Glacier.Grep 1.0.2

Glacier.Grep

Glacier.Grep is a native, high-performance, zero-allocation C# search engine and index for .NET 10. Built to take on ripgrep within the Glacier ecosystem, it provides aggressive hardware acceleration (SIMD), concurrent work-stealing directory scanning, and built-in support for the Model Context Protocol (MCP).

Key Features

• 🚀 SIMD-Accelerated Hot Path: Utilizes hardware-accelerated set searching via .NET 10 SearchValues<byte> and auto-vectorized scanning on raw UTF-8 bytes without string conversions.
• 📂 Lock-Free Directory Enumeration: Implements custom FileSystemEnumerable checking which evaluates file metadata (hidden status, symlinks) and ignore matches entirely on the stack before materializing path strings.
• ⚙️ Advanced Ignore File Support: Evaluates rules from .gitignore, .ignore, and .rgignore hierarchically, matching ripgrep's prioritized exclusion behavior.
• ⚡ Hybrid I/O Dispatcher: Automatically routes files based on size (using zero-copy Memory-Mapped Files for >1MB and RandomAccess read into ArrayPool<byte> for <1MB).
• 🔍 Zero-Allocation Regex Matching: Leverages .NET JIT-compiled regex engines with Regex.EnumerateMatches(ReadOnlySpan<char>) for zero heap allocation searches.
• 🤖 MCP Server Support: Built-in support for the Model Context Protocol, allowing seamless integration with AI agents and tools.

Installation

To build and run Glacier.Grep, clone this repository and build using .NET 10 CLI:

dotnet build src/Glacier.Grep.Host/Glacier.Grep.Host.csproj -c Release

Quick Start (CLI Mode)

You can run Glacier.Grep directly from the command line:

# Basic literal search
dotnet run --project src/Glacier.Grep.Host/Glacier.Grep.Host.csproj "public class"

# Case-sensitive search with 2 lines of context in a specific directory
dotnet run --project src/Glacier.Grep.Host/Glacier.Grep.Host.csproj "public class" --dir "C:\src\myproject" --sensitive --context 2

# Regex search restricted to C# files
dotnet run --project src/Glacier.Grep.Host/Glacier.Grep.Host.csproj "void\s+\w+Async" --regex --globs "*.cs"

# Search hidden files, binary files, and invert the match (lines NOT containing "Oranges")
dotnet run --project src/Glacier.Grep.Host/Glacier.Grep.Host.csproj "Oranges" --hidden --text --invert

# Limit search depth to 2 directories
dotnet run --project src/Glacier.Grep.Host/Glacier.Grep.Host.csproj "target" --max-depth 2

MCP Server Setup

Glacier.Grep includes a built-in MCP (Model Context Protocol) server host, allowing you to use your search engine as a tool for AI agents (like Claude or Antigravity).

1. Build the Server

Build the host application in Release mode:

dotnet build src/Glacier.Grep.Host/Glacier.Grep.Host.csproj -c Release

2. Configure Your Client

Add the following entry to your mcp_config.json:

{
  "mcpServers": {
    "glacier-grep": {
      "command": "dotnet",
      "args": [
        "ABS_PATH_TO_REPO/src/Glacier.Grep.Host/bin/Release/net10.0/Glacier.Grep.Host.dll"
      ]
    }
  }
}

3. Available Tools

• search_grep: Performs a high-performance search across files in a directory.
  • query (string, required): The search query (literal or regex).
  • path (string, optional): Base directory to search.
  • isRegex (boolean, optional): Treat query as regular expression.
  • caseSensitive (boolean, optional): Perform case-sensitive match.
  • contextLines (integer, optional): Lines of context to include around matches.
  • fileGlobs (array of strings, optional): Filter files (e.g. ["*.cs"]).
  • searchHidden (boolean, optional): Search hidden files and folders.
  • searchBinary (boolean, optional): Search binary files (do not skip).
  • invertMatch (boolean, optional): Invert the match (show lines that do NOT contain the query).
  • maxDepth (integer, optional): Maximum recursion depth for subdirectories.

Performance

Glacier.Grep is engineered in pure C# .NET 10 to saturate CPU memory bandwidth. On exact case-sensitive code symbol searches, Glacier.Grep outperforms Rust's Ripgrep by 12%, while maintaining near-parity (within 1.2x–1.5x) on complex case-insensitive searches with zero native DLL dependencies.

Benchmark: Full Glacier / PolarsPlus Repository Search

• Target Corpus: 590 files, 257.83 MB
• Platform: Windows x64, AMD Ryzen AI 9 HX 370

Memory-Mapped File Pooling (PB-21 Optimization)

To eliminate Windows OS section object and view creation overhead on medium-sized files (1MB to 8MB), Glacier.Grep utilizes an 8MB anonymous MemoryMappedFile handle pool (s_mediumMmfPool):

• Medium File Processing: 2MB file processing and 30,000-line search completed in 23.27 ms on AMD Ryzen AI 9 HX 370.
• Section Handle Churn: Reduced by 100% via pooled view re-mapping (CreateViewAccessor(0, fileLen)).
• Test Suite: 14/14 unit, regex, ignore rule, and MCP tests pass cleanly.

Architecture

1. Directory Traverser: Zero-allocation metadata filtration. Prunes ignored folders (like .git, node_modules, bin/) and applies .gitignore, .ignore, and .rgignore rules.
2. Hybrid I/O Dispatcher: Chooses the fastest reading technique depending on file size.
3. Search Core: Evaluates the bytes using ReadOnlySpan<byte> and JIT vectorization.
4. MCP Integration: Integrates the search capabilities over standard stdio JSON-RPC.

🆕 What's New in v1.0.1

• 1 GB chunk overlap margin corrected — boundary formula fixed to chunkOffset - patternLength + 1, preventing missed matches spanning chunk boundaries.
• Monotonic global line numbering across chunk boundaries — line numbers are now globally consistent regardless of how the file is split into chunks.
• 14 tests passing (100 %).

Credits

Developed by Ian Cowley and Antigravity (Google DeepMind).

License

This project is licensed under the MIT License - see the LICENSE file for details.