M2.3: find in files backend

crates/search: parallel ignore walk + grep-searcher with the project's
ignore rules (vendor searched as library, node_modules and .git skipped),
literal or regex, case, whole word, include/exclude globs, binary and
> 8 MiB skips, match cap, cancel token, buffer overlay.

Engine: search.find returns {searchId} at once and runs off the lock;
pump_vfs drains results into searchResults (one per file, at most 500
per pump) and searchDone; search.cancel. Testkit: generated search tree.
Tests: library unit and property tests, engine tests, TCP control test.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
danny 2026-10-16 22:03:17 +02:00
parent a6b2fd1300
commit 172693c8dd
21 changed files with 2231 additions and 5 deletions

View file

@ -0,0 +1,170 @@
//! Find in files over a real TCP control connection: `search.find` answers
//! at once, `searchResults` and `searchDone` arrive as event lines from the
//! pump, `search.cancel` ends a search.
#![allow(
clippy::unwrap_used,
clippy::expect_used,
clippy::indexing_slicing,
clippy::panic
)]
use std::io::{BufRead, BufReader, Write};
use std::net::TcpStream;
use std::sync::Arc;
use std::sync::atomic::{AtomicBool, Ordering};
use std::time::Duration;
use lutra_testkit::EngineBackend;
use serde_json::{Value, json};
/// A client that keeps responses and event lines apart.
struct Client {
reader: BufReader<TcpStream>,
writer: TcpStream,
events: Vec<Value>,
}
impl Client {
fn connect(addr: std::net::SocketAddr) -> Self {
let stream = TcpStream::connect(addr).unwrap();
stream
.set_read_timeout(Some(Duration::from_secs(10)))
.unwrap();
Self {
writer: stream.try_clone().unwrap(),
reader: BufReader::new(stream),
events: Vec::new(),
}
}
fn line(&mut self) -> Value {
let mut line = String::new();
self.reader.read_line(&mut line).expect("line within 10 s");
serde_json::from_str(&line).unwrap()
}
fn call(&mut self, request: Value) -> Value {
self.writer
.write_all(format!("{request}\n").as_bytes())
.unwrap();
loop {
let line = self.line();
if line.get("event").is_some() {
self.events.push(line["event"].clone());
} else {
return line;
}
}
}
/// Reads event lines until `searchDone` for `id`; returns the
/// `searchResults` of that search and the done event.
fn until_done(&mut self, id: u64) -> (Vec<Value>, Value) {
let mut results = Vec::new();
let mut pending = std::mem::take(&mut self.events).into_iter();
loop {
let event = match pending.next() {
Some(e) => e,
None => {
let line = self.line();
let Some(event) = line.get("event") else {
continue;
};
event.clone()
}
};
if event["searchId"] != id {
continue;
}
match event["kind"].as_str() {
Some("searchResults") => results.push(event),
Some("searchDone") => return (results, event),
_ => {}
}
}
}
}
fn execute(command: &str, params: Value) -> Value {
json!({"id": command, "method": "engine.execute", "params": {"command": command, "params": params}})
}
/// Pumps the backend every 20 ms, like the desktop's pump thread, until
/// dropped.
struct Pump(Arc<AtomicBool>);
impl Pump {
fn start(backend: Arc<EngineBackend>) -> Self {
let stop = Arc::new(AtomicBool::new(false));
let flag = stop.clone();
std::thread::spawn(move || {
while !flag.load(Ordering::Relaxed) {
backend.pump();
std::thread::sleep(Duration::from_millis(20));
}
});
Self(stop)
}
}
impl Drop for Pump {
fn drop(&mut self) {
self.0.store(true, Ordering::Relaxed);
}
}
#[test]
fn find_streams_results_then_done_and_cancels() {
let dir = tempfile::tempdir().unwrap();
let tree = lutra_testkit::generate_search_tree(dir.path(), 1_000);
let backend = Arc::new(EngineBackend::new());
let server = lutra_control::start(0, backend.clone()).unwrap();
let mut client = Client::connect(server.local_addr());
client.call(json!({"id": 1, "method": "events.subscribe"}));
let r = client.call(execute("project.open", json!({"path": dir.path()})));
assert!(r["result"]["root"].is_string(), "{r}");
let _pump = Pump::start(backend.clone());
let r = client.call(execute(
"search.find",
json!({"query": lutra_testkit::NEEDLE, "caseSensitive": true}),
));
let id = r["result"]["searchId"].as_u64().expect("searchId");
let (results, done) = client.until_done(id);
assert_eq!(results.len(), tree.needle_files);
for result in &results {
let m = &result["matches"][0];
assert!(m["line"].is_u64() && m["column"].is_u64(), "{result}");
assert!(
m["preview"]
.as_str()
.unwrap()
.contains(lutra_testkit::NEEDLE)
);
assert!(m["previewColumn"].is_u64());
let path = result["path"].as_str().unwrap();
assert!(!path.starts_with("var/") && !path.contains('\\'), "{path}");
}
assert_eq!(done["files"], tree.needle_files);
assert_eq!(done["cancelled"], false);
assert_eq!(done["truncated"], false);
assert!(done["elapsedMs"].is_u64());
let r = client.call(execute("search.find", json!({"query": "(", "regex": true})));
assert_eq!(r["error"]["code"], "invalid_params", "{r}");
let r = client.call(execute(
"search.find",
json!({"query": "function", "maxResults": 100000}),
));
let id = r["result"]["searchId"].as_u64().unwrap();
let r = client.call(execute("search.cancel", json!({"searchId": id})));
assert!(r["result"]["cancelled"].is_boolean(), "{r}");
let (_, done) = client.until_done(id);
assert_eq!(done["kind"], "searchDone");
assert!(
done["cancelled"] == true || r["result"]["cancelled"] == false,
"{done}"
);
}

View file

@ -12,6 +12,7 @@ publish.workspace = true
layer = 4
[dependencies]
lutra-search.workspace = true
lutra-text.workspace = true
lutra-vfs.workspace = true
serde = { version = "1.0.229", features = ["derive"] }
@ -22,5 +23,6 @@ thiserror = "2.0.21"
workspace = true
[dev-dependencies]
lutra-testkit.workspace = true
proptest.workspace = true
tempfile.workspace = true

View file

@ -6,6 +6,7 @@ mod file;
mod params;
mod project;
mod recovery;
pub mod search;
mod settings;
pub mod theme;
pub mod ui;
@ -39,5 +40,7 @@ pub fn register_builtins(registry: &mut Registry) -> Result<(), EngineError> {
registry.register(Box::new(theme::List))?;
registry.register(Box::new(theme::SystemPreference))?;
registry.register(Box::new(ui::RegisterCommands))?;
registry.register(Box::new(search::Find))?;
registry.register(Box::new(search::Cancel))?;
Ok(())
}

View file

@ -0,0 +1,304 @@
//! Find in files: `search.find` and `search.cancel`.
//!
//! `search.find` compiles the query, takes a snapshot of the dirty buffers
//! and starts the search on its own threads ([`lutra_search::spawn`]), then
//! returns `{searchId}` at once; the engine lock is never held while files
//! are read. Results wait in a channel per search until
//! [`crate::Engine::pump_vfs`] calls [`pump`], which turns them into
//! [`Event::SearchResults`] (one per file) and a final
//! [`Event::SearchDone`]. Hosts already publish the pump's events, so
//! results reach the window and control-channel subscribers in the
//! engine's event order, at most one batch per pump (every 100 ms on the
//! desktop) and at most [`MAX_RESULTS_PER_PUMP`] files per batch, so a
//! subscriber's queue never overflows. Searches whose project is no longer
//! open are cancelled by the next pump and end with `cancelled: true`.
use std::collections::BTreeMap;
use std::path::{Component, PathBuf};
use std::sync::mpsc::{Receiver, TryRecvError};
use std::time::Instant;
use lutra_search::{CancelToken, Overlay, Query, Search, SearchEvent};
use serde::Deserialize;
use serde_json::{Value, json};
use crate::cmd::params::parse;
use crate::{Command, CommandSpec, EngineError, Event, Outcome, SearchMatch, State};
/// Most `searchResults` events one pump emits, over all searches; the rest
/// wait for the next pump.
pub const MAX_RESULTS_PER_PUMP: usize = 500;
/// Most include or exclude globs one query may carry.
pub const MAX_GLOBS: usize = 256;
/// The running searches.
#[derive(Debug, Default)]
pub struct Searches {
/// The id of the last search started; ids are never reused.
last_id: u64,
/// Searches whose `searchDone` has not been emitted yet, by id.
running: BTreeMap<u64, Running>,
}
impl Searches {
/// Number of searches whose `searchDone` is still to come.
pub fn running(&self) -> usize {
self.running.len()
}
}
/// One search in flight.
#[derive(Debug)]
struct Running {
/// The project root the search walks.
root: PathBuf,
/// Stops it.
cancel: CancelToken,
/// Its results and summary.
events: Receiver<SearchEvent>,
/// When it started, for a summary the search thread never sent.
started: Instant,
/// Files emitted so far.
files: usize,
/// Matches emitted so far.
matches: usize,
}
/// Params of `search.find`.
#[derive(Deserialize)]
#[serde(deny_unknown_fields, rename_all = "camelCase")]
struct FindParams {
/// The text or regular expression.
query: String,
/// `query` is a regular expression.
#[serde(default)]
regex: bool,
/// Match letter case exactly.
#[serde(default)]
case_sensitive: bool,
/// Whole words only.
#[serde(default)]
whole_word: bool,
/// Only files matching one of these globs.
#[serde(default)]
include: Vec<String>,
/// Skip files and directories matching one of these globs.
#[serde(default)]
exclude: Vec<String>,
/// Cap on matches.
max_results: Option<usize>,
}
/// `search.find`: starts a search in the project.
pub struct Find;
impl Command for Find {
fn spec(&self) -> &CommandSpec {
&CommandSpec {
id: "search.find",
label: "Find in Files",
menu_path: None,
default_shortcut: None,
params: "{ query, regex?, caseSensitive?, wholeWord?, include?: [glob], exclude?: [glob], maxResults? (default 10000, at most 100000) }; returns { searchId } at once, then events searchResults { searchId, path, matches: [{ line, column, length, preview, previewColumn }] } per file and searchDone { searchId, files, matches, truncated, cancelled, elapsedMs }; invalid_params for an empty query, a bad regex or glob",
}
}
fn enabled(&self, state: &State) -> bool {
state.has_project()
}
fn run(&self, state: &mut State, params: Value) -> Result<Outcome, EngineError> {
let id = self.spec().id;
let p: FindParams = parse(id, params)?;
if p.include.len() > MAX_GLOBS || p.exclude.len() > MAX_GLOBS {
return Err(EngineError::invalid_params(
id,
format!("at most {MAX_GLOBS} include and {MAX_GLOBS} exclude globs"),
));
}
let query = Query {
pattern: p.query,
regex: p.regex,
case_sensitive: p.case_sensitive,
whole_word: p.whole_word,
include: p.include,
exclude: p.exclude,
max_results: p.max_results.unwrap_or(lutra_search::DEFAULT_MAX_RESULTS),
};
let search =
Search::new(&query).map_err(|e| EngineError::invalid_params(id, e.to_string()))?;
let root = state
.project
.as_ref()
.ok_or_else(|| EngineError::Disabled(id.to_string()))?
.project
.root()
.to_path_buf();
let cancel = CancelToken::new();
let events = lutra_search::spawn(root.clone(), search, overlay(state), cancel.clone())
.map_err(|e| EngineError::Failed(format!("cannot start the search: {e}")))?;
let searches = &mut state.searches;
searches.last_id = searches.last_id.saturating_add(1);
let search_id = searches.last_id;
searches.running.insert(
search_id,
Running {
root,
cancel,
events,
started: Instant::now(),
files: 0,
matches: 0,
},
);
Ok(Outcome::value(json!({ "searchId": search_id })))
}
}
/// Params of `search.cancel`.
#[derive(Deserialize)]
#[serde(deny_unknown_fields, rename_all = "camelCase")]
struct CancelParams {
/// The search to stop.
search_id: u64,
}
/// `search.cancel`: stops a running search.
pub struct Cancel;
impl Command for Cancel {
fn spec(&self) -> &CommandSpec {
&CommandSpec {
id: "search.cancel",
label: "Cancel Search",
menu_path: None,
default_shortcut: None,
params: "{ searchId }; returns { cancelled }: false if the search already ended or never existed; a cancelled search still ends with searchDone (cancelled: true)",
}
}
fn run(&self, state: &mut State, params: Value) -> Result<Outcome, EngineError> {
let p: CancelParams = parse(self.spec().id, params)?;
let running = state.searches.running.get(&p.search_id);
if let Some(running) = running {
running.cancel.cancel();
}
Ok(Outcome::value(json!({ "cancelled": running.is_some() })))
}
}
/// The text of every dirty buffer by project-relative path, searched
/// instead of the file on disk. A path that is not plain Unicode cannot
/// match a walked file and is left out.
fn overlay(state: &State) -> Overlay {
state
.buffers
.dirty_ids()
.into_iter()
.filter_map(|id| state.buffers.get(id))
.filter_map(|b| Some((slash_path(&b.path)?, b.buffer.text())))
.collect()
}
/// A project-relative path with `/` separators.
fn slash_path(path: &std::path::Path) -> Option<String> {
let parts: Option<Vec<&str>> = path
.components()
.map(|c| match c {
Component::Normal(name) => name.to_str(),
_ => None,
})
.collect();
Some(parts?.join("/"))
}
/// Moves finished results of running searches into events: at most
/// [`MAX_RESULTS_PER_PUMP`] `searchResults`, and `searchDone` for every
/// search that ended. Searches of a project that is no longer open are
/// cancelled; their remaining results are dropped. A search thread that
/// vanished without a summary ends with `cancelled: true` and the counts
/// emitted so far.
pub fn pump(state: &mut State) {
let root = state
.project
.as_ref()
.map(|w| w.project.root().to_path_buf());
let mut budget = MAX_RESULTS_PER_PUMP;
let mut events = Vec::new();
let mut finished = Vec::new();
for (&search_id, running) in state.searches.running.iter_mut() {
let stale = root.as_ref() != Some(&running.root);
if stale {
running.cancel.cancel();
}
loop {
if budget == 0 && !stale {
break;
}
match running.events.try_recv() {
Ok(SearchEvent::File(file)) => {
if stale {
continue;
}
budget = budget.saturating_sub(1);
running.files = running.files.saturating_add(1);
running.matches = running.matches.saturating_add(file.matches.len());
events.push(Event::SearchResults {
search_id,
path: file.path,
matches: file.matches.into_iter().map(SearchMatch::from).collect(),
});
}
Ok(SearchEvent::Done(summary)) => {
events.push(Event::SearchDone {
search_id,
files: summary.files,
matches: summary.matches,
truncated: summary.truncated,
cancelled: summary.cancelled || stale,
elapsed_ms: millis(summary.elapsed),
});
finished.push(search_id);
break;
}
Err(TryRecvError::Empty) => break,
Err(TryRecvError::Disconnected) => {
events.push(Event::SearchDone {
search_id,
files: running.files,
matches: running.matches,
truncated: false,
cancelled: true,
elapsed_ms: millis(running.started.elapsed()),
});
finished.push(search_id);
break;
}
}
}
}
for search_id in finished {
state.searches.running.remove(&search_id);
}
for event in events {
state.emit(event);
}
}
/// Whole milliseconds of `elapsed`, saturating.
fn millis(elapsed: std::time::Duration) -> u64 {
u64::try_from(elapsed.as_millis()).unwrap_or(u64::MAX)
}
impl From<lutra_search::Match> for SearchMatch {
fn from(m: lutra_search::Match) -> Self {
Self {
line: m.line,
column: m.column,
length: m.length,
preview: m.preview,
preview_column: m.preview_column,
}
}
}

View file

@ -130,7 +130,8 @@ impl Engine {
/// Applies pending changes on disk (reloads clean buffers, reports
/// changed directories and conflicts), writes due recovery-journal
/// entries (see [`crate::journal`]) and returns every queued event.
/// entries (see [`crate::journal`]), moves finished search results into
/// events (see [`cmd::search`]) and returns every queued event.
/// Hosts call it regularly, e.g. every 100 ms from a background thread.
///
/// # Errors
@ -141,6 +142,7 @@ impl Engine {
guard::run(|| {
crate::workspace::pump(state);
crate::journal::tick(state, std::time::Instant::now());
cmd::search::pump(state);
Ok(())
})?;
Ok(state.take_events())
@ -373,6 +375,8 @@ mod tests {
"recovery.discard",
"recovery.list",
"recovery.restore",
"search.cancel",
"search.find",
"settings.get",
"settings.list",
"settings.reset",

View file

@ -144,6 +144,49 @@ pub enum Event {
/// Project-relative paths with `/`, sorted.
paths: Vec<String>,
},
/// Matches of a running search (`search.find`) in one file. A file is
/// reported once per search; files arrive in no particular order.
SearchResults {
/// The search, as `search.find` returned it.
search_id: u64,
/// Project-relative path of the file.
path: String,
/// The matches in line order; never empty.
matches: Vec<SearchMatch>,
},
/// A search ended; no `searchResults` for it follow.
SearchDone {
/// The search.
search_id: u64,
/// Files with reported matches.
files: usize,
/// Matches reported.
matches: usize,
/// The match cap stopped the search; more matches exist.
truncated: bool,
/// `search.cancel` (or closing the project) stopped the search.
cancelled: bool,
/// Wall time of the search in milliseconds.
elapsed_ms: u64,
},
}
/// One match of [`Event::SearchResults`]. Positions are 0-based; columns
/// and lengths count UTF-16 code units.
#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct SearchMatch {
/// Line of the match.
pub line: usize,
/// Start of the match in the line.
pub column: usize,
/// Length of the match; never 0.
pub length: usize,
/// The line without its line break, or a window of at most 300
/// characters around the match for longer lines.
pub preview: String,
/// Start of the match inside `preview`.
pub preview_column: usize,
}
/// Severity of an [`Event::Notification`].
@ -224,6 +267,39 @@ mod tests {
);
}
#[test]
fn search_shapes() {
let e = Event::SearchResults {
search_id: 2,
path: "src/a.php".into(),
matches: vec![super::SearchMatch {
line: 0,
column: 6,
length: 4,
preview: "<?php echo".into(),
preview_column: 6,
}],
};
assert_eq!(
serde_json::to_value(e).unwrap(),
json!({"kind": "searchResults", "searchId": 2, "path": "src/a.php", "matches": [
{"line": 0, "column": 6, "length": 4, "preview": "<?php echo", "previewColumn": 6}
]})
);
let e = Event::SearchDone {
search_id: 2,
files: 1,
matches: 1,
truncated: false,
cancelled: false,
elapsed_ms: 12,
};
assert_eq!(
serde_json::to_value(e).unwrap(),
json!({"kind": "searchDone", "searchId": 2, "files": 1, "matches": 1, "truncated": false, "cancelled": false, "elapsedMs": 12})
);
}
#[test]
fn ui_command_shape() {
let e = Event::UiCommand {

