rust/crates/ra_ide_db/src/symbol_index.rs

436 lines
13 KiB
Rust
Raw Normal View History

2019-01-08 13:33:36 -06:00
//! This module handles fuzzy-searching of functions, structs and other symbols
//! by name across the whole workspace and dependencies.
//!
//! It works by building an incrementally-updated text-search index of all
//! symbols. The backbone of the index is the **awesome** `fst` crate by
//! @BurntSushi.
//!
2019-02-11 10:18:27 -06:00
//! In a nutshell, you give a set of strings to `fst`, and it builds a
2019-01-20 20:35:18 -06:00
//! finite state machine describing this set of strings. The strings which
2019-01-08 13:33:36 -06:00
//! could fuzzy-match a pattern can also be described by a finite state machine.
2019-02-11 10:18:27 -06:00
//! What is freaking cool is that you can now traverse both state machines in
2019-01-08 13:33:36 -06:00
//! lock-step to enumerate the strings which are both in the input set and
2019-01-20 20:35:18 -06:00
//! fuzz-match the query. Or, more formally, given two languages described by
2019-02-11 10:18:27 -06:00
//! FSTs, one can build a product FST which describes the intersection of the
2019-01-08 13:33:36 -06:00
//! languages.
//!
//! `fst` does not support cheap updating of the index, but it supports unioning
2019-02-11 10:18:27 -06:00
//! of state machines. So, to account for changing source code, we build an FST
//! for each library (which is assumed to never change) and an FST for each Rust
2019-01-20 20:35:18 -06:00
//! file in the current workspace, and run a query against the union of all
2019-02-11 10:18:27 -06:00
//! those FSTs.
2020-02-06 08:08:31 -06:00
2019-01-08 13:33:36 -06:00
use std::{
2020-02-18 07:39:24 -06:00
cmp::Ordering,
fmt,
2019-01-08 13:33:36 -06:00
hash::{Hash, Hasher},
2019-01-25 08:20:52 -06:00
mem,
sync::Arc,
2019-01-08 13:33:36 -06:00
};
use fst::{self, Streamer};
use hir::db::DefDatabase;
2019-01-17 05:11:00 -06:00
use ra_db::{
salsa::{self, ParallelDatabase},
CrateId, FileId, SourceDatabaseExt, SourceRootId,
};
use ra_prof::profile;
use ra_syntax::{
ast::{self, NameOwner},
2019-10-05 09:03:03 -05:00
match_ast, AstNode, Parse, SmolStr, SourceFile,
SyntaxKind::{self, *},
SyntaxNode, SyntaxNodePtr, TextRange, WalkEvent,
2019-01-17 05:11:00 -06:00
};
2019-01-08 13:33:36 -06:00
use rayon::prelude::*;
2020-06-11 04:04:09 -05:00
use rustc_hash::{FxHashMap, FxHashSet};
2019-01-08 13:33:36 -06:00
2020-02-06 05:43:56 -06:00
use crate::RootDatabase;
2020-02-06 05:24:13 -06:00
#[derive(Debug)]
pub struct Query {
query: String,
lowercased: String,
only_types: bool,
libs: bool,
exact: bool,
limit: usize,
}
impl Query {
pub fn new(query: String) -> Query {
let lowercased = query.to_lowercase();
Query {
query,
lowercased,
only_types: false,
libs: false,
exact: false,
limit: usize::max_value(),
}
}
pub fn only_types(&mut self) {
self.only_types = true;
}
pub fn libs(&mut self) {
self.libs = true;
}
pub fn exact(&mut self) {
self.exact = true;
}
pub fn limit(&mut self, limit: usize) {
self.limit = limit
}
}
2019-01-08 13:33:36 -06:00
2019-01-25 14:27:16 -06:00
#[salsa::query_group(SymbolsDatabaseStorage)]
pub trait SymbolsDatabase: hir::db::HirDatabase + SourceDatabaseExt + ParallelDatabase {
2019-01-17 05:11:00 -06:00
fn file_symbols(&self, file_id: FileId) -> Arc<SymbolIndex>;
fn library_symbols(&self) -> Arc<FxHashMap<SourceRootId, SymbolIndex>>;
2019-01-26 02:17:05 -06:00
/// The set of "local" (that is, from the current workspace) roots.
/// Files in local roots are assumed to change frequently.
#[salsa::input]
2020-06-11 04:04:09 -05:00
fn local_roots(&self) -> Arc<FxHashSet<SourceRootId>>;
2019-01-26 02:17:05 -06:00
/// The set of roots for crates.io libraries.
