Skip to content

PERFORMANCE.md

Latest commit

 

History

History
126 lines (94 loc) · 3.57 KB

PERFORMANCE.md

File metadata and controls

126 lines (94 loc) · 3.57 KB
Raw

Performance Optimizations

What I Optimized

1. Non-Blocking Completion Provider

Before: Used async/await which blocked other completion providers
After: Returns a promise immediately, allowing other providers to show results first

2. Caching

Added: 2-second cache for completion items at the same position
Benefit: Repeated triggers (typing multiple dots) use cached results

3. Timeouts

Code Actions: 1 second timeout
Navigation Commands: 500ms timeout per provider
Benefit: Won't hang if language server is slow

4. Parallel Fetching

Before: Navigation commands fetched sequentially (slow)
After: All navigation providers fetched in parallel
Benefit: 4x faster for navigation commands

5. Cancellation Support

Added: Checks token.isCancellationRequested throughout
Benefit: Stops processing if user types more or closes completion

6. Smart Filtering

Added: Filters out lazy actions immediately
Benefit: Less processing, cleaner results

Performance Characteristics

Fast Path (Cached)

  • < 1ms - Returns cached results immediately

Normal Path (Not Cached)

  • Code Actions: Up to 1000ms (with timeout)
  • Navigation: Up to 500ms (with timeout)
  • Total: ~1.5 seconds max

Dart/Flutter Considerations

Dart language server can be slower, especially for:

  • Large projects
  • First completion after opening file
  • Complex widget trees

The timeouts ensure we never block longer than 1.5 seconds total.

User Experience

What Users See

  1. Type . → Other autocomplete appears immediately
  2. Wait ~1 second → Our commands appear below
  3. Type more → Cancels our fetch, no wasted work

Why This Works

  • Other completion providers aren't blocked
  • Normal autocomplete shows first
  • Our commands appear after (sorted with ~ prefix)
  • Users can start typing immediately

Configuration Options (Future)

Could add settings for:

  • commandCompletion.timeout - Adjust timeout (default: 1000ms)
  • commandCompletion.cacheTimeout - Adjust cache TTL (default: 2000ms)
  • commandCompletion.enableNavigation - Toggle navigation commands
  • commandCompletion.enableCodeActions - Toggle code actions

Monitoring Performance

Check Debug Console for:

Command completion triggered at position: ...
Found X dot(s)
Using cached completion items  ← Cache hit!
Found X code actions
Code action fetch timed out  ← Timeout triggered

If Still Slow

Dart/Flutter Specific

  1. Ensure Dart/Flutter extension is up to date
  2. Check if project analysis is complete
  3. Try smaller timeout: Reduce from 1000ms to 500ms

General

  1. Check if other extensions are slow too
  2. Disable other completion extensions temporarily
  3. Check CPU usage during completion

Technical Details

Cache Implementation

  • Map-based cache with position as key
  • 2-second TTL (Time To Live)
  • Max 50 entries (LRU-style cleanup)
  • Cleared on file changes (implicit via position change)

Timeout Implementation

  • Promise.race() between actual fetch and timeout
  • Returns empty array on timeout
  • Doesn't throw errors, just returns partial results

Parallel Fetching

  • Promise.all() for navigation commands
  • Each wrapped in Promise.race() with timeout
  • All 4 providers fetch simultaneously

Comparison

Before Optimization

Type . → Wait 3-5 seconds → All completions appear together

After Optimization

Type . → Normal autocomplete appears immediately
      → Our commands appear 1-1.5 seconds later (below)

This is the correct behavior - we don't block other providers!