mirror of
https://github.com/astral-sh/ruff.git
synced 2025-07-23 04:55:21 +00:00

## Summary This PR updates the parser test infrastructure to validate the token ranges. From the code documentation: ``` /// Verifies that: /// * the ranges are strictly increasing when loop the tokens in insertion order /// * all ranges are within the length of the source code ``` Follow-up from #12016 and #12017 resolves: #11938 ## Test Plan Make sure that there are no failures.
327 lines
10 KiB
Rust
327 lines
10 KiB
Rust
use std::cmp::Ordering;
|
|
use std::fmt::{Formatter, Write};
|
|
use std::fs;
|
|
use std::path::Path;
|
|
|
|
use annotate_snippets::display_list::{DisplayList, FormatOptions};
|
|
use annotate_snippets::snippet::{AnnotationType, Slice, Snippet, SourceAnnotation};
|
|
|
|
use ruff_python_ast::visitor::source_order::{walk_module, SourceOrderVisitor, TraversalSignal};
|
|
use ruff_python_ast::{AnyNodeRef, Mod};
|
|
use ruff_python_parser::{parse_unchecked, Mode, ParseErrorType, Token};
|
|
use ruff_source_file::{LineIndex, OneIndexed, SourceCode};
|
|
use ruff_text_size::{Ranged, TextLen, TextRange, TextSize};
|
|
|
|
#[test]
|
|
fn valid_syntax() {
|
|
insta::glob!("../resources", "valid/**/*.py", test_valid_syntax);
|
|
}
|
|
|
|
#[test]
|
|
fn invalid_syntax() {
|
|
insta::glob!("../resources", "invalid/**/*.py", test_invalid_syntax);
|
|
}
|
|
|
|
#[test]
|
|
fn inline_ok() {
|
|
insta::glob!("../resources/inline", "ok/**/*.py", test_valid_syntax);
|
|
}
|
|
|
|
#[test]
|
|
fn inline_err() {
|
|
insta::glob!("../resources/inline", "err/**/*.py", test_invalid_syntax);
|
|
}
|
|
|
|
/// Asserts that the parser generates no syntax errors for a valid program.
|
|
/// Snapshots the AST.
|
|
fn test_valid_syntax(input_path: &Path) {
|
|
let source = fs::read_to_string(input_path).expect("Expected test file to exist");
|
|
let parsed = parse_unchecked(&source, Mode::Module);
|
|
|
|
if !parsed.is_valid() {
|
|
let line_index = LineIndex::from_source_text(&source);
|
|
let source_code = SourceCode::new(&source, &line_index);
|
|
|
|
let mut message = "Expected no syntax errors for a valid program but the parser generated the following errors:\n".to_string();
|
|
|
|
for error in parsed.errors() {
|
|
writeln!(
|
|
&mut message,
|
|
"{}\n",
|
|
CodeFrame {
|
|
range: error.location,
|
|
error,
|
|
source_code: &source_code,
|
|
}
|
|
)
|
|
.unwrap();
|
|
}
|
|
|
|
panic!("{input_path:?}: {message}");
|
|
}
|
|
|
|
validate_tokens(parsed.tokens(), source.text_len(), input_path);
|
|
validate_ast(parsed.syntax(), source.text_len(), input_path);
|
|
|
|
let mut output = String::new();
|
|
writeln!(&mut output, "## AST").unwrap();
|
|
writeln!(&mut output, "\n```\n{:#?}\n```", parsed.syntax()).unwrap();
|
|
|
|
insta::with_settings!({
|
|
omit_expression => true,
|
|
input_file => input_path,
|
|
prepend_module_to_snapshot => false,
|
|
}, {
|
|
insta::assert_snapshot!(output);
|
|
});
|
|
}
|
|
|
|
/// Assert that the parser generates at least one syntax error for the given input file.
|
|
/// Snapshots the AST and the error messages.
|
|
fn test_invalid_syntax(input_path: &Path) {
|
|
let source = fs::read_to_string(input_path).expect("Expected test file to exist");
|
|
let parsed = parse_unchecked(&source, Mode::Module);
|
|
|
|
assert!(
|
|
!parsed.is_valid(),
|
|
"{input_path:?}: Expected parser to generate at least one syntax error for a program containing syntax errors."
|
|
);
|
|
|
|
validate_tokens(parsed.tokens(), source.text_len(), input_path);
|
|
validate_ast(parsed.syntax(), source.text_len(), input_path);
|
|
|
|
let mut output = String::new();
|
|
writeln!(&mut output, "## AST").unwrap();
|
|
writeln!(&mut output, "\n```\n{:#?}\n```", parsed.syntax()).unwrap();
|
|
|
|
writeln!(&mut output, "## Errors\n").unwrap();
|
|
|
|
let line_index = LineIndex::from_source_text(&source);
|
|
let source_code = SourceCode::new(&source, &line_index);
|
|
|
|
for error in parsed.errors() {
|
|
writeln!(
|
|
&mut output,
|
|
"{}\n",
|
|
CodeFrame {
|
|
range: error.location,
|
|
error,
|
|
source_code: &source_code,
|
|
}
|
|
)
|
|
.unwrap();
|
|
}
|
|
|
|
insta::with_settings!({
|
|
omit_expression => true,
|
|
input_file => input_path,
|
|
prepend_module_to_snapshot => false,
|
|
}, {
|
|
insta::assert_snapshot!(output);
|
|
});
|
|
}
|
|
|
|
// Test that is intentionally ignored by default.
|
|
// Use it for quickly debugging a parser issue.
|
|
#[test]
|
|
#[ignore]
|
|
#[allow(clippy::print_stdout)]
|
|
fn parser_quick_test() {
|
|
let source = "\
|
|
f'{'
|
|
f'{foo!r'
|
|
";
|
|
|
|
let parsed = parse_unchecked(source, Mode::Module);
|
|
|
|
println!("AST:\n----\n{:#?}", parsed.syntax());
|
|
println!("Tokens:\n-------\n{:#?}", parsed.tokens());
|
|
|
|
if !parsed.is_valid() {
|
|
println!("Errors:\n-------");
|
|
|
|
let line_index = LineIndex::from_source_text(source);
|
|
let source_code = SourceCode::new(source, &line_index);
|
|
|
|
for error in parsed.errors() {
|
|
// Sometimes the code frame doesn't show the error message, so we print
|
|
// the message as well.
|
|
println!("Syntax Error: {error}");
|
|
println!(
|
|
"{}\n",
|
|
CodeFrame {
|
|
range: error.location,
|
|
error,
|
|
source_code: &source_code,
|
|
}
|
|
);
|
|
}
|
|
|
|
println!();
|
|
}
|
|
}
|
|
|
|
struct CodeFrame<'a> {
|
|
range: TextRange,
|
|
error: &'a ParseErrorType,
|
|
source_code: &'a SourceCode<'a, 'a>,
|
|
}
|
|
|
|
impl std::fmt::Display for CodeFrame<'_> {
|
|
fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
|
|
// Copied and modified from ruff_linter/src/message/text.rs
|
|
let content_start_index = self.source_code.line_index(self.range.start());
|
|
let mut start_index = content_start_index.saturating_sub(2);
|
|
|
|
// Trim leading empty lines.
|
|
while start_index < content_start_index {
|
|
if !self.source_code.line_text(start_index).trim().is_empty() {
|
|
break;
|
|
}
|
|
start_index = start_index.saturating_add(1);
|
|
}
|
|
|
|
let content_end_index = self.source_code.line_index(self.range.end());
|
|
let mut end_index = content_end_index
|
|
.saturating_add(2)
|
|
.min(OneIndexed::from_zero_indexed(self.source_code.line_count()));
|
|
|
|
// Trim trailing empty lines.
|
|
while end_index > content_end_index {
|
|
if !self.source_code.line_text(end_index).trim().is_empty() {
|
|
break;
|
|
}
|
|
|
|
end_index = end_index.saturating_sub(1);
|
|
}
|
|
|
|
let start_offset = self.source_code.line_start(start_index);
|
|
let end_offset = self.source_code.line_end(end_index);
|
|
|
|
let annotation_range = self.range - start_offset;
|
|
let source = self
|
|
.source_code
|
|
.slice(TextRange::new(start_offset, end_offset));
|
|
|
|
let start_char = source[TextRange::up_to(annotation_range.start())]
|
|
.chars()
|
|
.count();
|
|
|
|
let char_length = source[annotation_range].chars().count();
|
|
let label = format!("Syntax Error: {error}", error = self.error);
|
|
|
|
let snippet = Snippet {
|
|
title: None,
|
|
slices: vec![Slice {
|
|
source,
|
|
line_start: start_index.get(),
|
|
annotations: vec![SourceAnnotation {
|
|
label: &label,
|
|
annotation_type: AnnotationType::Error,
|
|
range: (start_char, start_char + char_length),
|
|
}],
|
|
// The origin (file name, line number, and column number) is already encoded
|
|
// in the `label`.
|
|
origin: None,
|
|
fold: false,
|
|
}],
|
|
footer: Vec::new(),
|
|
opt: FormatOptions::default(),
|
|
};
|
|
|
|
writeln!(f, "{message}", message = DisplayList::from(snippet))
|
|
}
|
|
}
|
|
|
|
/// Verifies that:
|
|
/// * the ranges are strictly increasing when loop the tokens in insertion order
|
|
/// * all ranges are within the length of the source code
|
|
fn validate_tokens(tokens: &[Token], source_length: TextSize, test_path: &Path) {
|
|
let mut previous: Option<&Token> = None;
|
|
|
|
for token in tokens {
|
|
assert!(
|
|
token.end() <= source_length,
|
|
"{path}: Token range exceeds the source code length. Token: {token:#?}",
|
|
path = test_path.display()
|
|
);
|
|
|
|
if let Some(previous) = previous {
|
|
assert_eq!(
|
|
previous.range().ordering(token.range()),
|
|
Ordering::Less,
|
|
"{path}: Token ranges are not in increasing order
|
|
Previous token: {previous:#?}
|
|
Current token: {token:#?}
|
|
Tokens: {tokens:#?}
|
|
",
|
|
path = test_path.display(),
|
|
);
|
|
}
|
|
|
|
previous = Some(token);
|
|
}
|
|
}
|
|
|
|
/// Verifies that:
|
|
/// * the range of the parent node fully encloses all its child nodes
|
|
/// * the ranges are strictly increasing when traversing the nodes in pre-order.
|
|
/// * all ranges are within the length of the source code.
|
|
fn validate_ast(root: &Mod, source_len: TextSize, test_path: &Path) {
|
|
walk_module(&mut ValidateAstVisitor::new(source_len, test_path), root);
|
|
}
|
|
|
|
#[derive(Debug)]
|
|
struct ValidateAstVisitor<'a> {
|
|
parents: Vec<AnyNodeRef<'a>>,
|
|
previous: Option<AnyNodeRef<'a>>,
|
|
source_length: TextSize,
|
|
test_path: &'a Path,
|
|
}
|
|
|
|
impl<'a> ValidateAstVisitor<'a> {
|
|
fn new(source_length: TextSize, test_path: &'a Path) -> Self {
|
|
Self {
|
|
parents: Vec::new(),
|
|
previous: None,
|
|
source_length,
|
|
test_path,
|
|
}
|
|
}
|
|
}
|
|
|
|
impl<'ast> SourceOrderVisitor<'ast> for ValidateAstVisitor<'ast> {
|
|
fn enter_node(&mut self, node: AnyNodeRef<'ast>) -> TraversalSignal {
|
|
assert!(
|
|
node.end() <= self.source_length,
|
|
"{path}: The range of the node exceeds the length of the source code. Node: {node:#?}",
|
|
path = self.test_path.display()
|
|
);
|
|
|
|
if let Some(previous) = self.previous {
|
|
assert_ne!(previous.range().ordering(node.range()), Ordering::Greater,
|
|
"{path}: The ranges of the nodes are not strictly increasing when traversing the AST in pre-order.\nPrevious node: {previous:#?}\n\nCurrent node: {node:#?}\n\nRoot: {root:#?}",
|
|
path = self.test_path.display(),
|
|
root = self.parents.first()
|
|
);
|
|
}
|
|
|
|
if let Some(parent) = self.parents.last() {
|
|
assert!(parent.range().contains_range(node.range()),
|
|
"{path}: The range of the parent node does not fully enclose the range of the child node.\nParent node: {parent:#?}\n\nChild node: {node:#?}\n\nRoot: {root:#?}",
|
|
path = self.test_path.display(),
|
|
root = self.parents.first()
|
|
);
|
|
}
|
|
|
|
self.parents.push(node);
|
|
|
|
TraversalSignal::Traverse
|
|
}
|
|
|
|
fn leave_node(&mut self, node: AnyNodeRef<'ast>) {
|
|
self.parents.pop().expect("Expected tree to be balanced");
|
|
|
|
self.previous = Some(node);
|
|
}
|
|
}
|