/// Files in libraries are assumed to never change.
#[salsa::input]
2020-06-11 04:04:09 -05:00
fn library_roots(&self) -> Arc<FxHashSet<SourceRootId>>;
2019-01-08 13:33:36 -06:00
}
fn library_symbols(
db: &(impl SymbolsDatabase + ParallelDatabase),
) -> Arc<FxHashMap<SourceRootId, SymbolIndex>> {
let _p = profile("library_symbols");
let roots = db.library_roots();
let res = roots
.iter()
.map(|&root_id| {
let root = db.source_root(root_id);
let files = root
2020-06-11 04:04:09 -05:00
.iter()
.map(|it| (it, SourceDatabaseExt::file_text(db, it)))
.collect::<Vec<_>>();
let symbol_index = SymbolIndex::for_files(
files.into_par_iter().map(|(file, text)| (file, SourceFile::parse(&text))),
);
(root_id, symbol_index)
})
.collect();
Arc::new(res)
}
2019-01-15 09:19:09 -06:00
fn file_symbols(db: &impl SymbolsDatabase, file_id: FileId) -> Arc<SymbolIndex> {
2019-01-15 06:45:48 -06:00
db.check_canceled();
let parse = db.parse(file_id);
2019-07-19 04:56:47 -05:00
let symbols = source_file_to_file_symbols(&parse.tree(), file_id);
2019-01-08 13:33:36 -06:00
2019-03-23 02:53:48 -05:00
// FIXME: add macros here
2019-01-08 13:33:36 -06:00
2019-01-15 09:19:09 -06:00
Arc::new(SymbolIndex::new(symbols))
2019-01-08 13:33:36 -06:00
}
/// Need to wrap Snapshot to provide `Clone` impl for `map_with`
struct Snap<DB>(DB);
impl<DB: ParallelDatabase> Clone for Snap<salsa::Snapshot<DB>> {
fn clone(&self) -> Snap<salsa::Snapshot<DB>> {
Snap(self.0.snapshot())
}
}
2020-05-30 18:54:54 -05:00
// Feature: Workspace Symbol
//
// Uses fuzzy-search to find types, modules and functions by name across your
// project and dependencies. This is **the** most useful feature, which improves code
// navigation tremendously. It mostly works on top of the built-in LSP
// functionality, however `#` and `*` symbols can be used to narrow down the
// search. Specifically,
//
// - `Foo` searches for `Foo` type in the current workspace
// - `foo#` searches for `foo` function in the current workspace
// - `Foo*` searches for `Foo` type among dependencies, including `stdlib`
// - `foo#*` searches for `foo` function among dependencies
//
// That is, `#` switches from "types" to all symbols, `*` switches from the current
// workspace to dependencies.
//
// |===
// | Editor | Shortcut
//
// | VS Code | kbd:[Ctrl+T]
// |===
2020-02-06 05:43:56 -06:00
pub fn world_symbols(db: &RootDatabase, query: Query) -> Vec<FileSymbol> {
2020-06-08 06:00:31 -05:00
let _p = ra_prof::profile("world_symbols").detail(|| query.query.clone());
let tmp1;
let tmp2;
let buf: Vec<&SymbolIndex> = if query.libs {
tmp1 = db.library_symbols();
tmp1.values().collect()
2019-01-08 13:33:36 -06:00
} else {
let mut files = Vec::new();
for &root in db.local_roots().iter() {
let sr = db.source_root(root);
2020-06-11 04:04:09 -05:00
files.extend(sr.iter())
2019-01-08 13:33:36 -06:00
}
let snap = Snap(db.snapshot());
tmp2 = files
.par_iter()
.map_with(snap, |db, &file_id| db.0.file_symbols(file_id))
.collect::<Vec<_>>();
tmp2.iter().map(|it| &**it).collect()
2019-01-08 13:33:36 -06:00
};
2019-01-15 09:19:09 -06:00
query.search(&buf)
2019-01-08 13:33:36 -06:00
}
pub fn crate_symbols(db: &RootDatabase, krate: CrateId, query: Query) -> Vec<FileSymbol> {
2020-06-11 05:03:08 -05:00
// FIXME(#4842): This now depends on CrateDefMap, why not build the entire symbol index from
// that instead?
