slint/internal/compiler/tests/syntax_tests.rs
Leon Matthes 077d3b6972
Some checks are pending
autofix.ci / format_fix (push) Waiting to run
autofix.ci / lint_typecheck (push) Waiting to run
autofix.ci / ci (push) Blocked by required conditions
Add length to spans and diagnostics (#9703)
This allows diagnostics to cover not just a single character, but the
entire relevant code region.

ChangeLog: Diagnostics now indicate the range of the error/warning, not just the starting character
2025-12-03 17:56:01 +01:00

573 lines
20 KiB
Rust

// Copyright © SixtyFPS GmbH <info@slint.dev>
// SPDX-License-Identifier: GPL-3.0-only OR LicenseRef-Slint-Royalty-free-2.0 OR LicenseRef-Slint-Software-3.0
//! This test is trying to compile all the *.slint files in the sub directories and check that compilation
//! errors are properly reported.
//!
//! The .slint files can have comments like this:
//! ```ignore
//! hi foo
//! // > <error{expected_message}
//! ```
//!
//! Meaning that there must an error with that error message spanning the characters on the line above between the `>` and `<` characters.
//! The `>` and `<` indicators may also appear individually, if the diagnostic spans multiple lines.
//! A `^` character means that the diagnostic must only span that single character.
//!
//! If there are additional `^` characters: `> <^error{expected_message}` then it means the comment refers to a diagnostic two lines above, instead of one, and so on with more carets.
//!
//! If there are additional `<` characters: `> <<error{expected_message}` then it means the range
//! should be an additional character to the left, which is useful if the diagnostic starts or ends
//! in the first or second column, where otherwise the `//` is located.
//!
//! `> <warning{expected_message}` is also supported.
//!
//! The newlines are replaced by `↵` in the error message. Also the manifest dir (CARGO_MANIFEST_DIR) is replaced by `📂`.
//!
//! When the env variable `SLINT_SYNTAX_TEST_UPDATE` is set to `1`, the source code will be modified to add the comments
//! The env variable `SLINT_TEST_FILTER` accepts a regexp and will filter out tests not maching that pattern
use i_slint_compiler::ComponentSelection;
use i_slint_compiler::diagnostics::{
self, BuildDiagnostics, ByteFormat, Diagnostic, DiagnosticLevel,
};
use std::ops::Range;
use std::path::{Path, PathBuf};
#[test]
fn syntax_tests() -> std::io::Result<()> {
use rayon::prelude::*;
let update = std::env::args().any(|arg| arg == "--update")
|| std::env::var("SLINT_SYNTAX_TEST_UPDATE").is_ok_and(|v| v == "1");
if let Some(specific_test) =
std::env::args().skip(1).find(|arg| !(arg.starts_with("--") || arg == "syntax_tests"))
{
let mut path = std::path::PathBuf::from(env!("CARGO_MANIFEST_DIR"));
path.push("tests");
path.push(specific_test);
assert!(process_file(&path, update)?);
return Ok(());
}
let pattern = std::env::var("SLINT_TEST_FILTER").ok().map(|p| regex::Regex::new(&p).unwrap());
let mut test_entries = Vec::new();
for entry in std::fs::read_dir(format!("{}/tests/syntax", env!("CARGO_MANIFEST_DIR")))? {
let entry = entry?;
if entry.file_type().is_ok_and(|f| f.is_dir()) {
let path = entry.path();
for test_entry in path.read_dir()? {
let test_entry = test_entry?;
let path = test_entry.path();
if let Some(ext) = path.extension() {
if (ext == "60" || ext == "slint")
&& pattern
.as_ref()
.map(|p| p.is_match(&path.to_string_lossy()))
.unwrap_or(true)
{
test_entries.push(path);
}
}
}
}
}
let success = test_entries
.par_iter()
.try_fold(
|| true,
|mut success, path| {
success &= process_file(path, update)?;
Ok::<bool, std::io::Error>(success)
},
)
.try_reduce(|| true, |success, result| Ok(success & result))?;
assert!(success);
Ok(())
}
fn process_file(path: &std::path::Path, update: bool) -> std::io::Result<bool> {
let source = std::fs::read_to_string(path)?;
if path.to_str().unwrap_or("").contains("bom-") && !source.starts_with("\u{FEFF}") {
// make sure that the file still contains BOM and it wasn't remove by some tools
return Err(std::io::Error::other(format!(
"{path:?} does not contains BOM while it should"
)));
}
std::panic::catch_unwind(|| process_file_source(path, source, false, update)).unwrap_or_else(
|err| {
println!("Panic while processing {}: {:?}", path.display(), err);
Ok(false)
},
)
}
struct ExpectedDiagnostic {
start: Option<usize>,
end: Option<usize>,
level: DiagnosticLevel,
message: String,
comment_range: Range<usize>,
}
fn extract_expected_diags(source: &str) -> Vec<ExpectedDiagnostic> {
let mut expected = Vec::new();
// Find expected errors in the file. The first caret (^) points to the expected column. The number of
// carets refers to the number of lines to go back. This is useful when one line of code produces multiple
// errors or warnings.
let re = regex::Regex::new(
r"\n *//[^\n\^<>]*((\^)|((>)?( *<)?))(\^*)(<*)(error|warning)\{([^\n]*)\}",
)
.unwrap();
// regex::Regex::new(r"\n *//[^\n\^<>]*(\^|>|<)(\^*)(error|warning)\{([^\n]*)\}").unwrap();
for m in re.captures_iter(source) {
let line_begin_offset = m.get(0).unwrap().start();
let start_column = m.get(1).unwrap().start()
- line_begin_offset
// Allow shifting columns with <
- m.get(7).map(|group| group.as_str().len()).unwrap_or_default();
let lines_to_source = m.get(6).map(|group| group.as_str().len()).unwrap_or_default() + 1;
let warning_or_error = m.get(8).unwrap().as_str();
let expected_message =
m.get(9).unwrap().as_str().replace('↵', "\n").replace('📂', env!("CARGO_MANIFEST_DIR"));
let comment_range = m.get(0).unwrap().range();
let mut line_counter = 0;
let mut line_offset = source[..line_begin_offset].rfind('\n').unwrap_or(0);
let mut offset = loop {
line_counter += 1;
if line_counter >= lines_to_source {
break line_offset + start_column;
}
if let Some(o) = source[..line_offset].rfind('\n') {
line_offset = o;
} else {
break 1;
};
};
let mut start = None;
let mut end = None;
if m.get(2).is_some() {
// ^warning{...}
start = Some(offset);
end = Some(offset);
} else {
// >
if m.get(4).is_some() {
start = Some(offset);
offset += 1;
}
// < (including spaces before)
if let Some(range_length) = m.get(5).map(|group| group.as_str().len()) {
end = Some(offset + range_length - 1);
}
}
// Windows edge-case, if the end falls on a newline, it should span the entire
// newline character, which is two characters, not one.
if let Some(end_offset) = end {
if source.get(end_offset..=(end_offset + 1)) == Some("\r\n") {
end = Some(end_offset + 1)
};
}
let expected_diag_level = match warning_or_error {
"warning" => DiagnosticLevel::Warning,
"error" => DiagnosticLevel::Error,
_ => panic!("Unsupported diagnostic level {warning_or_error}"),
};
expected.push(ExpectedDiagnostic {
start,
end,
level: expected_diag_level,
message: expected_message,
comment_range,
});
}
expected
}
fn process_diagnostics(
compile_diagnostics: &BuildDiagnostics,
path: &Path,
source: &str,
_silent: bool,
update: bool,
) -> std::io::Result<bool> {
let mut success = true;
let path = canonical(path);
let diags = compile_diagnostics
.iter()
.filter(|d| {
canonical(
d.source_file()
.unwrap_or_else(|| panic!("{path:?}: Error without a source file {d:?}",)),
) == path
})
.collect::<Vec<_>>();
let lines = source
.bytes()
.enumerate()
.filter_map(|(i, c)| if c == b'\n' { Some(i) } else { None })
.collect::<Vec<usize>>();
let mut expected = extract_expected_diags(source);
let captures: Vec<_> =
expected.iter().map(|expected| &expected.comment_range).cloned().collect();
// Find expected errors in the file. The first caret (^) points to the expected column. The number of
// carets refers to the number of lines to go back. This is useful when one line of code produces multiple
// errors or warnings.
for diag in &diags {
fn compare_message(message: &str, expected_message: &str) -> bool {
if message == expected_message {
return true;
}
// The error message might contain path that might have other character, so replace them on windows
#[cfg(target_os = "windows")]
if message.replace('\\', "/") == expected_message.replace('\\', "/") {
return true;
}
false
}
let (l, c) = diag.line_column();
let diag_start = lines.get(l.wrapping_sub(2)).unwrap_or(&0) + c;
// end_line_column is not (yet) available via the public API, so use the private API
// instead.
let (l, c) = diagnostics::diagnostic_end_line_column_with_format(diag, ByteFormat::Utf8);
let diag_end = lines.get(l.wrapping_sub(2)).unwrap_or(&0) + c - 1;
let expected_start = expected.iter().position(|expected| {
Some(diag_start) == expected.start
&& compare_message(diag.message(), &expected.message)
&& diag.level() == expected.level
&& (expected.end.is_none() || Some(diag_end) == expected.end)
});
let expected_end = expected.iter().position(|expected| {
Some(diag_end) == expected.end
&& compare_message(diag.message(), &expected.message)
&& diag.level() == expected.level
&& (expected.start.is_none() || Some(diag_start) == expected.start)
});
let found_match = match (expected_start, expected_end) {
(Some(start), Some(end)) => {
// Found both start and end, success!
// Make sure to remove the larger index first, so the
// smaller index remains valid.
expected.remove(start.max(end));
if start != end {
expected.remove(start.min(end));
}
true
}
(Some(start), None) => {
println!("{path:?}: Could not find end of error/warning: {diag:#?}");
expected.remove(start);
false
}
(None, Some(end)) => {
println!("{path:?}: Could not find start of error/warning: {diag:#?}");
expected.remove(end);
false
}
// TODO: Remove start/end if only one was found
(None, None) => {
println!("{path:?}: Unexpected error/warning: {diag:#?}, {diag_start}, {diag_end}",);
false
}
};
if !found_match {
success = false;
#[cfg(feature = "display-diagnostics")]
if !_silent {
let mut to_report = BuildDiagnostics::default();
to_report.push_compiler_error((*diag).clone());
to_report.print();
}
}
}
for expected in expected {
success = false;
println!(
"{path:?}: {level:?} not found at offset {start:?}-{end:?}: {message:?}",
level = expected.level,
start = expected.start,
end = expected.end,
message = expected.message
);
}
if !success && update {
let mut source = source.to_string();
self::update(&diags, &mut source, lines, &captures);
std::fs::write(path, source).unwrap();
}
Ok(success)
}
/// Rewrite the source to remove the old comments and add accurate error comments
fn update(
diags: &[&Diagnostic],
source: &mut String,
mut lines: Vec<usize>,
to_remove: &[std::ops::Range<usize>],
) {
for to_remove in to_remove.iter().rev() {
source.drain(to_remove.clone());
for l in &mut lines {
if *l > to_remove.start {
*l -= to_remove.end - to_remove.start;
}
}
}
let mut last_line_adjust = Vec::from_iter(std::iter::repeat_n(0, lines.len()));
for d in diags {
let mut insert_range_at = |range: &str, l, c: usize| {
let column_adjust = if c < 3 { "<".repeat(3 - c) } else { "".to_string() };
let byte_offset = lines[l - 1] + 1;
let to_insert = format!(
"//{indent}{range}{adjust}{column_adjust}{error_or_warning}{{{message}}}\n",
indent = " ".repeat(c.max(3) - 3),
adjust = "^".repeat(last_line_adjust[l - 1]),
error_or_warning =
if d.level() == DiagnosticLevel::Error { "error" } else { "warning" },
message = d.message().replace('\n', "").replace(env!("CARGO_MANIFEST_DIR"), "📂")
);
if byte_offset > source.len() {
source.push('\n');
}
source.insert_str(byte_offset, &to_insert);
for line in (l - 1)..lines.len() {
lines[line] += to_insert.len();
}
last_line_adjust[l - 1] += 1;
};
let (line_start, column_start) = d.line_column();
// end_line_column is not (yet) available via the public API, so use the private API
// instead.
let (line_end, column_end) =
diagnostics::diagnostic_end_line_column_with_format(d, ByteFormat::Utf8);
// The end column is exclusive, therefore use - 1 here
let range = if d.length() <= 1 {
// Single-character diagnostic, use "^" for the marker
"^".to_owned()
} else {
let end = if line_start == line_end {
// Same line, we can insert the closing "<"
" ".repeat(column_end - column_start - 2) + "<"
} else {
// End is on a different line, we'll emit it later
"".to_owned()
};
format!(">{end}")
};
insert_range_at(&range, line_start, column_start);
// Insert the closing `<` at another line if necessary
// Edge-case: If a single-character diagnostic is on a newline character (\n), its
// end_line_column is technically on a new line, but the single ^ marker is enough, so no
// closing character is needed.
if line_start != line_end && d.length() > 1 {
insert_range_at("<", line_end, column_end - 1);
}
}
}
fn canonical(path: &Path) -> PathBuf {
path.canonicalize().unwrap_or_else(|_| path.to_owned())
}
fn process_file_source(
path: &std::path::Path,
source: String,
silent: bool,
update: bool,
) -> std::io::Result<bool> {
let mut parse_diagnostics = BuildDiagnostics::default();
let syntax_node =
i_slint_compiler::parser::parse(source.clone(), Some(path), &mut parse_diagnostics);
let has_parse_error = parse_diagnostics.has_errors();
let mut compiler_config = i_slint_compiler::CompilerConfiguration::new(
i_slint_compiler::generator::OutputFormat::Interpreter,
);
compiler_config.library_paths = [(
"test-lib".into(),
concat!(env!("CARGO_MANIFEST_DIR"), "/tests/typeloader/library").into(),
)]
.into_iter()
.collect();
compiler_config.embed_resources = i_slint_compiler::EmbedResourcesKind::OnlyBuiltinResources;
compiler_config.enable_experimental = true;
compiler_config.style = Some("fluent".into());
compiler_config.components_to_generate =
if source.contains("config:generate_all_exported_windows") {
ComponentSelection::ExportedWindows
} else {
// Otherwise we'd have lots of warnings about not inheriting Window
ComponentSelection::LastExported
};
let compile_diagnostics = if !parse_diagnostics.has_errors() {
let (_, build_diags, _) = spin_on::spin_on(i_slint_compiler::compile_syntax_node(
syntax_node.clone(),
parse_diagnostics,
compiler_config.clone(),
));
build_diags
} else {
parse_diagnostics
};
let mut success = true;
success &= process_diagnostics(&compile_diagnostics, path, &source, silent, update)?;
for p in &compile_diagnostics.all_loaded_files {
let source = if p.is_absolute() {
std::fs::read_to_string(p)?
} else {
// probably std-widgets.slint
String::new()
};
success &= process_diagnostics(&compile_diagnostics, p, &source, silent, update)?;
}
if has_parse_error {
// Still try to compile to make sure it doesn't panic
spin_on::spin_on(i_slint_compiler::compile_syntax_node(
syntax_node,
compile_diagnostics,
compiler_config,
));
}
Ok(success)
}
#[test]
/// Test that this actually fail when it should
fn self_test() -> std::io::Result<()> {
let fake_path = std::path::Path::new("fake.slint");
let process = |str: &str| process_file_source(fake_path, str.into(), true, false);
// this should succeed
assert!(process(
r#"
export component Foo inherits Window { width: 10px; }
"#
)?);
// unless we expected an error
assert!(!process(
r#"
export component Foo inherits Window { width: 10px; }
// ^error{i want an error}
"#
)?);
// An error should fail
assert!(!process(
r#"
export component Foo inherits Window foo { width: 10px; }
"#
)?);
// An error with the proper comment should pass
assert!(process(
r#"
export component Foo inherits Window foo { width: 10px; }
// > <error{Syntax error: expected '{'}
"#
)?);
// also when it's shifted up by an additional ^
assert!(process(
r#"
export component Foo inherits Window foo { width: 10px; }
// > <^error{Syntax error: expected '{'}
"#
)?);
// also when it's shifted left by additional <
assert!(process(
r#"
export component Foo inherits Window foo { width: 10px; }
// > <<<<error{Syntax error: expected '{'}
"#
)?);
// or split into multiple lines
assert!(process(
r#"
export component Foo inherits Window foo { width: 10px; }
// >error{Syntax error: expected '{'}
// <^error{Syntax error: expected '{'}
"#
)?);
// But not if it is at the wrong position
assert!(!process(
r#"
export component Foo inherits Window foo { width: 10px; }
// > <error{Syntax error: expected '{'}
"#
)?);
// or the wrong line
assert!(!process(
r#"
export component Foo inherits Window foo { width: 10px; }
// > <error{Syntax error: expected '{'}
"#
)?);
// or the wrong message
assert!(!process(
r#"
export component Foo inherits Window foo { width: 10px; }
// > <error{foo_bar}
"#
)?);
// or the wrong line because two carets
assert!(!process(
r#"
export component Foo inherits Window foo { width: 10px; }
// > <^^error{Syntax error: expected '{'}
"#
)?);
// Even on windows, it should work
assert!(process(
"\r\n\
export component Foo inherits Window foo { width: 10px; }\r\n\
// > <error{Syntax error: expected '{'}\r\n\
"
)?);
Ok(())
}