Bugfixes

[Dojdik]

Bug Fixes for UI Freeze Issues

This document describes the critical bugs that were causing UI freezes and the fixes applied.

Summary of Changes

1. tree-sitter.ts - File Size Limits and Parse Timeout

Problem:

• No file size limit before parsing, causing blocks on minified/generated files (10MB+)
• Synchronous parser.parse() could block the event loop indefinitely on large files
• Recursive functions (serializeShape, countNodes, collectHashableNodes) had excessive depth limits

Fixes:

• Added 1MB file size limit in parseFile() - files larger than this are skipped
• Wrapped parser.parse() in a 5-second timeout using Promise.race()
• Reduced recursion depth from 40 to 20 in serializeShape() and countNodes()
• Reduced recursion depth from 10 to 5 in collectHashableNodes()
• Added MAX_CHILDREN = 50 limit to prevent exponential blowup in recursive functions
• Added MAX_RESULTS = 200 limit in collectHashableNodes() to cap memory usage
• Fixed memory leak in findImports() by wrapping re-export query in try/finally to ensure tree.delete() is always called

2. regex.ts - String/Comment Awareness and Size Limits

Problem:

• extractBraceScope() counted braces inside strings and comments, causing incorrect scope extraction
• No size limits on file processing, causing blocks on large files

Fixes:

• Completely rewrote extractBraceScope() to track string state (", ', `) and comment state (// and /* */)
• Braces inside strings and comments are now correctly ignored
• Added 1MB file size limit to findSymbols(), findImports(), and findExports()
• Added 50k line limit to prevent processing extremely large files

3. router.ts - Reduced Timeouts and Directory Scan Limits

Problem:

• 30-second operation timeout was too long for UI responsiveness
• Synchronous readdirSync BFS in detectAllLanguages() and findProbeFile() could scan thousands of directories

Fixes:

• Reduced operation timeout from 30s to 10s in executeWithFallbackTracked()
• Reduced detectAllLanguages() limits:
  • MAX_DEPTH: 8→6 (JVM), 3→2 (others)
  • MAX_DIRS: 500→200 (JVM), 100→50 (others)
• Reduced findProbeFile() limits:
  • MAX_DEPTH: 10→8 (JVM), 4→3 (others)
  • MAX_DIRS: 500→300 (JVM), 200→100 (others)

4. ts-morph.ts - File Size Limits

Problem:

• No file size limit in getSourceFile(), allowing very large files to be loaded into memory
• Large files could cause getPreEmitDiagnostics() to block for minutes

Fixes:

• Added 1MB file size check in getSourceFile() using statSync()
• Files larger than 1MB are now skipped to prevent blocking

Impact

These fixes address the root causes of UI freezes:

1. Large file parsing - Files > 1MB are now skipped entirely, preventing multi-second blocks
2. Parse timeouts - Tree-sitter parsing is capped at 5 seconds, preventing indefinite blocking
3. Recursion depth - Reduced limits prevent stack overflow and excessive computation on deeply nested code
4. Memory leaks - Fixed tree deletion in error paths prevents memory buildup
5. Directory scanning - Reduced limits prevent scanning thousands of directories synchronously
6. String/comment handling - Correct brace counting prevents incorrect scope extraction
7. Operation timeouts - 10s timeout ensures UI remains responsive

Testing Recommendations

Test with:

• Large minified files (>1MB) - should be skipped
• Deeply nested code (>20 levels) - should not cause stack overflow
• Files with braces in strings/comments - should extract correct scopes
• Large monorepos (1000+ directories) - should not block during language detection
• Complex TypeScript projects - should not block during diagnostics

Performance Characteristics

After fixes:

• Single file operations: <5 seconds (with timeout)
• Directory scanning: <2 seconds (with reduced limits)
• Memory usage: Bounded by result limits (200 nodes, 50 children)
• UI responsiveness: Maintained with 10s operation timeout

Modified files

• src/core/intelligence/backends/regex.ts
• src/core/intelligence/backends/tree-sitter.ts
• src/core/intelligence/backends/ts-morph.ts
• src/core/intelligence/router.ts

[proxysoul]

Thanks @Dojdik — great diagnosis, this is fixed in e294b6c. 🙏

Landed as proposed: the 1MB file caps in all three backends (the big win — a 4.5MB minified bundle went from a ~118s synchronous block in ts-morph to an 87ms skip), the findImports tree leak (plus the same leak in getFileOutline), and string/comment-aware brace matching in extractBraceScope.

Two adjustments after verification: the parse timeout uses tree-sitter's native progressCallback with a 5s deadline, since Promise.race can't preempt a synchronous WASM parse — and the router keeps its BFS depth limits (they're load-bearing for JVM project detection) but now yields to the event loop during scans, with per-op timeouts (10s default, 30s for project-wide ops). Recursion depths were left intact to preserve clone-detection hash semantics.

Verified end-to-end with headless A/B runs against a pre-fix worktree — startup scan, symbol reads, and oversized-file fallback all behave. Thanks again for the detailed report!