WIP: Implement unified palette with all 5 modes

[kathbigra]

Self-Review Checklist:

• I've reviewed my own diff for quality, security, and reliability
• Unsafe blocks (if any) have justifying comments
• The content is consistent with the UI/UX checklist
• Tests cover the new/changed behavior (39 tests passing)
• Performance impact has been considered and is acceptable

Closes #15896

Release Notes:

• Added unified palette that combines file finder, command palette, go-to-line, project symbols, and outline into a single interface with mode switching via prefixes (>, #, @, :)

---

Overview

This PR implements a unified command palette that consolidates 5 separate pickers (File Finder, Command Palette, Go To Line, Project Symbols, Outline) into a single modal with prefix-based mode switching.

Implementation Journey

Initial Approach

Started by attempting to reuse existing picker delegates, but encountered fundamental blockers:

• Existing delegates are tightly coupled to their specific picker types via Context<Picker<SpecificDelegate>>
• Private constructors and internal state made composition impossible
• Sub-picker approach failed due to type system constraints

Final Solution

Built a new unified delegate from scratch with:

• Single UnifiedPaletteDelegate handling all 5 modes
• Mode detection from query prefixes
• Async search with proper cancellation
• Shared match storage with mode-specific rendering

Features Implemented

Core Modes

File Finder (default)

• Opens with Cmd+P
• Fuzzy search with match highlighting
• File history (shows 10 most recent on empty query)
• Path parsing: file.rs:42:10 navigates to line/column
• File type icons
• Enter to open, Cmd+Enter for split pane

Command Palette (> prefix)

• Fuzzy search through available commands
• Keyboard shortcuts displayed
• Match highlighting shows why commands matched

Go To Line (: prefix)

• Type :42 to jump to line 42
• Works with active editor

Project Symbols (# prefix)

• LSP-based symbol search across project
• Two-line display: symbol name + file path with line number
• Path truncation for long paths (middle elision)
• Opens file and navigates to symbol

Outline (@ prefix)

• Searches symbols in current file
• Fuzzy matching with match highlighting
• Live preview: navigates as you arrow through results
• Single-line display format

Advanced Features

• Async search with cancellation: Prevents stale results from slow searches
• File history persistence: Stored in Workspace, survives palette close/reopen
• Race condition prevention: Cancel flags in all async search paths
• Split pane support: Cmd+Enter opens in adjacent pane
• Mode switching: Seamless transition between modes, cancels pending searches

Issues Discovered & Fixed

P0 Critical Issues (Post-Implementation)

1. History Persistence Bug

• Problem: file_history stored in delegate, wiped on every palette open
• Root cause: New delegate created each time in register()
• Fix: Moved file_history to Workspace with accessor methods
• Tests: 4 tests (empty query, tracking, limit, ordering)

2. Missing Cancel Flags in Symbols/Outline

• Problem: Async tasks could update stale/dismissed pickers
• Root cause: Only file/command searches had cancellation logic
• Fix: Added cancel_flag to search_project_symbols() and search_outline()
• Tests: 4 tests (rapid typing, mode switching, symbols, outline)

3. Outline Match Highlighting Missing

• Problem: Outline showed syntax highlighting but not fuzzy match positions
• Root cause: Discarded match positions from fuzzy search results
• Fix: Store positions in SymbolMatch, use HighlightedLabel for rendering
• Tests: Visual verification (existing tests cover functionality)

P1 UX Issues

4. Path Overflow

• Problem: Long file paths overflow UI, filename invisible
• Fix: Middle elision for paths > 60 chars (start...end)

Compromises & Deferred Work

Deferred to Future PRs

1. Symbol kind icons (ƒ, C, etc.)

   • Reason: No existing icon mapping in codebase
   • Would require design decision on icon set

2. Command descriptions ("Editor: Save File")

   • Reason: Requires extracting metadata from action registry
   • Non-trivial implementation, low priority

3. Selection reset optimization

   • Reason: Complex logic, low user impact
   • Current behavior is standard for pickers

Technical Decisions

Why not reuse existing pickers?

• Type system constraints made composition impossible
• Delegates expect specific Context<Picker<TheirDelegate>> types
• Private constructors prevented instantiation
• Building from scratch gave better control and cleaner code

Why store history in Workspace?

• Delegate is recreated on every palette open
• Workspace persists for entire session
• Natural location for cross-session state

Why cancel flags AND search IDs?

• Search IDs prevent out-of-order results
• Cancel flags stop unnecessary work early
• Both needed for robust async handling

Code Quality

Structure

• Refactored rendering: Split 110-line render_match into 5 focused functions (10-35 lines each)
• Proper error handling: No unwrap(), all errors propagated with ? or .log_err()
• Async safety: Cancel flags prevent race conditions
• Type safety: Strong typing throughout, no Any types

Test Coverage

39 tests passing covering:

• All 5 modes (basic functionality)
• Mode switching and cancellation (4 tests)
• File history persistence (4 tests)
• Match highlighting (2 tests)
• Edge cases (no editor, empty queries, special characters)
• Split pane navigation
• Path parsing with line/column

Performance

• Async search prevents UI blocking
• Cancel flags stop unnecessary work
• Fuzzy matching limited to 100 results per mode
• File history limited to 10 items

Files Modified

• crates/workspace/src/workspace.rs - Added file_history field and accessor methods
• crates/unified_palette/unified_palette.rs - Complete implementation (960 lines)
• crates/unified_palette/unified_palette_tests.rs - Test suite (1200+ lines)
• crates/unified_palette/Cargo.toml - Dependencies
• crates/zed/src/main.rs - Module initialization

Migration Notes

• Existing Cmd+P keybinding now opens unified palette
• All original functionality preserved
• No breaking changes to user workflows