View file

@ -23,6 +23,6 @@ pub use buffers::{BufferId, LineEndingName, OpenBuffer};
pub use command::{Command, CommandInfo, CommandSpec, Effect, Outcome};
pub use engine::{Engine, EngineConfig};
pub use error::{EngineError, ErrorPayload};
pub use event::{Event, NotificationLevel};
pub use event::{Event, NotificationLevel, SearchMatch};
pub use registry::Registry;
pub use state::State;

View file

@ -35,6 +35,8 @@ pub struct State {
pub engine_commands: BTreeSet<&'static str>,
/// UI commands the frontend registered (`ui.registerCommands`), by id.
pub ui_commands: BTreeMap<String, UiCommand>,
/// Running searches (`search.find`).
pub searches: crate::cmd::search::Searches,
/// Number of notifications emitted, for their ids.
notifications: u64,
/// Events queued for the host, oldest first.

View file

@ -0,0 +1,207 @@
//! `search.find` and `search.cancel`: results arrive through the pump.
#![allow(
clippy::unwrap_used,
clippy::expect_used,
clippy::panic,
clippy::indexing_slicing
)]
use std::fs;
use std::time::{Duration, Instant};
use lutra_engine::cmd::search::MAX_RESULTS_PER_PUMP;
use lutra_engine::{Engine, Event};
use serde_json::{Value, json};
use tempfile::TempDir;
/// A temp folder with `files`, opened in a fresh engine.
fn open_project(files: &[(&str, &str)]) -> (TempDir, Engine) {
let dir = tempfile::tempdir().unwrap();
for (path, text) in files {
let abs = dir.path().join(path);
fs::create_dir_all(abs.parent().unwrap()).unwrap();
fs::write(abs, text).unwrap();
}
let mut engine = Engine::new().unwrap();
engine
.execute("project.open", json!({"path": dir.path()}))
.unwrap();
(dir, engine)
}
fn find(engine: &mut Engine, params: Value) -> u64 {
let out = engine.execute("search.find", params).unwrap();
assert!(out.events.is_empty(), "results only come from the pump");
out.value["searchId"].as_u64().unwrap()
}
/// Pumps until `searchDone` for `id`; returns the results (sorted by path)
/// and the done event, and the number of pumps.
fn collect(engine: &mut Engine, id: u64) -> (Vec<Event>, Event, usize) {
let deadline = Instant::now() + Duration::from_secs(10);
let mut results = Vec::new();
let mut pumps = 0;
while Instant::now() < deadline {
pumps += 1;
for event in engine.pump_vfs().unwrap() {
match &event {
Event::SearchResults { search_id, .. } if *search_id == id => results.push(event),
Event::SearchDone { search_id, .. } if *search_id == id => {
results.sort_by_key(path);
return (results, event, pumps);
}
_ => {}
}
}
std::thread::sleep(Duration::from_millis(10));
}
panic!("no searchDone within 10 s");
}
fn path(event: &Event) -> String {
match event {
Event::SearchResults { path, .. } => path.clone(),
_ => String::new(),
}
}
#[test]
fn finds_in_files_and_dirty_buffers() {
let (_dir, mut engine) = open_project(&[
("src/a.php", "<?php\necho 'needle';\n"),
("src/b.php", "<?php\n"),
("vendor/x/c.php", "<?php // needle\n"),
("node_modules/d.js", "needle\n"),
]);
let out = engine
.execute("file.open", json!({"path": "src/b.php"}))
.unwrap();
let buffer = out.value["id"].as_u64().unwrap();
engine
.execute(
"buffer.edit",
json!({"id": buffer, "baseVersion": 1, "edits": [{"from": 6, "to": 6, "insert": "$needle = 1;\n"}]}),
)
.unwrap();
let id = find(&mut engine, json!({"query": "NEEDLE"}));
let (results, done, _) = collect(&mut engine, id);
let paths: Vec<String> = results.iter().map(path).collect();
assert_eq!(paths, ["src/a.php", "src/b.php", "vendor/x/c.php"]);
let Event::SearchResults { matches, .. } = &results[1] else {
panic!()
};
assert_eq!(
(matches[0].line, matches[0].column, matches[0].length),
(1, 1, 6)
);
assert_eq!(matches[0].preview, "$needle = 1;");
let Event::SearchDone {
files,
matches,
truncated,
cancelled,
..
} = done
else {
panic!()
};
assert_eq!((files, matches, truncated, cancelled), (3, 3, false, false));
let id2 = find(
&mut engine,
json!({"query": "needle", "caseSensitive": true, "include": ["src/*.php"]}),
);
assert!(id2 > id);
let (results, _, _) = collect(&mut engine, id2);
assert_eq!(results.len(), 2);
}
#[test]
fn invalid_queries_fail_with_invalid_params() {
let (_dir, mut engine) = open_project(&[("a.php", "x")]);
for params in [
json!({"query": "(", "regex": true}),
json!({"query": ""}),
json!({"query": "x", "include": ["["]}),
json!({"query": "x", "bogus": 1}),
json!({}),
] {
let err = engine.execute("search.find", params.clone()).unwrap_err();
assert_eq!(err.code(), "invalid_params", "{params}");
}
assert_eq!(engine.state().searches.running(), 0);
engine.execute("project.close", json!({})).unwrap();
let err = engine
.execute("search.find", json!({"query": "x"}))
.unwrap_err();
assert_eq!(err.code(), "disabled");
}
#[test]
fn cancel_and_project_close_end_searches() {
let dir = tempfile::tempdir().unwrap();
lutra_testkit::generate_search_tree(dir.path(), 2_000);
let mut engine = Engine::new().unwrap();
engine
.execute("project.open", json!({"path": dir.path()}))
.unwrap();
let id = find(&mut engine, json!({"query": "function"}));
let out = engine
.execute("search.cancel", json!({"searchId": id}))
.unwrap();
assert_eq!(out.value["cancelled"], true);
let (_, done, _) = collect(&mut engine, id);
let Event::SearchDone {
cancelled, matches, ..
} = done
else {
panic!()
};
assert!(
cancelled || matches == 10_000,
"cancelled unless it had already finished"
);
let out = engine
.execute("search.cancel", json!({"searchId": id}))
.unwrap();
assert_eq!(out.value["cancelled"], false);
let id = find(&mut engine, json!({"query": "function"}));
engine.execute("project.close", json!({})).unwrap();
let (results, done, _) = collect(&mut engine, id);
assert!(results.is_empty());
assert!(matches!(
done,
Event::SearchDone {
cancelled: true,
..
}
));
assert_eq!(engine.state().searches.running(), 0);
}
#[test]
fn pumps_spread_many_results() {
let dir = tempfile::tempdir().unwrap();
for i in 0..(MAX_RESULTS_PER_PUMP + 100) {
fs::write(dir.path().join(format!("f{i}.txt")), "hit\n").unwrap();
}
let mut engine = Engine::new().unwrap();
engine
.execute("project.open", json!({"path": dir.path()}))
.unwrap();
let id = find(&mut engine, json!({"query": "hit"}));
std::thread::sleep(Duration::from_millis(500));
let first = engine.pump_vfs().unwrap();
let batch = first
.iter()
.filter(|e| matches!(e, Event::SearchResults { .. }))
.count();
assert!(batch <= MAX_RESULTS_PER_PUMP, "{batch}");
let (rest, done, _) = collect(&mut engine, id);
assert_eq!(batch + rest.len(), MAX_RESULTS_PER_PUMP + 100);
assert!(matches!(done, Event::SearchDone { files, .. } if files == MAX_RESULTS_PER_PUMP + 100));
}

View file

@ -12,6 +12,17 @@ publish.workspace = true
layer = 2
[dependencies]
globset = "0.4.18"
grep-matcher = "0.1.7"
grep-regex = "0.1.13"
grep-searcher = "0.1.14"
ignore.workspace = true
thiserror = "2.0.21"
[dev-dependencies]
lutra-testkit.workspace = true
proptest.workspace = true
tempfile.workspace = true
[lints]
workspace = true

View file

@ -2,3 +2,234 @@
//!
//! Streams matches from the project files off the UI thread, honouring the
//! ignore rules of the project model.
//!
//! # Query
//! A [`Query`] is compiled once into a [`Search`] ([`Search::new`]): a
//! literal or a regular expression (Rust `regex` syntax), case sensitive or
//! not, optionally whole words only, plus include and exclude globs. An
//! invalid pattern or glob fails with [`SearchError`] before anything runs.
//! Patterns never match across lines.
//!
//! # Which files are searched
//! The walk follows the rules of `lutra-vfs`:
//! - `.gitignore`, `.ignore`, `.git/info/exclude` and the global git excludes
//! apply whether or not the folder is a git repository; ignored files and
//! everything below an ignored directory are skipped. Ignore files above
//! the project root are not read.
//! - `.git` (directory or file) at any depth and every `node_modules`
//! directory are skipped.
//! - `vendor/` directly under the root is library code and is searched even
//! when an ignore file lists it (Composer projects usually do); ignore
//! files inside it still apply.
//! - Hidden (dot) files are searched, as the project tree shows them.
//! Symbolic links are not followed.
//! - Include globs keep only files that match one of them, exclude globs
//! drop files and whole directories. A glob without `/` matches the file
//! or directory name (`*.php`), a glob with `/` the project-relative path
//! (`src/**/*.php`).
//! - Files larger than [`MAX_FILE_SIZE`], binary files (a NUL byte) and
//! files that cannot be read are skipped.
//!
//! # Results
//! Each file with matches is reported once as [`FileMatches`], in no
//! particular order (the walk is parallel). Positions are 0-based; columns
//! and lengths count UTF-16 code units. Lines end at `\n` (a `\r` before it
//! is not part of the line). The preview is the line, or a window of at most
//! [`MAX_PREVIEW_CHARS`] characters around the match for longer lines.
//! At most [`Query::max_results`] matches are reported in total; the
//! [`Summary`] says whether the cap cut the search short.
//!
//! # Threads and cancellation
//! [`run`] blocks the calling thread and searches on a pool of up to
//! [`MAX_THREADS`] walker threads; [`spawn`] runs it on a thread of its own
//! and delivers [`SearchEvent`]s over a channel. A [`CancelToken`] stops a
//! running search between files and between lines.
//!
//! # Open buffers
//! An [`Overlay`] maps project-relative paths to text that replaces the file
//! on disk, so unsaved edits are searched instead of the stale file. Only
//! files the walk visits are looked up; text for a path that is not on disk
//! (or is skipped by the rules) is not searched.
mod lines;
mod query;
mod walk;
use std::collections::HashMap;
use std::path::PathBuf;
use std::sync::Arc;
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::mpsc::{Receiver, channel};
use std::time::Duration;
pub use query::Search;
pub use walk::run;
/// Largest file searched, in bytes (8 MiB).
pub const MAX_FILE_SIZE: u64 = 8 * 1024 * 1024;
/// Default cap on the number of matches of one search.
pub const DEFAULT_MAX_RESULTS: usize = 10_000;
/// Highest cap a query may ask for.
pub const MAX_RESULTS_LIMIT: usize = 100_000;
/// Longest preview, in characters (Unicode scalar values).
pub const MAX_PREVIEW_CHARS: usize = 300;
/// Most walker threads of one search.
pub const MAX_THREADS: usize = 8;
/// What to search for and where.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct Query {
/// The text or regular expression; must not be empty.
pub pattern: String,
/// Treat `pattern` as a regular expression instead of literal text.
pub regex: bool,
/// Match letter case exactly.
pub case_sensitive: bool,
/// Only matches that start and end at word boundaries.
pub whole_word: bool,
/// Only files matching one of these globs; all files when empty.
pub include: Vec<String>,
/// Skip files and directories matching one of these globs.
pub exclude: Vec<String>,
/// Stop after this many matches; clamped to `1..=`[`MAX_RESULTS_LIMIT`].
pub max_results: usize,
}
impl Query {
/// A literal, case-insensitive query for `pattern` with the default cap.
pub fn literal(pattern: impl Into<String>) -> Self {
Self {
pattern: pattern.into(),
regex: false,
case_sensitive: false,
whole_word: false,
include: Vec::new(),
exclude: Vec::new(),
max_results: DEFAULT_MAX_RESULTS,
}
}
}
/// Why a query cannot run.
#[derive(Debug, Clone, PartialEq, Eq, thiserror::Error)]
pub enum SearchError {
/// The pattern is empty.
#[error("the search text is empty")]
EmptyPattern,
/// The pattern is not a valid regular expression, or matches a line
/// break.
#[error("invalid pattern: {0}")]
InvalidPattern(String),
/// An include or exclude glob is invalid.
#[error("invalid glob: {0}")]
InvalidGlob(String),
}
/// One match.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct Match {
/// 0-based line.
pub line: usize,
/// Start in the line, in UTF-16 code units.
pub column: usize,
/// Length in UTF-16 code units; never 0.
pub length: usize,
/// The line, or a window of it around the match, without the line break.
pub preview: String,
/// Start of the match inside `preview`, in UTF-16 code units.
pub preview_column: usize,
}
/// The matches in one file, in line order.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct FileMatches {
/// Project-relative path with `/` separators.
pub path: String,
/// The matches; never empty.
pub matches: Vec<Match>,
}
/// How a search ended.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
pub struct Summary {
/// Files with at least one reported match.
pub files: usize,
/// Matches reported.
pub matches: usize,
/// The match cap stopped the search; more matches may exist.
pub truncated: bool,
/// The search was cancelled before it finished.
pub cancelled: bool,
/// Wall time from start to end.
pub elapsed: Duration,
}
/// What a search started with [`spawn`] delivers, in order: any number of
/// files, then exactly one summary.
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum SearchEvent {
/// A file with matches.
File(FileMatches),
/// The search ended; nothing follows.
Done(Summary),
}
/// Text of open buffers by project-relative path (`/` separators), searched
/// instead of the file on disk.
pub type Overlay = HashMap<String, String>;
/// Stops a running search. Clones share the flag.
#[derive(Debug, Clone, Default)]
pub struct CancelToken(Arc<AtomicBool>);
impl CancelToken {
/// A token that is not cancelled.
pub fn new() -> Self {
Self::default()
}
/// Asks the search to stop; it ends with [`Summary::cancelled`] set.
pub fn cancel(&self) {
self.0.store(true, Ordering::Relaxed);
}
/// Whether [`CancelToken::cancel`] was called.
pub fn is_cancelled(&self) -> bool {
self.0.load(Ordering::Relaxed)
}
}
/// Runs [`run`] on a new thread and returns the channel of its events. A
/// receiver that is dropped does not stop the search; cancel it with
/// `cancel`.
///
/// # Parameters
/// - `root`: the project root (absolute).
/// - `search`: the compiled query.
/// - `overlay`: buffer texts that replace files on disk.
/// - `cancel`: stops the search.
///
/// # Errors
/// The OS refused to start the thread.
pub fn spawn(
root: PathBuf,
search: Search,
overlay: Overlay,
cancel: CancelToken,
) -> std::io::Result<Receiver<SearchEvent>> {
let (tx, rx) = channel();
std::thread::Builder::new()
.name("lutra-search".into())
.spawn(move || {
let files = tx.clone();
let summary = run(&root, &search, &overlay, &cancel, &move |file| {
let _receiver_gone = files.send(SearchEvent::File(file));
});
let _receiver_gone = tx.send(SearchEvent::Done(summary));
})?;
Ok(rx)
}

195
crates/search/src/lines.rs Normal file
View file

@ -0,0 +1,195 @@
//! Turning a matching line into [`Match`]es: UTF-16 columns and previews.
use grep_matcher::Matcher;
use grep_regex::RegexMatcher;
use crate::{MAX_PREVIEW_CHARS, Match};
/// Characters shown before the match when a long line is cut to a window.
const PREVIEW_LEAD: usize = 80;
/// Appends the matches of one line to `out`, at most `limit` in total.
///
/// # Parameters
/// - `matcher`: the compiled pattern.
/// - `line`: 0-based line number.
/// - `bytes`: the line as read, with or without its line break. Invalid
/// UTF-8 is replaced by U+FFFD before matching, so positions refer to the
/// text an editor shows.
/// - `out`: receives the matches; zero-length matches are skipped.
/// - `limit`: stop once `out` holds this many matches.
///
/// A matcher error (none is expected for an in-memory regex) ends the line
/// early; the matches found so far are kept.
pub(crate) fn collect(
matcher: &RegexMatcher,
line: usize,
bytes: &[u8],
out: &mut Vec<Match>,
limit: usize,
) {
let bytes = trim_line_break(bytes);
let text = String::from_utf8_lossy(bytes);
let mut spans = Vec::new();
let _matcher_error = matcher.find_iter(text.as_bytes(), |m| {
if m.start() < m.end() {
spans.push((m.start(), m.end()));
}
out.len().saturating_add(spans.len()) < limit
});
if spans.is_empty() {
return;
}
let window = Window::new(&text);
let mut cursor = Cursor::default();
for (start, end) in spans {
let Some(matched) = text.get(start..end) else {
continue;
};
if !cursor.advance(&text, start) {
continue;
}
let (preview, preview_column) = window.preview(&text, &cursor, start);
out.push(Match {
line,
column: cursor.utf16,
length: utf16_len(matched),
preview,
preview_column,
});
}
}
/// `bytes` without a trailing `\n` or `\r\n`.
fn trim_line_break(bytes: &[u8]) -> &[u8] {
let bytes = bytes.strip_suffix(b"\n").unwrap_or(bytes);
bytes.strip_suffix(b"\r").unwrap_or(bytes)
}
/// Number of UTF-16 code units of `s`.
fn utf16_len(s: &str) -> usize {
s.chars().map(char::len_utf16).sum()
}
/// A position in the line, counted three ways; only moves forward.
#[derive(Default)]
struct Cursor {
/// Byte offset.
byte: usize,
/// Characters before `byte`.
chars: usize,
/// UTF-16 code units before `byte`.
utf16: usize,
}
impl Cursor {
/// Moves to byte offset `to`; false if `to` lies behind the cursor or
/// not on a character boundary.
fn advance(&mut self, text: &str, to: usize) -> bool {
let Some(between) = text.get(self.byte..to) else {
return false;
};
for c in between.chars() {
self.chars = self.chars.saturating_add(1);
self.utf16 = self.utf16.saturating_add(c.len_utf16());
}
self.byte = to;
true
}
}
/// How previews of one line are cut.
enum Window {
/// The line is short enough to be the preview of every match.
Whole,
/// The line is longer than [`MAX_PREVIEW_CHARS`]: the byte offset of
/// every character, plus the text length at the end.
Cut(Vec<usize>),
}
impl Window {
/// Prepares previews for `text`.
fn new(text: &str) -> Self {
if text.len() <= MAX_PREVIEW_CHARS || text.chars().count() <= MAX_PREVIEW_CHARS {
return Self::Whole;
}
let mut offsets: Vec<usize> = text.char_indices().map(|(i, _)| i).collect();
offsets.push(text.len());
Self::Cut(offsets)
}
/// The preview for a match at `cursor` (byte offset `start`), and the
/// match's UTF-16 column inside it.
fn preview(&self, text: &str, cursor: &Cursor, start: usize) -> (String, usize) {
let Self::Cut(offsets) = self else {
return (text.to_string(), cursor.utf16);
};
let total = offsets.len().saturating_sub(1);
let first = cursor
.chars
.saturating_sub(PREVIEW_LEAD)
.min(total.saturating_sub(MAX_PREVIEW_CHARS));
let last = first.saturating_add(MAX_PREVIEW_CHARS).min(total);
let from = offsets.get(first).copied().unwrap_or(0);
let to = offsets.get(last).copied().unwrap_or(text.len());
let preview = text.get(from..to).unwrap_or_default().to_string();
let column = text.get(from..start).map(utf16_len).unwrap_or(0);
(preview, column)
}
}
#[cfg(test)]
mod tests {
use grep_regex::RegexMatcherBuilder;
use super::*;
fn matcher(pattern: &str) -> RegexMatcher {
RegexMatcherBuilder::new()
.fixed_strings(true)
.line_terminator(Some(b'\n'))
.build(pattern)
.unwrap()
}
#[test]
fn columns_count_utf16_units() {
let mut out = Vec::new();
collect(&matcher("x"), 4, "ä😀x and x\r\n".as_bytes(), &mut out, 10);
assert_eq!(out.len(), 2);
assert_eq!((out[0].line, out[0].column, out[0].length), (4, 3, 1));
assert_eq!(out[0].preview, "ä😀x and x");
assert_eq!(out[0].preview_column, 3);
assert_eq!(out[1].column, 9);
}
#[test]
fn long_lines_are_cut_around_the_match() {
let line = format!("{}needle{}", "a".repeat(1000), "b".repeat(1000));
let mut out = Vec::new();
collect(&matcher("needle"), 0, line.as_bytes(), &mut out, 10);
let m = &out[0];
assert_eq!(m.column, 1000);
assert_eq!(m.preview.chars().count(), MAX_PREVIEW_CHARS);
assert_eq!(
m.preview.get(m.preview_column..m.preview_column + 6),
Some("needle")
);
assert_eq!(m.preview_column, PREVIEW_LEAD);
let tail = format!("{}needle", "a".repeat(1000));
out.clear();
collect(&matcher("needle"), 0, tail.as_bytes(), &mut out, 10);
assert!(out[0].preview.ends_with("needle"));
assert_eq!(out[0].preview.chars().count(), MAX_PREVIEW_CHARS);
}
#[test]
fn invalid_utf8_and_limits() {
let mut out = Vec::new();
collect(&matcher("ab"), 0, b"\xffab ab ab\n", &mut out, 2);
assert_eq!(out.len(), 2);
assert_eq!(out[0].column, 1);
assert_eq!(out[0].preview, "\u{fffd}ab ab ab");
}
}

139
crates/search/src/query.rs Normal file
View file

@ -0,0 +1,139 @@
//! Compiling a [`Query`]: the matcher and the path filters.
use globset::{GlobBuilder, GlobSet, GlobSetBuilder};
use grep_regex::{RegexMatcher, RegexMatcherBuilder};
use crate::{MAX_RESULTS_LIMIT, Query, SearchError};
/// Largest compiled regex, in bytes; bigger patterns are rejected instead of
/// using unbounded memory.
const REGEX_SIZE_LIMIT: usize = 16 * 1024 * 1024;
/// Most nesting a pattern may use.
const REGEX_NEST_LIMIT: u32 = 100;
/// A compiled query, ready to run.
#[derive(Debug, Clone)]
pub struct Search {
/// Matches one line at a time.
pub(crate) matcher: RegexMatcher,
/// Include and exclude globs.
pub(crate) filter: PathFilter,
/// Cap on reported matches.
pub(crate) max_results: usize,
}
impl Search {
/// Compiles `query`.
///
/// # Errors
/// [`SearchError::EmptyPattern`], [`SearchError::InvalidPattern`] for a
/// bad regular expression (or one that can match a line break),
/// [`SearchError::InvalidGlob`] for a bad include or exclude glob.
pub fn new(query: &Query) -> Result<Self, SearchError> {
if query.pattern.is_empty() {
return Err(SearchError::EmptyPattern);
}
let matcher = RegexMatcherBuilder::new()
.fixed_strings(!query.regex)
.case_insensitive(!query.case_sensitive)
.word(query.whole_word)
.line_terminator(Some(b'\n'))
.crlf(true)
.size_limit(REGEX_SIZE_LIMIT)
.dfa_size_limit(REGEX_SIZE_LIMIT)
.nest_limit(REGEX_NEST_LIMIT)
.build(&query.pattern)
.map_err(|e| SearchError::InvalidPattern(e.to_string()))?;
Ok(Self {
matcher,
filter: PathFilter::new(&query.include, &query.exclude)?,
max_results: query.max_results.clamp(1, MAX_RESULTS_LIMIT),
})
}
}
/// Include and exclude globs, split by whether they match a name or a path.
#[derive(Debug, Clone)]
pub(crate) struct PathFilter {
/// Include globs; `None` when every file is included.
include: Option<Globs>,
/// Exclude globs; `None` when nothing is excluded.
exclude: Option<Globs>,
}
/// Globs matched against the file name and against the relative path.
#[derive(Debug, Clone)]
struct Globs {
/// Globs without `/`.
names: GlobSet,
/// Globs with `/`.
paths: GlobSet,
}
impl PathFilter {
/// Compiles the globs; blank ones are ignored.
fn new(include: &[String], exclude: &[String]) -> Result<Self, SearchError> {
Ok(Self {
include: Globs::new(include)?,
exclude: Globs::new(exclude)?,
})
}
/// Whether the file at `rel` (project-relative, `/` separators) passes
/// the include and exclude globs.
pub(crate) fn file(&self, rel: &str) -> bool {
let included = self.include.as_ref().is_none_or(|g| g.matches(rel));
included && !self.excluded(rel)
}
/// Whether an exclude glob matches the file or directory at `rel`.
pub(crate) fn excluded(&self, rel: &str) -> bool {
self.exclude.as_ref().is_some_and(|g| g.matches(rel))
}
}
impl Globs {
/// Compiles `globs`; `None` if there is none. Leading and trailing `/`
/// are dropped, so `tests/` excludes every directory named `tests`.
fn new(globs: &[String]) -> Result<Option<Self>, SearchError> {
let mut names = GlobSetBuilder::new();
let mut paths = GlobSetBuilder::new();
let mut any = false;
for glob in globs.iter().map(|g| g.trim()).filter(|g| !g.is_empty()) {
let pattern = glob.trim_matches('/');
if pattern.is_empty() {
continue;
}
let compiled = GlobBuilder::new(pattern)
.literal_separator(true)
.backslash_escape(true)
.build()
.map_err(|e| SearchError::InvalidGlob(e.to_string()))?;
if pattern.contains('/') {
paths.add(compiled);
} else {
names.add(compiled);
}
any = true;
}
if !any {
return Ok(None);
}
let build = |b: GlobSetBuilder| {
b.build()
.map_err(|e| SearchError::InvalidGlob(e.to_string()))
};
Ok(Some(Self {
names: build(names)?,
paths: build(paths)?,
}))
}
/// Whether `rel`'s last component matches a name glob or `rel` a path
/// glob.
fn matches(&self, rel: &str) -> bool {
let name = rel.rsplit('/').next().unwrap_or(rel);
self.names.is_match(name) || self.paths.is_match(rel)
}
}

