Replace LSP-based symbol search with tree-sitter symbol index - #61950
Open
gaojunran wants to merge 10 commits into
Open
Replace LSP-based symbol search with tree-sitter symbol index#61950gaojunran wants to merge 10 commits into
gaojunran wants to merge 10 commits into
Conversation
Introduce a new symbol_index crate that uses tree-sitter outline queries to extract and fuzzy-search symbols, replacing the LSP-based project symbols picker. symbol_index crate: - SymbolIndex with extract_symbols via tree-sitter outline queries - IndexSnapshot for zero-copy concurrent search - fuzzy_nucleo-based search with cancellation support - Batch update API for efficient bulk indexing SymbolIndexManager (project crate): - GPUI entity integrated into Project via lazy init - Subscribes to WorktreeStore events for incremental updates - Batch indexing with progress tracking - Fs trait for testability (FakeFs support) project_symbols picker: - Replaced LSP symbol search with symbol_index snapshot search - Search cancellation guards against stale results overwriting - Byte-accurate cursor positioning via Point (not PointUtf16)
- IndexedSymbol gains name_range for syntax highlight runs - render_match colors context as keyword and name by SymbolKind - PreviewSource::Path carries optional position for symbol highlighting - picker_preview derives MatchLocation from position after buffer loads
…ions - Lazy snapshot rebuild via dirty flag instead of per-batch O(n) clone (C1) - Parallel file processing with buffer_unordered for 8x faster indexing (C2) - Batch file removal with single retain pass (M1) - Eliminate triple string storage: Arc<str> name + context, compute display on demand (M2) - Single HashSet retain in update_files_batch instead of per-file O(n) (M3) - Pre-filter files by language before collecting into file list (M4) - Group files by extension and load each language grammar once (M5)
cla-bot
Bot
added
the
cla-signed
zed-community-bot
Bot
added
the
first contribution
@context captures appearing after @name in source (e.g. "(" and ")" in TypeScript function_declarations) were collected unordered and joined with spaces, producing "async ( ) mmdata" instead of "async function mmdata". Now extract context text from source between the first capture and the name node, preserving original token order.
Mirror the initial scan strategy in on_updated_entries: group changed files by extension, load each language grammar once, then process with buffer_unordered. Avoids per-file language loading on bulk changes like git checkout.
gaojunran
marked this pull request as ready for review
July 31, 2026 09:36
Replace context-only extraction with full source range from first to last relevant capture (@context, @name, @OPEN, @close). Display text now includes tokens after @name (e.g., "()" in TypeScript functions), producing "async function mmdata()" instead of "async function mmdata". Data model change: IndexedSymbol.context -> display_text + name_range, where name_range tracks the name's byte position within display_text.
Take raw source range directly and normalize whitespace as a whole, instead of splitting into before/after parts and rejoining with explicit spaces. This preserves original spacing like "name()" instead of "name ()".
Collect (byte_range, is_name) for each @context and @name capture, then build display text by concatenating captured text in order. A space is inserted only when there's a source gap between adjacent captures. This preserves "mmdata()" (no gap) and "async function" (gap), while excluding uncaptured tokens like "{" in destructuring. @OPEN and @close captures are excluded from display text, matching outline panel's next_outline_item behavior.
Contributor
Author
|
@maxbrunsfeld Any idea for this? This is a follow up for #45719 which you previously reviewed! Sorry for the noise :) |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Replace LSP-based symbol search (
workspace/symbol) with a tree-sitter symbol index that provides client-side fuzzy matching usingfuzzy_nucleo. This enables camelCase initialism queries (e.g.ibfo→initBookForOfficial), works without any LSP running, and indexes all languages in a single pass.Discussion: #61880
Follow-Up: Use the same tree-sitter index way in agent ui
@symbolsearch.Showcase
(For manually test you can use https://github.com/gaojunran/zed/actions/runs/30794496633):
This PR: Tree-sitter version (quicker index, can index before actually opening a file, can fuzzy search, has grammar highlight on candidates, just like outline panel)
2026-07-31.17.32.26.mov
Current version (index speed depends on LSP and is always slower, cannot index before opening a file, no grammar highlights on candidates, must fully match, cannot search in many markup languages)
2026-07-31.13.26.28.mov
Solution
New crate
crates/symbol_index/— Core index with no dependency onprojectorworktreecrates.SymbolIndexholds a flatVec<IndexedSymbol>withArc<str>fields.extract_symbols()parses source text with tree-sitter and runs theoutline.scmquery.IndexSnapshotprovidesArc<[...]>for lock-free background search. Snapshot rebuild is deferred via a dirty flag — O(n) per search, not per batch.crates/project/src/symbol_index_manager.rs— GPUI entity managing the indexing lifecycle. Initial scan processes files in parallel (buffer_unordered(num_cpus * 2)), pre-filters by language, groups by extension to load each grammar once. Incremental updates viaWorktreeStoreevents use the same strategy. Batch operations use singleHashSetretain passes.crates/project_symbols/src/project_symbols.rs— Rewritten picker delegate. Search runs on in-memoryIndexSnapshotwith no LSP round-trips. Cancel flag prevents stale results. Labels are syntax-highlighted: context in keyword color, name colored bySymbolKind. Fuzzy match positions are offset to align with display text.crates/picker/+crates/picker_preview/—PreviewSource::Pathgainsposition: Option<Point>for symbol highlighting. NewPreviewUpdate::from_path_with_position()constructor;from_path()kept for backward compatibility.picker_previewderivesMatchLocationfrom position, extending to end-of-line.Self-Review Checklist:
Tests
symbol_index(extraction, initialism matching, add/remove, sorted results, empty query, concurrent mutation)project_symbols(tree-sitter search, struct/enum/function matching)pickerandpicker_previewtests pass--no-deps --lib)Release Notes: