ruff/crates/ruff_linter/src/noqa.rs

2987 lines
93 KiB
Rust

use std::collections::BTreeMap;
use std::error::Error;
use std::fmt::Display;
use std::fs;
use std::ops::Add;
use std::path::Path;
use anyhow::Result;
use itertools::Itertools;
use log::warn;
use ruff_db::diagnostic::{Diagnostic, SecondaryCode};
use ruff_python_trivia::{CommentRanges, Cursor, indentation_at_offset};
use ruff_source_file::{LineEnding, LineRanges};
use ruff_text_size::{Ranged, TextLen, TextRange, TextSize};
use rustc_hash::FxHashSet;
use crate::Edit;
use crate::Locator;
use crate::fs::relativize_path;
use crate::registry::Rule;
use crate::rule_redirects::get_redirect_target;
/// Generates an array of edits that matches the length of `messages`.
/// Each potential edit in the array is paired, in order, with the associated diagnostic.
/// Each edit will add a `noqa` comment to the appropriate line in the source to hide
/// the diagnostic. These edits may conflict with each other and should not be applied
/// simultaneously.
pub fn generate_noqa_edits(
path: &Path,
diagnostics: &[Diagnostic],
locator: &Locator,
comment_ranges: &CommentRanges,
external: &[String],
noqa_line_for: &NoqaMapping,
line_ending: LineEnding,
) -> Vec<Option<Edit>> {
let file_directives = FileNoqaDirectives::extract(locator, comment_ranges, external, path);
let exemption = FileExemption::from(&file_directives);
let directives = NoqaDirectives::from_commented_ranges(comment_ranges, external, path, locator);
let comments = find_noqa_comments(diagnostics, locator, &exemption, &directives, noqa_line_for);
build_noqa_edits_by_diagnostic(comments, locator, line_ending)
}
/// A directive to ignore a set of rules either for a given line of Python source code or an entire file (e.g.,
/// `# noqa: F401, F841` or `#ruff: noqa: F401, F841`).
#[derive(Debug)]
pub(crate) enum Directive<'a> {
/// The `noqa` directive ignores all rules (e.g., `# noqa`).
All(All),
/// The `noqa` directive ignores specific rules (e.g., `# noqa: F401, F841`).
Codes(Codes<'a>),
}
#[derive(Debug)]
pub(crate) struct All {
range: TextRange,
}
impl Ranged for All {
/// The range of the `noqa` directive.
fn range(&self) -> TextRange {
self.range
}
}
/// An individual rule code in a `noqa` directive (e.g., `F401`).
#[derive(Debug)]
pub(crate) struct Code<'a> {
code: &'a str,
range: TextRange,
}
impl<'a> Code<'a> {
/// The code that is ignored by the `noqa` directive.
pub(crate) fn as_str(&self) -> &'a str {
self.code
}
}
impl Display for Code<'_> {
fn fmt(&self, fmt: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
fmt.write_str(self.code)
}
}
impl Ranged for Code<'_> {
/// The range of the rule code.
fn range(&self) -> TextRange {
self.range
}
}
#[derive(Debug)]
pub(crate) struct Codes<'a> {
range: TextRange,
codes: Vec<Code<'a>>,
}
impl Codes<'_> {
/// Returns an iterator over the [`Code`]s in the `noqa` directive.
pub(crate) fn iter(&self) -> std::slice::Iter<'_, Code<'_>> {
self.codes.iter()
}
/// Returns `true` if the string list of `codes` includes `code` (or an alias
/// thereof).
pub(crate) fn includes<T: for<'a> PartialEq<&'a str>>(&self, needle: &T) -> bool {
self.iter()
.any(|code| *needle == get_redirect_target(code.as_str()).unwrap_or(code.as_str()))
}
}
impl Ranged for Codes<'_> {
/// The range of the `noqa` directive.
fn range(&self) -> TextRange {
self.range
}
}
/// Returns `true` if the given [`Rule`] is ignored at the specified `lineno`.
pub(crate) fn rule_is_ignored(
code: Rule,
offset: TextSize,
noqa_line_for: &NoqaMapping,
comment_ranges: &CommentRanges,
locator: &Locator,
) -> bool {
let offset = noqa_line_for.resolve(offset);
let line_range = locator.line_range(offset);
let &[comment_range] = comment_ranges.comments_in_range(line_range) else {
return false;
};
match lex_inline_noqa(comment_range, locator.contents()) {
Ok(Some(NoqaLexerOutput {
directive: Directive::All(_),
..
})) => true,
Ok(Some(NoqaLexerOutput {
directive: Directive::Codes(codes),
..
})) => codes.includes(&code.noqa_code()),
_ => false,
}
}
/// A summary of the file-level exemption as extracted from [`FileNoqaDirectives`].
#[derive(Debug)]
pub(crate) enum FileExemption {
/// The file is exempt from all rules.
All(Vec<Rule>),
/// The file is exempt from the given rules.
Codes(Vec<Rule>),
}
impl FileExemption {
/// Returns `true` if the file is exempt from the given rule, as identified by its noqa code.
pub(crate) fn contains_secondary_code(&self, needle: &SecondaryCode) -> bool {
match self {
FileExemption::All(_) => true,
FileExemption::Codes(codes) => codes.iter().any(|code| *needle == code.noqa_code()),
}
}
/// Returns `true` if the file is exempt from the given rule.
pub(crate) fn includes(&self, needle: Rule) -> bool {
match self {
FileExemption::All(_) => true,
FileExemption::Codes(codes) => codes.contains(&needle),
}
}
/// Returns `true` if the file exemption lists the rule directly, rather than via a blanket
/// exemption.
pub(crate) fn enumerates(&self, needle: Rule) -> bool {
let codes = match self {
FileExemption::All(codes) => codes,
FileExemption::Codes(codes) => codes,
};
codes.contains(&needle)
}
}
impl<'a> From<&'a FileNoqaDirectives<'a>> for FileExemption {
fn from(directives: &'a FileNoqaDirectives) -> Self {
let codes = directives
.lines()
.iter()
.flat_map(|line| &line.matches)
.copied()
.collect();
if directives
.lines()
.iter()
.any(|line| matches!(line.parsed_file_exemption, Directive::All(_)))
{
FileExemption::All(codes)
} else {
FileExemption::Codes(codes)
}
}
}
/// The directive for a file-level exemption (e.g., `# ruff: noqa`) from an individual line.
#[derive(Debug)]
pub(crate) struct FileNoqaDirectiveLine<'a> {
/// The range of the text line for which the noqa directive applies.
pub(crate) range: TextRange,
/// The blanket noqa directive.
pub(crate) parsed_file_exemption: Directive<'a>,
/// The codes that are ignored by the parsed exemptions.
pub(crate) matches: Vec<Rule>,
}
impl Ranged for FileNoqaDirectiveLine<'_> {
/// The range of the `noqa` directive.
fn range(&self) -> TextRange {
self.range
}
}
/// All file-level exemptions (e.g., `# ruff: noqa`) from a given Python file.
#[derive(Debug)]
pub(crate) struct FileNoqaDirectives<'a>(Vec<FileNoqaDirectiveLine<'a>>);
impl<'a> FileNoqaDirectives<'a> {
/// Extract the [`FileNoqaDirectives`] for a given Python source file, enumerating any rules
/// that are globally ignored within the file.
pub(crate) fn extract(
locator: &Locator<'a>,
comment_ranges: &CommentRanges,
external: &[String],
path: &Path,
) -> Self {
let mut lines = vec![];
for range in comment_ranges {
let lexed = lex_file_exemption(range, locator.contents());
match lexed {
Ok(Some(NoqaLexerOutput {
warnings,
directive,
})) => {
let no_indentation_at_offset =
indentation_at_offset(range.start(), locator.contents()).is_none();
if !warnings.is_empty() || no_indentation_at_offset {
#[expect(deprecated)]
let line = locator.compute_line_index(range.start());
let path_display = relativize_path(path);
for warning in warnings {
warn!(
"Missing or joined rule code(s) at {path_display}:{line}: {warning}"
);
}
if no_indentation_at_offset {
warn!(
"Unexpected `# ruff: noqa` directive at {path_display}:{line}. File-level suppression comments must appear on their own line. For line-level suppression, omit the `ruff:` prefix."
);
continue;
}
}
let matches = match &directive {
Directive::All(_) => {
vec![]
}
Directive::Codes(codes) => {
codes.iter().filter_map(|code| {
let code = code.as_str();
// Ignore externally-defined rules.
if external.iter().any(|external| code.starts_with(external)) {
return None;
}
if let Ok(rule) = Rule::from_code(get_redirect_target(code).unwrap_or(code))
{
Some(rule)
} else {
#[expect(deprecated)]
let line = locator.compute_line_index(range.start());
let path_display = relativize_path(path);
warn!("Invalid rule code provided to `# ruff: noqa` at {path_display}:{line}: {code}");
None
}
}).collect()
}
};
lines.push(FileNoqaDirectiveLine {
range,
parsed_file_exemption: directive,
matches,
});
}
Err(err) => {
#[expect(deprecated)]
let line = locator.compute_line_index(range.start());
let path_display = relativize_path(path);
warn!("Invalid `# ruff: noqa` directive at {path_display}:{line}: {err}");
}
Ok(None) => {}
}
}
Self(lines)
}
pub(crate) fn lines(&self) -> &[FileNoqaDirectiveLine<'_>] {
&self.0
}
pub(crate) fn is_empty(&self) -> bool {
self.0.is_empty()
}
}
/// Output of lexing a `noqa` directive.
#[derive(Debug)]
pub(crate) struct NoqaLexerOutput<'a> {
warnings: Vec<LexicalWarning>,
directive: Directive<'a>,
}
/// Lexes in-line `noqa` comment, e.g. `# noqa: F401`
fn lex_inline_noqa(
comment_range: TextRange,
source: &str,
) -> Result<Option<NoqaLexerOutput<'_>>, LexicalError> {
let lexer = NoqaLexer::in_range(comment_range, source);
lexer.lex_inline_noqa()
}
/// Lexes file-level exemption comment, e.g. `# ruff: noqa: F401`
fn lex_file_exemption(
comment_range: TextRange,
source: &str,
) -> Result<Option<NoqaLexerOutput<'_>>, LexicalError> {
let lexer = NoqaLexer::in_range(comment_range, source);
lexer.lex_file_exemption()
}
pub(crate) fn lex_codes(text: &str) -> Result<Vec<Code<'_>>, LexicalError> {
let mut lexer = NoqaLexer::in_range(TextRange::new(TextSize::new(0), text.text_len()), text);
lexer.lex_codes()?;
Ok(lexer.codes)
}
/// Lexer for `noqa` comment lines.
///
/// Assumed to be initialized with range or offset starting
/// at the `#` at the beginning of a `noqa` comment.
#[derive(Debug)]
struct NoqaLexer<'a> {
/// Length between offset and end of comment range
line_length: TextSize,
/// Contains convenience methods for lexing
cursor: Cursor<'a>,
/// Byte offset of the start of putative `noqa` comment
///
/// Note: This is updated in the course of lexing in the case
/// where there are multiple comments in a comment range.
/// Ex) `# comment # noqa: F401`
/// start^ ^-- changed to here during lexing
offset: TextSize,
/// Non-fatal warnings collected during lexing
warnings: Vec<LexicalWarning>,
/// Tracks whether we are lexing in a context with a missing delimiter
/// e.g. at `C` in `F401C402`.
missing_delimiter: bool,
/// Rule codes collected during lexing
codes: Vec<Code<'a>>,
}
impl<'a> NoqaLexer<'a> {
/// Initialize [`NoqaLexer`] in the given range of source text.
fn in_range(range: TextRange, source: &'a str) -> Self {
Self {
line_length: range.len(),
offset: range.start(),
cursor: Cursor::new(&source[range]),
warnings: Vec::new(),
missing_delimiter: false,
codes: Vec::new(),
}
}
fn lex_inline_noqa(mut self) -> Result<Option<NoqaLexerOutput<'a>>, LexicalError> {
while !self.cursor.is_eof() {
// Skip over any leading content.
self.cursor.eat_while(|c| c != '#');
// This updates the offset and line length in the case of
// multiple comments in a range, e.g.
// `# First # Second # noqa`
self.offset += self.cursor.token_len();
self.line_length -= self.cursor.token_len();
self.cursor.start_token();
if !self.cursor.eat_char('#') {
continue;
}
self.eat_whitespace();
if !is_noqa_uncased(self.cursor.as_str()) {
continue;
}
self.cursor.skip_bytes("noqa".len());
return Ok(Some(self.lex_directive()?));
}
Ok(None)
}
fn lex_file_exemption(mut self) -> Result<Option<NoqaLexerOutput<'a>>, LexicalError> {
while !self.cursor.is_eof() {
// Skip over any leading content
self.cursor.eat_while(|c| c != '#');
// This updates the offset and line length in the case of
// multiple comments in a range, e.g.
// # First # Second # noqa
self.offset += self.cursor.token_len();
self.line_length -= self.cursor.token_len();
self.cursor.start_token();
// The remaining logic implements a simple regex
// `#\s*(?:ruff|flake8)\s*:\s*(?i)noqa`
// ^
if !self.cursor.eat_char('#') {
continue;
}
// `#\s*(?:ruff|flake8)\s*:\s*(?i)noqa`
// ^^^
self.eat_whitespace();
// `#\s*(?:ruff|flake8)\s*:\s*(?i)noqa`
// ^^^^^^^^^^^^^
if self.cursor.as_str().starts_with("ruff") {
self.cursor.skip_bytes("ruff".len());
} else if self.cursor.as_str().starts_with("flake8") {
self.cursor.skip_bytes("flake8".len());
} else {
continue;
}
// `#\s*(?:ruff|flake8)\s*:\s*(?i)noqa`
// ^^^
self.eat_whitespace();
// `#\s*(?:ruff|flake8)\s*:\s*(?i)noqa`
// ^
if !self.cursor.eat_char(':') {
continue;
}
// `#\s*(?:ruff|flake8)\s*:\s*(?i)noqa`
// ^^^
self.eat_whitespace();
// `#\s*(?:ruff|flake8)\s*:\s*(?i)noqa`
// ^^^^^^^
if !is_noqa_uncased(self.cursor.as_str()) {
continue;
}
self.cursor.skip_bytes("noqa".len());
return Ok(Some(self.lex_directive()?));
}
Ok(None)
}
/// Collect codes in `noqa` comment.
fn lex_directive(mut self) -> Result<NoqaLexerOutput<'a>, LexicalError> {
let range = TextRange::at(self.offset, self.position());
match self.cursor.bump() {
// End of comment
// Ex) # noqa# A comment
None | Some('#') => {
return Ok(NoqaLexerOutput {
warnings: self.warnings,
directive: Directive::All(All { range }),
});
}
// Ex) # noqa: F401,F842
Some(':') => self.lex_codes()?,
// Ex) # noqa A comment
// Ex) # noqa : F401
Some(c) if c.is_whitespace() => {
self.eat_whitespace();
match self.cursor.bump() {
Some(':') => self.lex_codes()?,
_ => {
return Ok(NoqaLexerOutput {
warnings: self.warnings,
directive: Directive::All(All { range }),
});
}
}
}
// Ex) #noqaA comment
// Ex) #noqaF401
_ => {
return Err(LexicalError::InvalidSuffix);
}
}
let Some(last) = self.codes.last() else {
return Err(LexicalError::MissingCodes);
};
Ok(NoqaLexerOutput {
warnings: self.warnings,
directive: Directive::Codes(Codes {
range: TextRange::new(self.offset, last.range.end()),
codes: self.codes,
}),
})
}
fn lex_codes(&mut self) -> Result<(), LexicalError> {
// SAFETY: Every call to `lex_code` advances the cursor at least once.
while !self.cursor.is_eof() {
self.lex_code()?;
}
Ok(())
}
fn lex_code(&mut self) -> Result<(), LexicalError> {
self.cursor.start_token();
self.eat_whitespace();
// Ex) # noqa: F401, ,F841
// ^^^^
if self.cursor.first() == ',' {
self.warnings.push(LexicalWarning::MissingItem(
self.token_range().add(self.offset),
));
self.cursor.eat_char(',');
return Ok(());
}
let before_code = self.cursor.as_str();
// Reset start of token so it does not include whitespace
self.cursor.start_token();
match self.cursor.bump() {
// Ex) # noqa: F401
// ^
Some(c) if c.is_ascii_uppercase() => {
self.cursor.eat_while(|chr| chr.is_ascii_uppercase());
if !self.cursor.eat_if(|c| c.is_ascii_digit()) {
// Fail hard if we're already attempting
// to lex squashed codes, e.g. `F401Fabc`
if self.missing_delimiter {
return Err(LexicalError::InvalidCodeSuffix);
}
// Otherwise we've reached the first invalid code,
// so it could be a comment and we just stop parsing.
// Ex) #noqa: F401 A comment
// we're here^
self.cursor.skip_bytes(self.cursor.as_str().len());
return Ok(());
}
self.cursor.eat_while(|chr| chr.is_ascii_digit());
let code = &before_code[..self.cursor.token_len().to_usize()];
self.push_code(code);
if self.cursor.is_eof() {
return Ok(());
}
self.missing_delimiter = match self.cursor.first() {
',' => {
self.cursor.eat_char(',');
false
}
// Whitespace is an allowed delimiter or the end of the `noqa`.
c if c.is_whitespace() => false,
// e.g. #noqa:F401F842
// ^
// Push a warning and update the `missing_delimiter`
// state but don't consume the character since it's
// part of the next code.
c if c.is_ascii_uppercase() => {
self.warnings
.push(LexicalWarning::MissingDelimiter(TextRange::empty(
self.offset + self.position(),
)));
true
}
// Start of a new comment
// e.g. #noqa: F401#A comment
// ^
'#' => false,
_ => return Err(LexicalError::InvalidCodeSuffix),
};
}
Some(_) => {
// The first time we hit an evidently invalid code,
// it's probably a trailing comment. So stop lexing,
// but don't push an error.
// Ex)
// # noqa: F401 we import this for a reason
// ^----- stop lexing but no error
self.cursor.skip_bytes(self.cursor.as_str().len());
}
None => {}
}
Ok(())
}
/// Push current token to stack of [`Code`] objects
fn push_code(&mut self, code: &'a str) {
self.codes.push(Code {
code,
range: self.token_range().add(self.offset),
});
}
/// Consume whitespace
#[inline]
fn eat_whitespace(&mut self) {
self.cursor.eat_while(char::is_whitespace);
}
/// Current token range relative to offset
#[inline]
fn token_range(&self) -> TextRange {
let end = self.position();
let len = self.cursor.token_len();
TextRange::at(end - len, len)
}
/// Retrieves the current position of the cursor within the line.
#[inline]
fn position(&self) -> TextSize {
self.line_length - self.cursor.text_len()
}
}
/// Helper to check if "noqa" (case-insensitive) appears at the given position
#[inline]
fn is_noqa_uncased(text: &str) -> bool {
matches!(
text.as_bytes(),
[b'n' | b'N', b'o' | b'O', b'q' | b'Q', b'a' | b'A', ..]
)
}
/// Indicates recoverable error encountered while lexing with [`NoqaLexer`]
#[derive(Debug, Clone, Copy)]
enum LexicalWarning {
MissingItem(TextRange),
MissingDelimiter(TextRange),
}
impl Display for LexicalWarning {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
match self {
Self::MissingItem(_) => f.write_str("expected rule code between commas"),
Self::MissingDelimiter(_) => {
f.write_str("expected comma or space delimiter between codes")
}
}
}
}
impl Ranged for LexicalWarning {
fn range(&self) -> TextRange {
match *self {
LexicalWarning::MissingItem(text_range) => text_range,
LexicalWarning::MissingDelimiter(text_range) => text_range,
}
}
}
/// Fatal error occurring while lexing a `noqa` comment as in [`NoqaLexer`]
#[derive(Debug)]
pub(crate) enum LexicalError {
/// The `noqa` directive was missing valid codes (e.g., `# noqa: unused-import` instead of `# noqa: F401`).
MissingCodes,
/// The `noqa` directive used an invalid suffix (e.g., `# noqa; F401` instead of `# noqa: F401`).
InvalidSuffix,
/// The `noqa` code matched the regex "[A-Z]+[0-9]+" with text remaining
/// (e.g. `# noqa: F401abc`)
InvalidCodeSuffix,
}
impl Display for LexicalError {
fn fmt(&self, fmt: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
match self {
LexicalError::MissingCodes => fmt.write_str("expected a comma-separated list of codes (e.g., `# noqa: F401, F841`)."),
LexicalError::InvalidSuffix => {
fmt.write_str("expected `:` followed by a comma-separated list of codes (e.g., `# noqa: F401, F841`).")
}
LexicalError::InvalidCodeSuffix => {
fmt.write_str("expected code to consist of uppercase letters followed by digits only (e.g. `F401`)")
}
}
}
}
impl Error for LexicalError {}
/// Adds noqa comments to suppress all messages of a file.
pub(crate) fn add_noqa(
path: &Path,
diagnostics: &[Diagnostic],
locator: &Locator,
comment_ranges: &CommentRanges,
external: &[String],
noqa_line_for: &NoqaMapping,
line_ending: LineEnding,
) -> Result<usize> {
let (count, output) = add_noqa_inner(
path,
diagnostics,
locator,
comment_ranges,
external,
noqa_line_for,
line_ending,
);
fs::write(path, output)?;
Ok(count)
}
fn add_noqa_inner(
path: &Path,
diagnostics: &[Diagnostic],
locator: &Locator,
comment_ranges: &CommentRanges,
external: &[String],
noqa_line_for: &NoqaMapping,
line_ending: LineEnding,
) -> (usize, String) {
let mut count = 0;
// Whether the file is exempted from all checks.
let directives = FileNoqaDirectives::extract(locator, comment_ranges, external, path);
let exemption = FileExemption::from(&directives);
let directives = NoqaDirectives::from_commented_ranges(comment_ranges, external, path, locator);
let comments = find_noqa_comments(diagnostics, locator, &exemption, &directives, noqa_line_for);
let edits = build_noqa_edits_by_line(comments, locator, line_ending);
let contents = locator.contents();
let mut output = String::with_capacity(contents.len());
let mut last_append = TextSize::default();
for (_, edit) in edits {
output.push_str(&contents[TextRange::new(last_append, edit.start())]);
edit.write(&mut output);
count += 1;
last_append = edit.end();
}
output.push_str(&contents[TextRange::new(last_append, TextSize::of(contents))]);
(count, output)
}
fn build_noqa_edits_by_diagnostic(
comments: Vec<Option<NoqaComment>>,
locator: &Locator,
line_ending: LineEnding,
) -> Vec<Option<Edit>> {
let mut edits = Vec::default();
for comment in comments {
match comment {
Some(comment) => {
if let Some(noqa_edit) = generate_noqa_edit(
comment.directive,
comment.line,
FxHashSet::from_iter([comment.code]),
locator,
line_ending,
) {
edits.push(Some(noqa_edit.into_edit()));
}
}
None => edits.push(None),
}
}
edits
}
fn build_noqa_edits_by_line<'a>(
comments: Vec<Option<NoqaComment<'a>>>,
locator: &Locator,
line_ending: LineEnding,
) -> BTreeMap<TextSize, NoqaEdit<'a>> {
let mut comments_by_line = BTreeMap::default();
for comment in comments.into_iter().flatten() {
comments_by_line
.entry(comment.line)
.or_insert_with(Vec::default)
.push(comment);
}
let mut edits = BTreeMap::default();
for (offset, matches) in comments_by_line {
let Some(first_match) = matches.first() else {
continue;
};
let directive = first_match.directive;
if let Some(edit) = generate_noqa_edit(
directive,
offset,
matches
.into_iter()
.map(|NoqaComment { code, .. }| code)
.collect(),
locator,
line_ending,
) {
edits.insert(offset, edit);
}
}
edits
}
struct NoqaComment<'a> {
line: TextSize,
code: &'a SecondaryCode,
directive: Option<&'a Directive<'a>>,
}
fn find_noqa_comments<'a>(
diagnostics: &'a [Diagnostic],
locator: &'a Locator,
exemption: &'a FileExemption,
directives: &'a NoqaDirectives,
noqa_line_for: &NoqaMapping,
) -> Vec<Option<NoqaComment<'a>>> {
// List of noqa comments, ordered to match up with `messages`
let mut comments_by_line: Vec<Option<NoqaComment<'a>>> = vec![];
// Mark any non-ignored diagnostics.