280
crates/search/src/walk.rs Normal file
View file

@ -0,0 +1,280 @@
//! The parallel walk over the project and the per-file search.
use std::path::{Path, PathBuf};
use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
use std::time::Instant;
use grep_matcher::LineTerminator;
use grep_regex::RegexMatcher;
use grep_searcher::{BinaryDetection, Searcher, SearcherBuilder, Sink, SinkMatch};
use ignore::{DirEntry, WalkBuilder, WalkState};
use crate::query::PathFilter;
use crate::{
CancelToken, FileMatches, MAX_FILE_SIZE, MAX_THREADS, Match, Overlay, Search, Summary, lines,
};
/// Searches the project at `root` and calls `on_file` for every file with
/// matches, from the walker threads, while the search runs. Blocks until
/// the walk ends, the cap is reached or `cancel` fires.
///
/// # Parameters
/// - `root`: the project root (absolute).
/// - `search`: the compiled query.
/// - `overlay`: buffer texts that replace files on disk.
/// - `cancel`: stops the search; checked between files and lines.
/// - `on_file`: receives each file's matches.
///
/// # Returns
/// The totals. Files that cannot be read are skipped silently: a search
/// reports what it can read, like the project tree lists what it can.
pub fn run(
root: &Path,
search: &Search,
overlay: &Overlay,
cancel: &CancelToken,
on_file: &(dyn Fn(FileMatches) + Sync),
) -> Summary {
let started = Instant::now();
let totals = Totals::default();
walker(root, &search.filter).build_parallel().run(|| {
let mut searcher = SearcherBuilder::new()
.line_terminator(LineTerminator::crlf())
.line_number(true)
.binary_detection(BinaryDetection::quit(0))
.build();
let totals = &totals;
Box::new(move |entry| {
if cancel.is_cancelled() || totals.full.load(Ordering::Relaxed) {
return WalkState::Quit;
}
let Ok(entry) = entry else {
return WalkState::Continue;
};
let Some(rel) = file_path(root, &entry, &search.filter) else {
return WalkState::Continue;
};
let room = search
.max_results
.saturating_sub(totals.matches.load(Ordering::Relaxed));
let Some(mut matches) = search_file(
&mut searcher,
&search.matcher,
entry.path(),
overlay.get(&rel),
room,
cancel,
) else {
return WalkState::Continue;
};
let granted = totals.reserve(matches.len(), search.max_results);
if granted < matches.len() {
matches.truncate(granted);
totals.full.store(true, Ordering::Relaxed);
}
if granted > 0 {
totals.files.fetch_add(1, Ordering::Relaxed);
on_file(FileMatches { path: rel, matches });
}
if totals.full.load(Ordering::Relaxed) {
WalkState::Quit
} else {
WalkState::Continue
}
})
});
let cancelled = cancel.is_cancelled();
Summary {
files: totals.files.load(Ordering::Relaxed),
matches: totals.matches.load(Ordering::Relaxed),
truncated: totals.full.load(Ordering::Relaxed) && !cancelled,
cancelled,
elapsed: started.elapsed(),
}
}
/// Counters shared by the walker threads.
#[derive(Default)]
struct Totals {
/// Files reported.
files: AtomicUsize,
/// Matches reported.
matches: AtomicUsize,
/// The cap was reached, or a file had more matches than fit.
full: AtomicBool,
}
impl Totals {
/// Takes up to `wanted` matches from what is left under `cap` and
/// returns how many were granted. The caller marks the totals full when
/// it got fewer than it wanted: only then is a match known to be left
/// out.
fn reserve(&self, wanted: usize, cap: usize) -> usize {
let mut used = self.matches.load(Ordering::Relaxed);
loop {
let granted = wanted.min(cap.saturating_sub(used));
match self.matches.compare_exchange_weak(
used,
used.saturating_add(granted),
Ordering::Relaxed,
Ordering::Relaxed,
) {
Ok(_) => return granted,
Err(now) => used = now,
}
}
}
}
/// The walker over `root` with the project's rules (see the crate docs).
fn walker(root: &Path, filter: &PathFilter) -> WalkBuilder {
let threads = std::thread::available_parallelism()
.map(|n| n.get())
.unwrap_or(2)
.clamp(1, MAX_THREADS);
let mut builder = WalkBuilder::new(root);
builder
.hidden(false)
.parents(false)
.ignore(true)
.git_ignore(true)
.git_exclude(true)
.git_global(true)
.require_git(false)
.follow_links(false)
.threads(threads);
let vendor = root.join("vendor");
let vendor_is_dir = std::fs::symlink_metadata(&vendor).is_ok_and(|m| m.is_dir());
if vendor_is_dir {
builder.add(&vendor);
}
let root = root.to_path_buf();
let filter = filter.clone();
builder.filter_entry(move |entry| keep_entry(&root, &vendor, &filter, entry));
builder
}
/// Whether the walk enters (directory) or visits (file) `entry`: `.git` and
/// `node_modules` are skipped, the root's `vendor` is reached through its
/// own walk root instead of the project root's walk, and exclude globs drop
/// whole directories.
fn keep_entry(root: &Path, vendor: &Path, filter: &PathFilter, entry: &DirEntry) -> bool {
let name = entry.file_name();
if name == ".git" {
return false;
}
let is_dir = entry.file_type().is_some_and(|t| t.is_dir());
if is_dir && name == "node_modules" {
return false;
}
if entry.depth() == 1 && entry.path() == vendor {
return false;
}
if is_dir && let Some(rel) = relative(root, entry.path()) {
return !filter.excluded(&rel);
}
true
}
/// The project-relative path of `entry` if it is a regular file that the
/// globs keep.
fn file_path(root: &Path, entry: &DirEntry, filter: &PathFilter) -> Option<String> {
if !entry.file_type().is_some_and(|t| t.is_file()) {
return None;
}
let rel = relative(root, entry.path())?;
filter.file(&rel).then_some(rel)
}
/// `path` relative to `root` with `/` separators; `None` outside the root or
/// for a name that is not valid Unicode.
fn relative(root: &Path, path: &Path) -> Option<String> {
let rel: PathBuf = path.strip_prefix(root).ok()?.to_path_buf();
let parts: Option<Vec<&str>> = rel.components().map(|c| c.as_os_str().to_str()).collect();
Some(parts?.join("/"))
}
/// Searches one file, or the buffer text that replaces it, and returns its
/// matches; `None` without matches, for a binary, oversized or unreadable
/// file, or when cancelled.
///
/// # Parameters
/// - `room`: most matches to collect; one more is collected when possible,
/// so the caller can tell that the cap cut the file short (with no room
/// left, one match is enough to know that).
fn search_file(
searcher: &mut Searcher,
matcher: &RegexMatcher,
path: &Path,
buffer: Option<&String>,
room: usize,
cancel: &CancelToken,
) -> Option<Vec<Match>> {
let mut sink = Collect {
matcher,
matches: Vec::new(),
limit: room.saturating_add(1),
binary: false,
cancel,
};
let result = match buffer {
Some(text) => {
if text.len() as u64 > MAX_FILE_SIZE {
return None;
}
searcher.search_slice(matcher, text.as_bytes(), &mut sink)
}
None => {
let size = std::fs::metadata(path).ok()?.len();
if size > MAX_FILE_SIZE {
return None;
}
searcher.search_path(matcher, path, &mut sink)
}
};
if result.is_err() || sink.binary || cancel.is_cancelled() || sink.matches.is_empty() {
return None;
}
Some(sink.matches)
}
/// Collects the matches of one file.
struct Collect<'a> {
/// The pattern, run again on each matching line to find every match.
matcher: &'a RegexMatcher,
/// Matches so far.
matches: Vec<Match>,
/// Stop at this many matches.
limit: usize,
/// The file turned out to be binary.
binary: bool,
/// Stops the file early.
cancel: &'a CancelToken,
}
impl Sink for Collect<'_> {
type Error = std::io::Error;
fn matched(&mut self, _searcher: &Searcher, mat: &SinkMatch<'_>) -> Result<bool, Self::Error> {
if self.cancel.is_cancelled() {
return Ok(false);
}
let line = mat
.line_number()
.and_then(|n| usize::try_from(n.saturating_sub(1)).ok())
.unwrap_or(0);
lines::collect(
self.matcher,
line,
mat.bytes(),
&mut self.matches,
self.limit,
);
Ok(self.matches.len() < self.limit)
}
fn binary_data(&mut self, _searcher: &Searcher, _offset: u64) -> Result<bool, Self::Error> {
self.binary = true;
Ok(false)
}
}

