in rules in
the flir/rules folder. Those are now directly read from the
folder where flir is installed. Consequently,
update_flir() has been removed as it has no purpose anymore
(#66).add_new_rule() to create the template for
a new rule in flir/rules/custom (#67).sample_int-4.BREAKING: flint is renamed
flir to avoid namespace conflict with the recent flint
package on CRAN. Thanks to Mikael Jagan for the warning (#63).
Consequences:
setup_flint() is renamed setup_flir() and
creates a folder named flir instead of
flint;setup_flint_gha() is renamed
setup_flir_gha() and creates a YAML file named
flir.yml instead of flint.yml;update_flint() is renamed
update_flir();flir-ignore
instead of flint-ignore;FLINT_ERROR_ON_LINT is renamed
FLIR_ERROR_ON_LINT;flint was used in a package,
.Rbuildignore must be updated to ignore the folder
flir instead of flint.FLINT_ERROR_ON_LINT to
determine whether flint should error if some lints were
found.No longer error about unavailable sourceMarkers when running
lint() in Positron.
Better detection of flint/config.yml when using
running flint on a package or a directory.
The cache used after setup_flir() is now invalidated
if the rules used change.
Do not print “empty data.table(…)” when no lints are detected.
New linter stopifnot_all_linter() to detect calls to
stopifnot(all(...)) since the all() is
unnecessary. This has an automatic fix available.
New linter list_comparison_linter() to detect a
comparison with a list, e.g. lapply(x, sum) > 10. No
automatic fix available.
Line breaks are removed from multi-line messages reported by
lint* functions.
matrix_apply_linter now detects when 1L
and 2L are used in the MARGIN
argument.
any_is_na_linter now reports cases like
NA %in% x, and can fix them to be anyNA(x)
instead.
library_call_linter no longer reports cases where
library() calls are wrapped in
suppressPackageStartupMessages().
Nested fixes no longer overlap. The fix*() functions
now run several times on the files containing nested fixes until there
are no more fixes to apply. This can be deactivated to run only once per
file by adding rerun = FALSE (#61).
any_is_na_linter wrongly reported cases like
any(is.na(x), y). Those are no longer reported.
No longer lint and fix
expect_equal(length(x), length(y)), which is more readable
than expect_length(x, length(y)).
No longer lint and fix
expect_equal(names(x), names(y)), which is more readable
than expect_named(x, names(y)).
sample(n, m) is now reported and can be rewritten as
sample.int(n, m) when n is a literal
integer.Rule names have been harmonized to use a dash instead of
underscore, e.g. any_duplicated-1 instead of
any_duplicated_1.
Replacement of redundant_ifelse_linter of the form
ifelse(cond, FALSE, TRUE) now works (#57).
absolute_path_linter was deactivated in 0.0.5 but
was still reported. It is now properly ignored.
Code like expect_equal(typeof(x), 'class') was
modified twice by expect_identical_linter and
expect_type_linter, which lead to a wrong rewrite. It is
now replaced by expect_type(x, 'class').
fix() and lint() now work correctly
when several paths are passed.
fix_package() and lint_package() used
all R files present from the root path, even those in folders that are
not typical of an R package.
fix_dir() and fix_package() now have
the arguments force and verbose, like
fix().
rep_len_linter(),
sample_int_linter() and
which_grepl_linter().fix() and its variants would
change some unstaged files.update_flint() now updates all rules and doesn’t only
add new rules anymore.condition_message_linter() and
expect_identical_linter().setup_flir() now puts built-in rules in
flint/rules/builtin.expect_comparison_linter and
package_hooks_linter.verbose to lint_package() and
lint_dir().seq_linter where
seq_len(x) is faster than seq(1, x).exclude_path.flint
(#22).function_return_linter and
todo_comment_linter.library_call_linter.overwrite to
setup_flir_gha().redundant_equals_linter.matrix_apply_linter.absolute_path_linter in default use as
there are too many false positives.unnecessary_nesting_linter.lint() and fix() showing
the number of files checked, lints found and/or fixed.T and
F in formulas.lengths_linter if
it was already present in code.lintr documentation in the manual
pages.for_loop_index,missing_argument.fix() has a new argument force
(FALSE by default). This is useful if Git was not detected,
fix() would modify several files, and it is run in a
non-interactive context. In this situation, set
force = TRUE to apply the fixes anyway.cli messages informing how many files are checked,
and how many contain lints (for lint_* functions) or were
modified (for fix_* functions).length_test linter.T and F in formulas
(#33).absolute_path,
duplicate_argument, empty_assignment,
expect_length, expect_not,
expect_null, expect_true_false,
expect_type, literal_coercion,
nested_ifelse, sort,
undesirable_operator.fix() and its variants.fix() on several files without using Git now
opens an interactive menu so that the user confirms they want to run
fix(). In case of non-interactive use, this errors.# nolint for compatibility with
lintr.New linters: expect_named,
numeric_leading_zero, outer_negation,
redundant_ifelse, undesirable_function,
unreachable_code.
fix_dir(), fix_package(),
lint_dir(), lint_package() now have arguments
to exclude paths, linters, and use cache.
Removed browser linter (it is now part of
undesriable_function).
Add support for a flint/config.yml file that
contains the list of linters to use so that one doesn’t need to
constantly specify them in lint() or
fix().
x %in% class(y) where x is
not a string as this is not
equivalent in some cases. Thanks Vincent Arel-Bundock for spotting
this.