for message in diagnostics {
let Some(code) = message.secondary_code() else {
comments_by_line.push(None);
continue;
};
if exemption.contains_secondary_code(code) {
comments_by_line.push(None);
continue;
}
// Is the violation ignored by a `noqa` directive on the parent line?
if let Some(parent) = message.parent() {
if let Some(directive_line) =
directives.find_line_with_directive(noqa_line_for.resolve(parent))
{
match &directive_line.directive {
Directive::All(_) => {
comments_by_line.push(None);
continue;
}
Directive::Codes(codes) => {
if codes.includes(code) {
comments_by_line.push(None);
continue;
}
}
}
}
}
let noqa_offset = noqa_line_for.resolve(message.expect_range().start());
// Or ignored by the directive itself?
if let Some(directive_line) = directives.find_line_with_directive(noqa_offset) {
match &directive_line.directive {
Directive::All(_) => {
comments_by_line.push(None);
continue;
}
directive @ Directive::Codes(codes) => {
if !codes.includes(code) {
comments_by_line.push(Some(NoqaComment {
line: directive_line.start(),
code,
directive: Some(directive),
}));
}
continue;
}
}
}
// There's no existing noqa directive that suppresses the diagnostic.
comments_by_line.push(Some(NoqaComment {
line: locator.line_start(noqa_offset),
code,
directive: None,
}));
}
comments_by_line
}
struct NoqaEdit<'a> {
edit_range: TextRange,
noqa_codes: FxHashSet<&'a SecondaryCode>,
codes: Option<&'a Codes<'a>>,
line_ending: LineEnding,
}
impl NoqaEdit<'_> {
fn into_edit(self) -> Edit {
let mut edit_content = String::new();
self.write(&mut edit_content);
Edit::range_replacement(edit_content, self.edit_range)
}
fn write(&self, writer: &mut impl std::fmt::Write) {
write!(writer, " # noqa: ").unwrap();
match self.codes {
Some(codes) => {
push_codes(
writer,
self.noqa_codes
.iter()
.map(|code| code.as_str())
.chain(codes.iter().map(Code::as_str))
.sorted_unstable(),
);
}
None => {
push_codes(writer, self.noqa_codes.iter().sorted_unstable());
}
}
write!(writer, "{}", self.line_ending.as_str()).unwrap();
}
}
impl Ranged for NoqaEdit<'_> {
fn range(&self) -> TextRange {
self.edit_range
}
}
fn generate_noqa_edit<'a>(
directive: Option<&'a Directive>,
offset: TextSize,
noqa_codes: FxHashSet<&'a SecondaryCode>,
locator: &Locator,
line_ending: LineEnding,
) -> Option<NoqaEdit<'a>> {
let line_range = locator.full_line_range(offset);
let edit_range;
let codes;
// Add codes.
match directive {
None => {
let trimmed_line = locator.slice(line_range).trim_end();
edit_range = TextRange::new(TextSize::of(trimmed_line), line_range.len()) + offset;
codes = None;
}
Some(Directive::Codes(existing_codes)) => {
// find trimmed line without the noqa
let trimmed_line = locator
.slice(TextRange::new(line_range.start(), existing_codes.start()))
.trim_end();
edit_range = TextRange::new(TextSize::of(trimmed_line), line_range.len()) + offset;
codes = Some(existing_codes);
}
Some(Directive::All(_)) => return None,
}
Some(NoqaEdit {
edit_range,
noqa_codes,
codes,
line_ending,
})
}
fn push_codes<I: Display>(writer: &mut dyn std::fmt::Write, codes: impl Iterator<Item = I>) {
let mut first = true;
for code in codes {
if !first {
write!(writer, ", ").unwrap();
}
write!(writer, "{code}").unwrap();
first = false;
}
}
#[derive(Debug)]
pub(crate) struct NoqaDirectiveLine<'a> {
/// The range of the text line for which the noqa directive applies.
pub(crate) range: TextRange,
/// The noqa directive.
pub(crate) directive: Directive<'a>,
/// The codes that are ignored by the directive.
pub(crate) matches: Vec<Rule>,
/// Whether the directive applies to `range.end`.
pub(crate) includes_end: bool,
}
impl Ranged for NoqaDirectiveLine<'_> {
/// The range of the `noqa` directive.
fn range(&self) -> TextRange {
self.range
}
}
#[derive(Debug, Default)]
pub(crate) struct NoqaDirectives<'a> {
inner: Vec<NoqaDirectiveLine<'a>>,
}
impl<'a> NoqaDirectives<'a> {
pub(crate) fn from_commented_ranges(
comment_ranges: &CommentRanges,
external: &[String],
path: &Path,
locator: &'a Locator<'a>,
) -> Self {
let mut directives = Vec::new();
for range in comment_ranges {
let lexed = lex_inline_noqa(range, locator.contents());
match lexed {
Ok(Some(NoqaLexerOutput {
warnings,
directive,
})) => {
if !warnings.is_empty() {
#[expect(deprecated)]
let line = locator.compute_line_index(range.start());
let path_display = relativize_path(path);
for warning in warnings {
warn!(
"Missing or joined rule code(s) at {path_display}:{line}: {warning}"
);
}
}
if let Directive::Codes(codes) = &directive {
// Warn on invalid rule codes.
for code in &codes.codes {
// Ignore externally-defined rules.
if !external
.iter()
.any(|external| code.as_str().starts_with(external))
{
if Rule::from_code(
get_redirect_target(code.as_str()).unwrap_or(code.as_str()),
)
.is_err()
{
#[expect(deprecated)]
let line = locator.compute_line_index(range.start());
let path_display = relativize_path(path);
warn!(
"Invalid rule code provided to `# noqa` at {path_display}:{line}: {code}"
);
}
}
}
}
// noqa comments are guaranteed to be single line.
