ParseZero

The High-Performance Zero-Allocation CSV Parser for .NET

ParseZero is a specialized CSV parsing library focusing on zero allocation. It uses Span<T> and Memory<T> to read data without creating strings until absolutely necessary, keeping GC pressure near zero.

Features

• 🚀 Zero Allocation - Uses Span<T>, Memory<T>, and ArrayPool<T> to minimize heap allocations
• ⚡ SIMD Acceleration - Hardware-accelerated delimiter scanning with AVX2/SSE2 on .NET 8+
• 📊 IDataReader Support - Plug directly into SqlBulkCopy.WriteToServer() for blazing-fast database inserts
• 🔄 Streaming - Process files of any size using System.IO.Pipelines
• 🎯 Multi-Targeted - Supports both .NET 8+ and .NET Framework 4.7.2+ (via .NET Standard 2.0)
• 🛡️ Robust - Handles quoted fields, escaped quotes, BOM, and mixed line endings

Installation

dotnet add package ParseZero

Quick Start

Basic Usage

using ParseZero;

// Simple row-by-row parsing
await foreach (var row in CsvReader.ReadAsync("data.csv"))
{
    int id = row[0].ParseInt32();
    string name = row[1].ToString();
    decimal price = row[2].ParseDecimal();
}

With Schema Mapping

using ParseZero;
using ParseZero.Schema;

public record Trade(int Id, string Symbol, decimal Price, DateTime Timestamp);

var schema = Schema.For<Trade>()
    .Column(t => t.Id)
    .Column(t => t.Symbol)
    .Column(t => t.Price, format: "C2")
    .Column(t => t.Timestamp, format: "yyyy-MM-dd HH:mm:ss");

await foreach (var trade in CsvReader.ReadAsync<Trade>("trades.csv", schema))
{
    Console.WriteLine($"{trade.Symbol}: {trade.Price}");
}

SqlBulkCopy Integration

using ParseZero;
using ParseZero.Data;
using Microsoft.Data.SqlClient;

await using var connection = new SqlConnection(connectionString);
await connection.OpenAsync();

await using var reader = CsvDataReader.Create("large-dataset.csv", options);

using var bulkCopy = new SqlBulkCopy(connection);
bulkCopy.DestinationTableName = "Trades";
await bulkCopy.WriteToServerAsync(reader);

Configuration Options

var options = new CsvOptions
{
    Delimiter = ',',
    HasHeader = true,
    Encoding = Encoding.UTF8,
    BufferSize = 4096,
    MaxLineLength = 64 * 1024,  // DoS protection
    TrimFields = true,
    AllowQuotedFields = true
};

await foreach (var row in CsvReader.ReadAsync("data.csv", options))
{
    // Process rows
}

Performance

ParseZero achieves zero allocation by:

1. Buffer Pooling - Reuses byte arrays from ArrayPool<T>
2. Span Slicing - Returns ReadOnlySpan<char> slices instead of allocating strings
3. SIMD Scanning - Uses AVX2/SSE2 to scan 32 bytes at a time for delimiters
4. Pipelines - Streams data through System.IO.Pipelines for optimal I/O

Benchmark Results

Parsing 100,000 rows (10 columns each) on .NET 8.0 with AVX2:

Method	Mean	Allocated	Alloc Ratio
ParseZero (ForEach)	22.1 ms	2.9 MB	0.27x
ParseZero (File)	23.1 ms	2.9 MB	0.27x
Sylvan.Data.Csv (File)	25.8 ms	4.3 MB	0.42x
Sylvan.Data.Csv (Stream)	38.6 ms	15.4 MB	1.51x
ParseZero (Stream)	67.4 ms	10.2 MB	1.00x (baseline)

Key takeaways:

• ParseZero's file-based parsing is 10% faster than Sylvan with 33% less memory
• The FieldCountOnly mode achieves near-zero allocation (280 bytes for 100K rows)
• SIMD-accelerated delimiter scanning on .NET 8+ provides additional throughput

Benchmark: BenchmarkDotNet, Intel Core i7-10870H, .NET 8.0, Release build

Supported Encodings

• UTF-8 (with and without BOM)
• UTF-16 LE/BE (with BOM detection)
• ISO-8859-1 (Latin-1)
• Windows-1252

Target Frameworks

Framework	SIMD Support
.NET 8.0+	✅ AVX2/SSE2
.NET Standard 2.0	❌ Scalar only
.NET Framework 4.7.2+	❌ Scalar only

License

MIT License - see LICENSE for details.

Contributing

Contributions are welcome! Please read our Contributing Guide for details.