Ataraxy-Labs/sem v0.8.0

on GitHub

Performance

The headline improvements:

• sem diff --patch: 61s → 410ms (149x faster)
• Graph resolution: O(n²) → O(n)
• 71K-file TS monorepo: DNF → 6.5s with cache

Under the hood:

• Streaming JSON output — results stream as they're computed instead of buffering in memory
• mimalloc allocator — significant win for allocation-heavy tree-sitter parsing
• Topology-only cache — caches dependency graph structure so repeated queries skip re-parsing unchanged files
• Content fingerprinting — skips re-analysis of files that haven't changed based on hash comparison
• Batched file fingerprint refresh — batched SQLite roundtrips instead of one-at-a-time
• Import table allocation reduction in bag-of-words resolver
• SQLite cache schema v4 with file_imports table for faster import resolution
• Partial topology cache for large impact runs — cold sem impact on 100K-file repos drops from 90s to 34s, warm runs under 2s
• Direct deps path for large impact runs
• Stabilized graph output for ambiguous calls

Consolidation

Major internal cleanup across the codebase:

• Consolidated identity matching, diff output filtering, diff path/patch parsing
• Consolidated cache and verify behavior
• Consolidated graph and scope resolution
• Consolidated TypeScript/JavaScript parser resolution
• Consolidated Swift extraction

Features

• LaTeX language support — regex-based parser for .tex, .latex, .cls, .sty files with preamble, command, section, and environment entity extraction

Fixes

• Skip binary and generated files in repo scans
• Handle cross-language file diffs
• Return MCP tool errors for bad inputs instead of generic failures
• MCP entity lookup failures now name the requested file and include candidate suggestions
• Stabilized sem graph --json output ordering
• Bounded sem log history scans
• Fixed subdirectory path resolution in CLI commands
• Portable format flag test across platforms