let range = locator.line_range(range.start());
directives.push(NoqaDirectiveLine {
range,
directive,
matches: Vec::new(),
includes_end: range.end() == locator.contents().text_len(),
});
}
Err(err) => {
#[expect(deprecated)]
let line = locator.compute_line_index(range.start());
let path_display = relativize_path(path);
warn!("Invalid `# noqa` directive on {path_display}:{line}: {err}");
}
Ok(None) => {}
}
}
Self { inner: directives }
}
pub(crate) fn find_line_with_directive(
&self,
offset: TextSize,
) -> Option<&NoqaDirectiveLine<'_>> {
self.find_line_index(offset).map(|index| &self.inner[index])
}
pub(crate) fn find_line_with_directive_mut(
&mut self,
offset: TextSize,
) -> Option<&mut NoqaDirectiveLine<'a>> {
if let Some(index) = self.find_line_index(offset) {
Some(&mut self.inner[index])
} else {
None
}
}
fn find_line_index(&self, offset: TextSize) -> Option<usize> {
self.inner
.binary_search_by(|directive| {
if directive.range.end() < offset {
std::cmp::Ordering::Less
} else if directive.range.start() > offset {
std::cmp::Ordering::Greater
}
// At this point, end >= offset, start <= offset
else if !directive.includes_end && directive.range.end() == offset {
std::cmp::Ordering::Less
} else {
std::cmp::Ordering::Equal
}
})
.ok()
}
pub(crate) fn lines(&self) -> &[NoqaDirectiveLine<'_>] {
&self.inner
}
pub(crate) fn is_empty(&self) -> bool {
self.inner.is_empty()
}
}
/// Remaps offsets falling into one of the ranges to instead check for a noqa comment on the
/// line specified by the offset.
#[derive(Debug, Default, PartialEq, Eq)]
pub struct NoqaMapping {
ranges: Vec<TextRange>,
}
impl NoqaMapping {
pub(crate) fn with_capacity(capacity: usize) -> Self {
Self {
ranges: Vec::with_capacity(capacity),
}
}
/// Returns the re-mapped position or `position` if no mapping exists.
pub(crate) fn resolve(&self, offset: TextSize) -> TextSize {
let index = self.ranges.binary_search_by(|range| {
if range.end() < offset {
std::cmp::Ordering::Less
} else if range.contains(offset) {
std::cmp::Ordering::Equal
} else {
std::cmp::Ordering::Greater
}
});
if let Ok(index) = index {
self.ranges[index].end()
} else {
offset
}
}
pub(crate) fn push_mapping(&mut self, range: TextRange) {
if let Some(last_range) = self.ranges.last_mut() {
// Strictly sorted insertion
if last_range.end() < range.start() {
// OK
} else if range.end() < last_range.start() {
// Incoming range is strictly before the last range which violates
// the function's contract.
panic!("Ranges must be inserted in sorted order")
} else {
// Here, it's guaranteed that `last_range` and `range` overlap
// in some way. We want to merge them into a single range.
*last_range = last_range.cover(range);
return;
}
}
self.ranges.push(range);
}
}
impl FromIterator<TextRange> for NoqaMapping {
fn from_iter<T: IntoIterator<Item = TextRange>>(iter: T) -> Self {
let mut mappings = NoqaMapping::default();
for range in iter {
mappings.push_mapping(range);
}
mappings
}
}
#[cfg(test)]
mod tests {
use std::path::Path;
use insta::assert_debug_snapshot;
use ruff_python_trivia::CommentRanges;
use ruff_source_file::{LineEnding, SourceFileBuilder};
use ruff_text_size::{TextLen, TextRange, TextSize};
use crate::noqa::{
Directive, LexicalError, NoqaLexerOutput, NoqaMapping, add_noqa_inner, lex_codes,
lex_file_exemption, lex_inline_noqa,
};
use crate::rules::pycodestyle::rules::{AmbiguousVariableName, UselessSemicolon};
use crate::rules::pyflakes::rules::UnusedVariable;
use crate::rules::pyupgrade::rules::PrintfStringFormatting;
use crate::{Edit, Violation};
use crate::{Locator, generate_noqa_edits};
fn assert_lexed_ranges_match_slices(
directive: Result<Option<NoqaLexerOutput>, LexicalError>,
source: &str,
) {
if let Ok(Some(NoqaLexerOutput {
warnings: _,
directive: Directive::Codes(codes),
})) = directive
{
for code in codes.iter() {
assert_eq!(&source[code.range], code.code);
}
}
}
#[test]
fn noqa_lex_codes() {
let source = " F401,,F402F403 # and so on";
assert_debug_snapshot!(lex_codes(source), @r#"
Ok(
[
Code {
code: "F401",
range: 1..5,
},
Code {
code: "F402",
range: 7..11,
},
Code {
code: "F403",
range: 11..15,
},
],
)
"#);
}
#[test]
fn noqa_all() {
let source = "# noqa";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..6,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_no_code() {
let source = "# noqa:";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_no_code_invalid_suffix() {
let source = "# noqa: foo";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_no_code_trailing_content() {
let source = "# noqa: # Foo";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn malformed_code_1() {
let source = "# noqa: F";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn malformed_code_2() {
let source = "# noqa: RUF001, F";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..14,
codes: [
Code {
code: "RUF001",
range: 8..14,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn malformed_code_3() {
let source = "# noqa: RUF001F";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Err(
InvalidCodeSuffix,
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_code() {
let source = "# noqa: F401";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..12,
codes: [
Code {
code: "F401",
range: 8..12,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_codes() {
let source = "# noqa: F401, F841";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..18,
codes: [
Code {
code: "F401",
range: 8..12,
},
Code {
code: "F841",
range: 14..18,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_all_case_insensitive() {
let source = "# NOQA";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..6,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_code_case_insensitive() {
let source = "# NOQA: F401";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..12,
codes: [
Code {
code: "F401",
range: 8..12,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_codes_case_insensitive() {
let source = "# NOQA: F401, F841";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..18,
codes: [
Code {
code: "F401",
range: 8..12,
},
Code {
code: "F841",
range: 14..18,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_leading_space() {
let source = "# # noqa: F401";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 4..16,
codes: [
Code {
code: "F401",
range: 12..16,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_trailing_space() {
let source = "# noqa: F401 #";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..12,
codes: [
Code {
code: "F401",
range: 8..12,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_all_no_space() {
let source = "#noqa";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..5,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_code_no_space() {
let source = "#noqa:F401";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..10,
codes: [
Code {
code: "F401",
range: 6..10,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_codes_no_space() {
let source = "#noqa:F401,F841";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..15,
codes: [
Code {
code: "F401",
range: 6..10,