View file

@ -0,0 +1,345 @@
//! Find in files over real directories: matching, rules, caps, cancel.
#![allow(
clippy::unwrap_used,
clippy::expect_used,
clippy::indexing_slicing,
clippy::panic
)]
use std::fs;
use std::path::Path;
use std::sync::Mutex;
use std::time::{Duration, Instant};
use lutra_search::{
CancelToken, FileMatches, MAX_FILE_SIZE, Overlay, Query, Search, SearchError, SearchEvent,
Summary, run, spawn,
};
use proptest::prelude::*;
/// Runs `query` over `root` and returns the files sorted by path.
fn find(root: &Path, query: &Query, overlay: &Overlay) -> (Vec<FileMatches>, Summary) {
let search = Search::new(query).unwrap();
let found = Mutex::new(Vec::new());
let summary = run(root, &search, overlay, &CancelToken::new(), &|f| {
found.lock().unwrap().push(f)
});
let mut files = found.into_inner().unwrap();
files.sort_by(|a, b| a.path.cmp(&b.path));
(files, summary)
}
fn paths(files: &[FileMatches]) -> Vec<&str> {
files.iter().map(|f| f.path.as_str()).collect()
}
fn put(root: &Path, rel: &str, text: &[u8]) {
let path = root.join(rel);
fs::create_dir_all(path.parent().unwrap()).unwrap();
fs::write(path, text).unwrap();
}
#[test]
fn literal_case_and_whole_word() {
let dir = tempfile::tempdir().unwrap();
put(
dir.path(),
"a.php",
b"<?php\n$user = new User();\n$users = [];\nUSER\n",
);
let mut q = Query::literal("user");
let (files, summary) = find(dir.path(), &q, &Overlay::new());
assert_eq!(summary.matches, 4);
let m = &files[0].matches;
assert_eq!((m[0].line, m[0].column, m[0].length), (1, 1, 4));
assert_eq!(m[0].preview, "$user = new User();");
assert_eq!((m[1].line, m[1].column), (1, 12));
q.case_sensitive = true;
assert_eq!(find(dir.path(), &q, &Overlay::new()).1.matches, 2);
q.whole_word = true;
let (files, _) = find(dir.path(), &q, &Overlay::new());
assert_eq!(files[0].matches.len(), 1);
assert_eq!(files[0].matches[0].line, 1);
let q = Query::literal("$users = [");
let (files, _) = find(dir.path(), &q, &Overlay::new());
assert_eq!(files[0].matches[0].line, 2, "regex characters are literal");
}
#[test]
fn regex_and_invalid_input() {
let dir = tempfile::tempdir().unwrap();
put(dir.path(), "a.php", b"foo1 foo22 bar\r\nfoo333\r\n");
let mut q = Query::literal(r"foo\d+");
q.regex = true;
let (files, summary) = find(dir.path(), &q, &Overlay::new());
assert_eq!(summary.matches, 3);
assert_eq!(files[0].matches[2].preview, "foo333", "no \\r in previews");
assert_eq!(files[0].matches[1].length, 5);
q.pattern = "(".into();
assert!(matches!(
Search::new(&q),
Err(SearchError::InvalidPattern(_))
));
q.pattern = String::new();
assert_eq!(Search::new(&q).unwrap_err(), SearchError::EmptyPattern);
let mut q = Query::literal("x");
q.include = vec!["[".into()];
assert!(matches!(Search::new(&q), Err(SearchError::InvalidGlob(_))));
}
#[test]
fn include_and_exclude_globs() {
let dir = tempfile::tempdir().unwrap();
for rel in ["src/a.php", "src/b.js", "tests/c.php", "src/deep/d.php"] {
put(dir.path(), rel, b"needle\n");
}
let mut q = Query::literal("needle");
q.include = vec!["*.php".into()];
let (files, _) = find(dir.path(), &q, &Overlay::new());
assert_eq!(
paths(&files),
["src/a.php", "src/deep/d.php", "tests/c.php"]
);
q.exclude = vec!["tests/".into()];
let (files, _) = find(dir.path(), &q, &Overlay::new());
assert_eq!(paths(&files), ["src/a.php", "src/deep/d.php"]);
q.include = vec!["src/*.php".into()];
q.exclude = vec![];
let (files, _) = find(dir.path(), &q, &Overlay::new());
assert_eq!(
paths(&files),
["src/a.php"],
"a path glob keeps to its level"
);
}
#[test]
fn honours_the_project_rules() {
let dir = tempfile::tempdir().unwrap();
let tree = lutra_testkit::generate_search_tree(dir.path(), 400);
put(dir.path(), "sub/.gitignore", b"skip.php\n");
put(dir.path(), "sub/skip.php", b"lutraNeedle\n");
put(dir.path(), "sub/.hidden.php", b"lutraNeedle\n");
put(dir.path(), ".git/HEAD", b"lutraNeedle\n");
let (files, summary) = find(
dir.path(),
&Query::literal(lutra_testkit::NEEDLE),
&Overlay::new(),
);
let found = paths(&files);
assert!(found.iter().all(|p| !p.starts_with("var/")), "{found:?}");
assert!(found.iter().all(|p| !p.starts_with("node_modules/")));
assert!(found.iter().all(|p| !p.starts_with(".git/")));
assert!(!found.contains(&"bin/data.bin"), "binary skipped");
assert!(!found.contains(&"debug.log"), "ignored file skipped");
assert!(!found.contains(&"sub/skip.php"), "nested .gitignore");
assert!(found.contains(&"sub/.hidden.php"), "dot files are searched");
let vendor = found.iter().filter(|p| p.starts_with("vendor/")).count();
assert_eq!(vendor, tree.needle_vendor, "vendor is searched");
assert!(vendor > 0);
assert_eq!(summary.files, tree.needle_files + 1);
assert!(!summary.truncated && !summary.cancelled);
}
#[test]
fn skips_large_and_binary_files() {
let dir = tempfile::tempdir().unwrap();
let mut big = b"needle\n".to_vec();
big.resize(usize::try_from(MAX_FILE_SIZE).unwrap() + 1, b'a');
put(dir.path(), "big.txt", &big);
let mut late_nul = b"needle\n".to_vec();
late_nul.extend(b"filler\n".repeat(50_000));
late_nul.push(0);
put(dir.path(), "late.bin", &late_nul);
put(dir.path(), "ok.txt", b"needle\n");
let (files, _) = find(dir.path(), &Query::literal("needle"), &Overlay::new());
assert_eq!(paths(&files), ["ok.txt"]);
}
#[test]
fn overlay_replaces_the_disk_text() {
let dir = tempfile::tempdir().unwrap();
put(dir.path(), "a.php", b"old text\n");
put(dir.path(), "b.php", b"needle on disk\n");
let mut overlay = Overlay::new();
overlay.insert("a.php".into(), "line\n needle in buffer\n".into());
overlay.insert("b.php".into(), "gone\n".into());
overlay.insert("missing.php".into(), "needle\n".into());
let (files, _) = find(dir.path(), &Query::literal("needle"), &overlay);
assert_eq!(paths(&files), ["a.php"]);
assert_eq!(files[0].matches[0].line, 1);
assert_eq!(files[0].matches[0].column, 2);
}
#[test]
fn caps_the_number_of_matches() {
let dir = tempfile::tempdir().unwrap();
for i in 0..20 {
put(dir.path(), &format!("f{i}.txt"), b"x x x x x\n");
}
let mut q = Query::literal("x");
q.max_results = 7;
let (files, summary) = find(dir.path(), &q, &Overlay::new());
assert_eq!(summary.matches, 7);
assert!(summary.truncated);
assert_eq!(files.iter().map(|f| f.matches.len()).sum::<usize>(), 7);
assert_eq!(summary.files, files.len());
q.max_results = 100;
let (_, summary) = find(dir.path(), &q, &Overlay::new());
assert_eq!(summary.matches, 100);
assert!(
!summary.truncated,
"exactly at the cap with nothing left out"
);
q.max_results = 99;
assert!(find(dir.path(), &q, &Overlay::new()).1.truncated);
}
#[test]
fn spawn_streams_and_cancels() {
let dir = tempfile::tempdir().unwrap();
let tree = lutra_testkit::generate_search_tree(dir.path(), 300);
let search = Search::new(&Query::literal(lutra_testkit::NEEDLE)).unwrap();
let rx = spawn(
dir.path().into(),
search.clone(),
Overlay::new(),
CancelToken::new(),
)
.unwrap();
let events: Vec<SearchEvent> = rx.iter().collect();
let files = events
.iter()
.filter(|e| matches!(e, SearchEvent::File(_)))
.count();
assert_eq!(files, tree.needle_files);
let Some(SearchEvent::Done(summary)) = events.last() else {
panic!("no summary last: {events:?}");
};
assert_eq!(summary.files, files);
let cancel = CancelToken::new();
cancel.cancel();
let rx = spawn(dir.path().into(), search, Overlay::new(), cancel).unwrap();
let events: Vec<SearchEvent> = rx.iter().collect();
assert_eq!(events.len(), 1);
let SearchEvent::Done(summary) = &events[0] else {
panic!("{events:?}");
};
assert!(summary.cancelled);
assert_eq!(summary.matches, 0);
}
/// Measures the first result and the total on a generated tree of 20,000
/// files. Run in release: `cargo test -p lutra-search --release --test
/// search -- --ignored --nocapture measure`.
#[test]
#[ignore = "measurement; run explicitly in release"]
fn measure_20k_tree() {
let dir = tempfile::tempdir().unwrap();
let made = Instant::now();
let tree = lutra_testkit::generate_search_tree(dir.path(), 20_000);
println!("generated {} files in {:?}", tree.files, made.elapsed());
for (label, pattern) in [
("common literal", "function"),
("rare literal", lutra_testkit::NEEDLE),
] {
for round in 0..3 {
let search = Search::new(&Query::literal(pattern)).unwrap();
let started = Instant::now();
let rx = spawn(
dir.path().into(),
search,
Overlay::new(),
CancelToken::new(),
)
.unwrap();
let mut first: Option<Duration> = None;
let mut summary = None;
for event in rx {
match event {
SearchEvent::File(_) => {
first.get_or_insert_with(|| started.elapsed());
}
SearchEvent::Done(s) => summary = Some(s),
}
}
let s = summary.unwrap();
println!(
"{label} round {round}: first {:?}, total {:?}, files {}, matches {}, truncated {}",
first.unwrap(),
s.elapsed,
s.files,
s.matches,
s.truncated
);
}
}
let mut q = Query::literal("function");
q.max_results = lutra_search::MAX_RESULTS_LIMIT;
let search = Search::new(&q).unwrap();
let s = run(
dir.path(),
&search,
&Overlay::new(),
&CancelToken::new(),
&|_| {},
);
println!(
"common literal, cap 100k: total {:?}, files {}, matches {}, truncated {}",
s.elapsed, s.files, s.matches, s.truncated
);
}
proptest! {
#![proptest_config(ProptestConfig::with_cases(64))]
/// Random patterns, flags and file contents never panic, and every
/// reported match lies inside its preview.
#[test]
fn random_queries_never_panic(
pattern in "\\PC{0,12}",
regex in any::<bool>(),
case_sensitive in any::<bool>(),
whole_word in any::<bool>(),
content in proptest::collection::vec(any::<u8>(), 0..400),
text in "\\PC{0,200}",
) {
let dir = tempfile::tempdir().unwrap();
fs::write(dir.path().join("bytes.txt"), &content).unwrap();
fs::write(dir.path().join("text.php"), &text).unwrap();
let query = Query {
pattern,
regex,
case_sensitive,
whole_word,
include: vec![],
exclude: vec![],
max_results: 50,
};
let Ok(search) = Search::new(&query) else {
return Ok(());
};
let found = Mutex::new(Vec::new());
let summary = run(dir.path(), &search, &Overlay::new(), &CancelToken::new(), &|f| {
found.lock().unwrap().push(f)
});
let found = found.into_inner().unwrap();
prop_assert!(summary.matches <= 50);
for file in found {
for m in file.matches {
prop_assert!(m.length > 0);
let units = m.preview.encode_utf16().count();
prop_assert!(m.preview_column + m.length <= units || m.preview_column <= units);
}
}
}
}