let def_map = db.crate_def_map(krate);
let mut files = Vec::new();
let mut modules = vec![def_map.root];
while let Some(module) = modules.pop() {
let data = &def_map[module];
files.extend(data.origin.file_id());
modules.extend(data.children.values());
}
let snap = Snap(db.snapshot());
let buf = files
.par_iter()
.map_with(snap, |db, &file_id| db.0.file_symbols(file_id))
.collect::<Vec<_>>();
let buf = buf.iter().map(|it| &**it).collect::<Vec<_>>();
query.search(&buf)
}
2020-02-06 05:43:56 -06:00
pub fn index_resolve(db: &RootDatabase, name_ref: &ast::NameRef) -> Vec<FileSymbol> {
2019-02-08 05:09:57 -06:00
let name = name_ref.text();
let mut query = Query::new(name.to_string());
query.exact();
query.limit(4);
2020-02-06 05:22:35 -06:00
world_symbols(db, query)
2019-02-08 05:09:57 -06:00
}
2019-03-05 08:09:48 -06:00
#[derive(Default)]
2020-02-06 05:43:56 -06:00
pub struct SymbolIndex {
2019-01-08 13:33:36 -06:00
symbols: Vec<FileSymbol>,
2020-03-10 08:55:23 -05:00
map: fst::Map<Vec<u8>>,
2019-01-08 13:33:36 -06:00
}
2019-03-05 08:09:48 -06:00
impl fmt::Debug for SymbolIndex {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
f.debug_struct("SymbolIndex").field("n_symbols", &self.symbols.len()).finish()
}
}
2019-01-08 13:33:36 -06:00
impl PartialEq for SymbolIndex {
fn eq(&self, other: &SymbolIndex) -> bool {
self.symbols == other.symbols
}
}
impl Eq for SymbolIndex {}
impl Hash for SymbolIndex {
fn hash<H: Hasher>(&self, hasher: &mut H) {
self.symbols.hash(hasher)
}
}
impl SymbolIndex {
fn new(mut symbols: Vec<FileSymbol>) -> SymbolIndex {
2020-02-18 07:39:24 -06:00
fn cmp(lhs: &FileSymbol, rhs: &FileSymbol) -> Ordering {
let lhs_chars = lhs.name.chars().map(|c| c.to_ascii_lowercase());
let rhs_chars = rhs.name.chars().map(|c| c.to_ascii_lowercase());
lhs_chars.cmp(rhs_chars)
2019-01-08 13:33:36 -06:00
}
2020-02-18 07:39:24 -06:00
symbols.par_sort_by(cmp);
let mut builder = fst::MapBuilder::memory();
let mut last_batch_start = 0;
for idx in 0..symbols.len() {
2020-02-18 07:39:24 -06:00
if let Some(next_symbol) = symbols.get(idx + 1) {
if cmp(&symbols[last_batch_start], next_symbol) == Ordering::Equal {
continue;
}
}
let start = last_batch_start;
let end = idx + 1;
last_batch_start = end;
2020-02-18 07:39:24 -06:00
let key = symbols[start].name.as_str().to_ascii_lowercase();
let value = SymbolIndex::range_to_map_value(start, end);
builder.insert(key, value).unwrap();
}
2020-03-10 08:55:23 -05:00
let map = fst::Map::new(builder.into_inner().unwrap()).unwrap();
2019-01-08 13:33:36 -06:00
SymbolIndex { symbols, map }
}
2020-02-06 05:43:56 -06:00
pub fn len(&self) -> usize {
2019-01-08 13:33:36 -06:00
self.symbols.len()
}
2020-02-06 05:43:56 -06:00
pub fn memory_size(&self) -> usize {
2019-01-25 12:10:28 -06:00
self.map.as_fst().size() + self.symbols.len() * mem::size_of::<FileSymbol>()
2019-01-25 08:20:52 -06:00
}
2019-07-18 14:29:20 -05:00
pub(crate) fn for_files(
files: impl ParallelIterator<Item = (FileId, Parse<ast::SourceFile>)>,
) -> SymbolIndex {
2019-01-08 13:33:36 -06:00
let symbols = files
2019-07-19 04:56:47 -05:00
.flat_map(|(file_id, file)| source_file_to_file_symbols(&file.tree(), file_id))
2019-01-08 13:33:36 -06:00
.collect::<Vec<_>>();
SymbolIndex::new(symbols)
}
fn range_to_map_value(start: usize, end: usize) -> u64 {
debug_assert![start <= (std::u32::MAX as usize)];
debug_assert![end <= (std::u32::MAX as usize)];
2019-04-23 15:19:45 -05:00
((start as u64) << 32) | end as u64
}