},
Code {
code: "F841",
range: 11..15,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_all_multi_space() {
let source = "# noqa";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..7,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_code_multi_space() {
let source = "# noqa: F401";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..13,
codes: [
Code {
code: "F401",
range: 9..13,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_codes_multi_space() {
let source = "# noqa: F401, F841";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..20,
codes: [
Code {
code: "F401",
range: 9..13,
},
Code {
code: "F841",
range: 16..20,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_code_leading_hashes() {
let source = "###noqa: F401";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 2..13,
codes: [
Code {
code: "F401",
range: 9..13,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_code_leading_hashes_with_spaces() {
let source = "# # # noqa: F401";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 6..18,
codes: [
Code {
code: "F401",
range: 14..18,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_all_leading_comment() {
let source = "# Some comment describing the noqa # noqa";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 35..41,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_code_leading_comment() {
let source = "# Some comment describing the noqa # noqa: F401";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 35..47,
codes: [
Code {
code: "F401",
range: 43..47,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_codes_leading_comment() {
let source = "# Some comment describing the noqa # noqa: F401, F841";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 35..53,
codes: [
Code {
code: "F401",
range: 43..47,
},
Code {
code: "F841",
range: 49..53,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_all_trailing_comment() {
let source = "# noqa # Some comment describing the noqa";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..6,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_code_trailing_comment() {
let source = "# noqa: F401 # Some comment describing the noqa";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..12,
codes: [
Code {
code: "F401",
range: 8..12,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_codes_trailing_comment() {
let source = "# noqa: F401, F841 # Some comment describing the noqa";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..18,
codes: [
Code {
code: "F401",
range: 8..12,
},
Code {
code: "F841",
range: 14..18,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_invalid_codes() {
let source = "# noqa: unused-import, F401, some other code";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_squashed_codes() {
let source = "# noqa: F401F841";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [
MissingDelimiter(
12..12,
),
],
directive: Codes(
Codes {
range: 0..16,
codes: [
Code {
code: "F401",
range: 8..12,
},
Code {
code: "F841",
range: 12..16,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_empty_comma() {
let source = "# noqa: F401,,F841";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [
MissingItem(
13..13,
),
],
directive: Codes(
Codes {
range: 0..18,
codes: [
Code {
code: "F401",
range: 8..12,
},
Code {
code: "F841",
range: 14..18,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_empty_comma_space() {
let source = "# noqa: F401, ,F841";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [
MissingItem(
13..14,
),
],
directive: Codes(
Codes {
range: 0..19,
codes: [
Code {
code: "F401",
range: 8..12,
},
Code {
code: "F841",
range: 15..19,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_non_code() {
let source = "# noqa: F401 We're ignoring an import";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..12,
codes: [
Code {
code: "F401",
range: 8..12,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_code_invalid_code_suffix() {
let source = "# noqa: F401abc";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Err(
InvalidCodeSuffix,
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn noqa_invalid_suffix() {
let source = "# noqa[F401]";
let directive = lex_inline_noqa(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Err(
InvalidSuffix,
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn flake8_exemption_all() {
let source = "# flake8: noqa";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..14,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn flake8_noqa_no_code() {
let source = "# flake8: noqa:";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn flake8_noqa_no_code_invalid_suffix() {
let source = "# flake8: noqa: foo";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn flake8_noqa_no_code_trailing_content() {
let source = "# flake8: noqa: # Foo";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn flake8_malformed_code_1() {
let source = "# flake8: noqa: F";
let directive = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn flake8_malformed_code_2() {
let source = "# flake8: noqa: RUF001, F";
let directive = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..22,
codes: [
Code {
code: "RUF001",
range: 16..22,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn flake8_malformed_code_3() {
let source = "# flake8: noqa: RUF001F";
let directive = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Err(
InvalidCodeSuffix,
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn ruff_exemption_all() {
let source = "# ruff: noqa";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..12,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_noqa_no_code() {
let source = "# ruff: noqa:";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_noqa_no_code_invalid_suffix() {
let source = "# ruff: noqa: foo";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_noqa_no_code_trailing_content() {
let source = "# ruff: noqa: # Foo";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_malformed_code_1() {
let source = "# ruff: noqa: F";
let directive = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Err(
MissingCodes,
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn ruff_malformed_code_2() {
let source = "# ruff: noqa: RUF001, F";
let directive = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..20,
codes: [
Code {
code: "RUF001",
range: 14..20,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn ruff_malformed_code_3() {
let source = "# ruff: noqa: RUF001F";
let directive = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(directive, @r"
Err(
InvalidCodeSuffix,
)
");
assert_lexed_ranges_match_slices(directive, source);
}
#[test]
fn flake8_exemption_all_no_space() {
let source = "#flake8:noqa";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..12,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_all_no_space() {
let source = "#ruff:noqa";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..10,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn flake8_exemption_codes() {
// Note: Flake8 doesn't support this; it's treated as a blanket exemption.