View file

@ -10,6 +10,8 @@
clippy::indexing_slicing
)]
pub mod tree;
use std::path::{Path, PathBuf};
use std::sync::Mutex;
@ -17,6 +19,8 @@ use lutra_control::{Backend, EventSink, Subscribers};
use lutra_engine::{CommandInfo, Effect, Engine, EngineConfig, EngineError};
use serde_json::Value;
pub use tree::{NEEDLE, SearchTree, generate_search_tree};
/// Engine directories inside `root` (`root/config`, `root/data`), for
/// [`Engine::with_config`] and [`EngineBackend::with_config`] in tests.
pub fn engine_config(root: &Path) -> EngineConfig {

151
crates/testkit/src/tree.rs Normal file
View file

@ -0,0 +1,151 @@
//! Generated project trees for search tests and measurements.
//!
//! [`generate_search_tree`] writes a deterministic PHP project of any size:
//! project sources, templates, scripts and docs, a gitignored `var/`, a
//! gitignored Composer `vendor/` (searched as library code), an excluded
//! `node_modules/` and a binary file. Every [`NEEDLE_EVERY`]th file of each
//! group contains [`NEEDLE`] once; [`SearchTree`] reports how many files of
//! each group have it, so tests know the expected result without searching
//! twice.
use std::fs;
use std::path::Path;
/// A token that occurs only where the generator puts it.
pub const NEEDLE: &str = "lutraNeedle";
/// Every this many files of a group contains [`NEEDLE`].
pub const NEEDLE_EVERY: usize = 50;
/// What [`generate_search_tree`] wrote.
#[derive(Debug, Clone, Copy, Default, PartialEq, Eq)]
pub struct SearchTree {
/// Files written in total, including `.gitignore` and the binary file.
pub files: usize,
/// Searchable files (sources, templates, scripts, docs, vendor) that
/// contain [`NEEDLE`].
pub needle_files: usize,
/// Of those, files under `vendor/`.
pub needle_vendor: usize,
/// Files containing [`NEEDLE`] that a search must skip (ignored,
/// excluded or binary).
pub needle_skipped: usize,
}
/// Writes a project of about `files` files below `root`.
///
/// Shares: 70 % PHP sources in `src/`, 8 % templates, 5 % JavaScript, 5 %
/// Markdown, 6 % ignored PHP in `var/cache/`, 6 % PHP in `vendor/`, plus a
/// few files in `node_modules/` and `bin/data.bin` (a NUL byte, then the
/// needle). Every PHP file contains `function` on several lines, the common
/// literal for measurements.
///
/// # Panics
/// If a file cannot be written (test helper).
pub fn generate_search_tree(root: &Path, files: usize) -> SearchTree {
let mut tree = SearchTree::default();
write(root, ".gitignore", "/var/\n/vendor/\n*.log\n", &mut tree);
let groups: [(&str, usize, Kind); 6] = [
("src", 70, Kind::Php),
("templates", 8, Kind::Twig),
("assets", 5, Kind::Js),
("docs", 5, Kind::Markdown),
("var/cache", 6, Kind::Php),
("vendor/acme/toolkit/src", 6, Kind::Php),
];
for (dir, share, kind) in groups {
let count = files.saturating_mul(share) / 100;
for i in 0..count {
let needle = i % NEEDLE_EVERY == 0;
let rel = format!("{dir}/Module{:03}/{}", i / 100, kind.file_name(i));
write(root, &rel, &kind.content(i, needle), &mut tree);
if !needle {
continue;
}
if dir == "var/cache" {
tree.needle_skipped += 1;
} else {
tree.needle_files += 1;
if dir.starts_with("vendor") {
tree.needle_vendor += 1;
}
}
}
}
for i in 0..3 {
let rel = format!("node_modules/pkg{i}/index.js");
write(root, &rel, &Kind::Js.content(i, true), &mut tree);
tree.needle_skipped += 1;
}
write(root, "debug.log", &format!("{NEEDLE}\n"), &mut tree);
tree.needle_skipped += 1;
let mut binary = b"\x89BIN\x00\x01\x02".to_vec();
binary.extend_from_slice(NEEDLE.as_bytes());
fs::create_dir_all(root.join("bin")).unwrap();
fs::write(root.join("bin/data.bin"), binary).unwrap();
tree.files += 1;
tree.needle_skipped += 1;
tree
}
/// Writes `text` to `root/rel`, creating directories, and counts it.
fn write(root: &Path, rel: &str, text: &str, tree: &mut SearchTree) {
let path = root.join(rel);
if let Some(parent) = path.parent() {
fs::create_dir_all(parent).unwrap();
}
fs::write(path, text).unwrap();
tree.files += 1;
}
/// The kind of a generated file.
#[derive(Clone, Copy)]
enum Kind {
/// A PHP class.
Php,
/// A Twig template.
Twig,
/// A JavaScript module.
Js,
/// A Markdown page.
Markdown,
}
impl Kind {
/// File name of the `i`th file.
fn file_name(self, i: usize) -> String {
match self {
Kind::Php => format!("Service{i}.php"),
Kind::Twig => format!("page{i}.html.twig"),
Kind::Js => format!("widget{i}.js"),
Kind::Markdown => format!("note{i}.md"),
}
}
/// Content of the `i`th file, with [`NEEDLE`] once if `needle`.
fn content(self, i: usize, needle: bool) -> String {
let marker = if needle { NEEDLE } else { "plain" };
match self {
Kind::Php => {
let mut s = format!(
"<?php\n\ndeclare(strict_types=1);\n\nnamespace App\\Module{};\n\nuse App\\Contracts\\Handler;\n\n/**\n * Service number {i}.\n */\nfinal class Service{i} implements Handler\n{{\n private array $items = [];\n\n",
i / 100
);
for m in 0..8 {
s.push_str(&format!(
" public function handle{m}(int $id, string $name): ?string\n {{\n $key = sprintf('%s-%d', $name, $id + {m});\n if (isset($this->items[$key])) {{\n return $this->items[$key];\n }}\n return null;\n }}\n\n"
));
}
s.push_str(&format!(" public const MARKER = '{marker}';\n}}\n"));
s
}
Kind::Twig => format!(
"{{% extends 'base.html.twig' %}}\n{{% block body %}}\n <h1>Page {i}</h1>\n <p class=\"{marker}\">{{{{ content }}}}</p>\n{{% endblock %}}\n"
),
Kind::Js => format!(
"export function widget{i}(el) {{\n el.dataset.kind = '{marker}';\n return el;\n}}\n"
),
Kind::Markdown => format!("# Note {i}\n\nSome text about {marker} and more text.\n"),
}
}
}