fn map_value_to_range(value: u64) -> (usize, usize) {
let end = value as u32 as usize;
let start = (value >> 32) as usize;
(start, end)
}
2019-01-08 13:33:36 -06:00
}
impl Query {
pub(crate) fn search(self, indices: &[&SymbolIndex]) -> Vec<FileSymbol> {
2019-01-08 13:33:36 -06:00
let mut op = fst::map::OpBuilder::new();
for file_symbols in indices.iter() {
let automaton = fst::automaton::Subsequence::new(&self.lowercased);
op = op.add(file_symbols.map.search(automaton))
}
let mut stream = op.union();
let mut res = Vec::new();
while let Some((_, indexed_values)) = stream.next() {
for indexed_value in indexed_values {
let symbol_index = &indices[indexed_value.index];
let (start, end) = SymbolIndex::map_value_to_range(indexed_value.value);
for symbol in &symbol_index.symbols[start..end] {
if self.only_types && !is_type(symbol.kind) {
continue;
}
if self.exact && symbol.name != self.query {
continue;
}
2020-06-08 06:00:53 -05:00
2020-06-10 05:37:00 -05:00
res.push(symbol.clone());
2020-06-08 06:00:53 -05:00
if res.len() >= self.limit {
return res;
}
2019-01-08 13:33:36 -06:00
}
}
}
res
}
}
fn is_type(kind: SyntaxKind) -> bool {
match kind {
STRUCT_DEF | ENUM_DEF | TRAIT_DEF | TYPE_ALIAS_DEF => true,
2019-01-08 13:33:36 -06:00
_ => false,
}
}
/// The actual data that is stored in the index. It should be as compact as
/// possible.
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
2020-02-06 05:43:56 -06:00
pub struct FileSymbol {
pub file_id: FileId,
pub name: SmolStr,
pub kind: SyntaxKind,
pub range: TextRange,
2020-02-06 05:43:56 -06:00
pub ptr: SyntaxNodePtr,
pub name_range: Option<TextRange>,
pub container_name: Option<SmolStr>,
}
fn source_file_to_file_symbols(source_file: &SourceFile, file_id: FileId) -> Vec<FileSymbol> {
let mut symbols = Vec::new();
let mut stack = Vec::new();
for event in source_file.syntax().preorder() {
match event {
WalkEvent::Enter(node) => {
2019-07-19 04:56:47 -05:00
if let Some(mut symbol) = to_file_symbol(&node, file_id) {
2019-02-13 09:42:15 -06:00
symbol.container_name = stack.last().cloned();
stack.push(symbol.name.clone());
symbols.push(symbol);
}
}
WalkEvent::Leave(node) => {
2019-07-19 04:56:47 -05:00
if to_symbol(&node).is_some() {
stack.pop();
}
}
}
}
symbols
2019-01-08 13:33:36 -06:00
}
fn to_symbol(node: &SyntaxNode) -> Option<(SmolStr, SyntaxNodePtr, TextRange)> {
2019-07-19 04:56:47 -05:00
fn decl<N: NameOwner>(node: N) -> Option<(SmolStr, SyntaxNodePtr, TextRange)> {
let name = node.name()?;
2019-07-20 04:58:27 -05:00
let name_range = name.syntax().text_range();
let name = name.text().clone();
2019-01-23 08:37:10 -06:00
let ptr = SyntaxNodePtr::new(node.syntax());
Some((name, ptr, name_range))
2019-01-08 13:33:36 -06:00
}
2019-10-05 09:03:03 -05:00
match_ast! {
match node {
2020-04-06 09:21:33 -05:00
ast::FnDef(it) => decl(it),
ast::StructDef(it) => decl(it),
ast::EnumDef(it) => decl(it),
ast::TraitDef(it) => decl(it),
ast::Module(it) => decl(it),
ast::TypeAliasDef(it) => decl(it),
ast::ConstDef(it) => decl(it),
ast::StaticDef(it) => decl(it),
2020-03-22 02:00:44 -05:00
ast::MacroCall(it) => {
if it.is_macro_rules().is_some() {
decl(it)
} else {
None
}
},
2019-10-05 09:03:03 -05:00
_ => None,
}
}
2019-01-08 13:33:36 -06:00
}
fn to_file_symbol(node: &SyntaxNode, file_id: FileId) -> Option<FileSymbol> {
to_symbol(node).map(move |(name, ptr, name_range)| FileSymbol {
name,
kind: node.kind(),
range: node.text_range(),
ptr,
file_id,
name_range: Some(name_range),
container_name: None,
})
}