let source = "# flake8: noqa: F401, F841";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..26,
codes: [
Code {
code: "F401",
range: 16..20,
},
Code {
code: "F841",
range: 22..26,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_codes() {
let source = "# ruff: noqa: F401, F841";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..24,
codes: [
Code {
code: "F401",
range: 14..18,
},
Code {
code: "F841",
range: 20..24,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_codes_leading_hashes() {
let source = "#### ruff: noqa: F401, F841";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 3..27,
codes: [
Code {
code: "F401",
range: 17..21,
},
Code {
code: "F841",
range: 23..27,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_squashed_codes() {
let source = "# ruff: noqa: F401F841";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [
MissingDelimiter(
18..18,
),
],
directive: Codes(
Codes {
range: 0..22,
codes: [
Code {
code: "F401",
range: 14..18,
},
Code {
code: "F841",
range: 18..22,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_empty_comma() {
let source = "# ruff: noqa: F401,,F841";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [
MissingItem(
19..19,
),
],
directive: Codes(
Codes {
range: 0..24,
codes: [
Code {
code: "F401",
range: 14..18,
},
Code {
code: "F841",
range: 20..24,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_empty_comma_space() {
let source = "# ruff: noqa: F401, ,F841";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [
MissingItem(
19..20,
),
],
directive: Codes(
Codes {
range: 0..25,
codes: [
Code {
code: "F401",
range: 14..18,
},
Code {
code: "F841",
range: 21..25,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_invalid_code_suffix() {
let source = "# ruff: noqa: F401abc";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Err(
InvalidCodeSuffix,
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_code_leading_comment() {
let source = "# Leading comment # ruff: noqa: F401";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 18..36,
codes: [
Code {
code: "F401",
range: 32..36,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_code_trailing_comment() {
let source = "# ruff: noqa: F401 # Trailing comment";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..18,
codes: [
Code {
code: "F401",
range: 14..18,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_all_leading_comment() {
let source = "# Leading comment # ruff: noqa";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 18..30,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_all_trailing_comment() {
let source = "# ruff: noqa # Trailing comment";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..12,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_code_trailing_comment_no_space() {
let source = "# ruff: noqa: F401# And another comment";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..18,
codes: [
Code {
code: "F401",
range: 14..18,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_all_trailing_comment_no_space() {
let source = "# ruff: noqa# Trailing comment";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..12,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_all_trailing_comment_no_hash() {
let source = "# ruff: noqa Trailing comment";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..12,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_code_trailing_comment_no_hash() {
let source = "# ruff: noqa: F401 Trailing comment";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r#"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: Codes(
Codes {
range: 0..18,
codes: [
Code {
code: "F401",
range: 14..18,
},
],
},
),
},
),
)
"#);
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn flake8_exemption_all_case_insensitive() {
let source = "# flake8: NoQa";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..14,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn ruff_exemption_all_case_insensitive() {
let source = "# ruff: NoQa";
let exemption = lex_file_exemption(TextRange::up_to(source.text_len()), source);
assert_debug_snapshot!(exemption, @r"
Ok(
Some(
NoqaLexerOutput {
warnings: [],
directive: All(
All {
range: 0..12,
},
),
},
),
)
");
assert_lexed_ranges_match_slices(exemption, source);
}
#[test]
fn modification() {
let path = Path::new("/tmp/foo.txt");
let contents = "x = 1";
let noqa_line_for = NoqaMapping::default();
let (count, output) = add_noqa_inner(
path,
&[],
&Locator::new(contents),
&CommentRanges::default(),
&[],
&noqa_line_for,
LineEnding::Lf,
);
assert_eq!(count, 0);
assert_eq!(output, format!("{contents}"));
let source_file = SourceFileBuilder::new(path.to_string_lossy(), contents).finish();
let messages = [UnusedVariable {
name: "x".to_string(),
}
.into_diagnostic(
TextRange::new(TextSize::from(0), TextSize::from(0)),
&source_file,
)];
let contents = "x = 1";
let noqa_line_for = NoqaMapping::default();
let (count, output) = add_noqa_inner(
path,
&messages,
&Locator::new(contents),
&CommentRanges::default(),
&[],
&noqa_line_for,
LineEnding::Lf,
);
assert_eq!(count, 1);
assert_eq!(output, "x = 1 # noqa: F841\n");
let source_file = SourceFileBuilder::new(path.to_string_lossy(), contents).finish();
let messages = [
AmbiguousVariableName("x".to_string()).into_diagnostic(
TextRange::new(TextSize::from(0), TextSize::from(0)),
&source_file,
),
UnusedVariable {
name: "x".to_string(),
}
.into_diagnostic(
TextRange::new(TextSize::from(0), TextSize::from(0)),
&source_file,
),
];
let contents = "x = 1 # noqa: E741\n";
let noqa_line_for = NoqaMapping::default();
let comment_ranges =
CommentRanges::new(vec![TextRange::new(TextSize::from(7), TextSize::from(19))]);
let (count, output) = add_noqa_inner(
path,
&messages,
&Locator::new(contents),
&comment_ranges,
&[],
&noqa_line_for,
LineEnding::Lf,
);
assert_eq!(count, 1);
assert_eq!(output, "x = 1 # noqa: E741, F841\n");
let source_file = SourceFileBuilder::new(path.to_string_lossy(), contents).finish();
let messages = [
AmbiguousVariableName("x".to_string()).into_diagnostic(
TextRange::new(TextSize::from(0), TextSize::from(0)),
&source_file,
),
UnusedVariable {
name: "x".to_string(),
}
.into_diagnostic(
TextRange::new(TextSize::from(0), TextSize::from(0)),
&source_file,
),
];
let contents = "x = 1 # noqa";
let noqa_line_for = NoqaMapping::default();
let comment_ranges =
CommentRanges::new(vec![TextRange::new(TextSize::from(7), TextSize::from(13))]);
let (count, output) = add_noqa_inner(
path,
&messages,
&Locator::new(contents),
&comment_ranges,
&[],
&noqa_line_for,
LineEnding::Lf,
);
assert_eq!(count, 0);
assert_eq!(output, "x = 1 # noqa");
}
#[test]
fn multiline_comment() {
let path = Path::new("/tmp/foo.txt");
let source = r#"
print(
"""First line
second line
third line
%s"""
% name
)
"#;
let noqa_line_for = [TextRange::new(8.into(), 68.into())].into_iter().collect();
let source_file = SourceFileBuilder::new(path.to_string_lossy(), source).finish();
let messages = [PrintfStringFormatting
.into_diagnostic(TextRange::new(12.into(), 79.into()), &source_file)];
let comment_ranges = CommentRanges::default();
let edits = generate_noqa_edits(
path,
&messages,
&Locator::new(source),
&comment_ranges,
&[],
&noqa_line_for,
LineEnding::Lf,
);
assert_eq!(
edits,
vec![Some(Edit::replacement(
" # noqa: UP031\n".to_string(),
68.into(),
69.into()
))]
);
}
#[test]
fn syntax_error() {
let path = Path::new("/tmp/foo.txt");
let source = "\
foo;
bar =
";
let source_file = SourceFileBuilder::new(path.to_string_lossy(), source).finish();
let messages =
[UselessSemicolon.into_diagnostic(TextRange::new(4.into(), 5.into()), &source_file)];
let noqa_line_for = NoqaMapping::default();
let comment_ranges = CommentRanges::default();
let edits = generate_noqa_edits(
path,
&messages,
&Locator::new(source),
&comment_ranges,
&[],
&noqa_line_for,
LineEnding::Lf,
);
assert_eq!(
edits,
vec![Some(Edit::replacement(
" # noqa: E703\n".to_string(),
4.into(),
5.into()
))]
);
}
}