Skip to content

leandromoh/RecordParser

Repository files navigation

Nuget GitHub Workflow Status (branch) GitHub GitHub code size in bytes

RecordParser - Simple, Fast, GC friendly & Extensible

RecordParser is a expression tree based parser that helps you to write maintainable parsers with high-performance and zero allocations, thanks to Span type. It makes easier for developers to do parsing by automating non-relevant code, which allow you to focus on the essentials of mapping.

🏆 2nd place in The fastest CSV parser in .NET blog post

Even the focus of this library being data mapping to objects (classes, structs, etc), it got an excellent result in the blog benchmark which tested how fast libraries can transform a CSV row into an array of strings. We got 1st place by parsing a 1 million lines file in 826ms.

RecordParser is a Zero Allocation Writer/Reader Parser for .NET Core

  1. It supports .NET 6, 7, 8 and .NET Standard 2.1
  2. It supports to parse individual records as well as whole files
  3. It has minimal heap allocations because it does intense use of Span type, a .NET type designed to have high-performance and reduce memory allocations (see benchmark)
  4. It is even more performant because the relevant code is generated using expression trees, which once compiled is fast as handwriting code
  5. It supports parse for ANY type: classes, structs, records, arrays, tuples etc
  6. It supports to map values for properties, fields, indexers, etc.
  7. It does not do boxing for structs.
  8. It is flexible: you can choose the most convenient way to configure each of your parsers: indexed or sequential configuration
  9. It is extensible: you can totally customize your parsing with lambdas/delegates
  10. It is even more extensible because you can easily create extension methods that wraps custom mappings
  11. It is efficient: you can take advantage of multicore to use parallel processing and speed up parsing
  12. It is not intrusive: all mapping configuration is done outside of the mapped type. It keeps your classes with minimised dependencies and low coupling
  13. It provides clean API with familiar methods: Parse, TryParse and TryFormat
  14. It is easy configurated with a builder object, even programmatically, because does not require to define a class each time you want to define a parser
  15. Compliant with RFC 4180 standard

Benchmark

Libraries always say themselves have great perfomance, but how often they show you a benchmark comparing with other libraries? Check the benchmark page to see RecordParser comparisons. If you miss some, a PR is welcome.

Third Party Benchmarks

Currently there are parsers for 2 record formats:

  1. Fixed length, common in positional/flat files, e.g. financial services, mainframe use, etc
  2. Variable length, common in delimited files, e.g. CSV, TSV files, etc

Custom Converters

  1. Readers
  2. Writers

*ă…¤you can use a "string pool" (function that converts a ReadOnlySpan<char> to string) to avoid creating multiple instances of strings with same content. This optimization is useful when there are a lot of repeated string values present. In this scenario, it may reduce allocated memory and speed-up processing time.

Parsing Files

  1. Readers
  2. Writers

NOTE: MOST EXAMPLES USE TUPLES FOR SIMPLICITY. PARSER ACTUALLY WORKS FOR ANY TYPE (CLASSES, STRUCTS, RECORDS, ARRAYS, TUPLES, ETC)

Fixed Length Reader

There are 2 flavors for mapping: indexed or sequential.

Indexed is useful when you want to map columns by its position: start/length.

[Fact]
public void Given_value_using_standard_format_should_parse_without_extra_configuration()
{
    var reader = new FixedLengthReaderBuilder<(string Name, DateTime Birthday, decimal Money)>()
        .Map(x => x.Name, startIndex: 0, length: 11)
        .Map(x => x.Birthday, 12, 10)
        .Map(x => x.Money, 23, 7)
        .Build();

    var result = reader.Parse("foo bar baz 2020.05.23 0123.45");

    result.Should().BeEquivalentTo((Name: "foo bar baz",
                                    Birthday: new DateTime(2020, 05, 23),
                                    Money: 123.45M));
}

Sequential is useful when you want to map columns by its order, so you just need specify the lengths.

[Fact]
public void Given_value_using_standard_format_should_parse_without_extra_configuration()
{
    var reader = new FixedLengthReaderSequentialBuilder<(string Name, DateTime Birthday, decimal Money)>()
        .Map(x => x.Name, length: 11)
        .Skip(1)
        .Map(x => x.Birthday, 10)
        .Skip(1)
        .Map(x => x.Money, 7)
        .Build();

    var result = reader.Parse("foo bar baz 2020.05.23 0123.45");

    result.Should().BeEquivalentTo((Name: "foo bar baz",
                                    Birthday: new DateTime(2020, 05, 23),
                                    Money: 123.45M));
}

Variable Length Reader

There are 2 flavors for mapping: indexed or sequential.

Indexed is useful when you want to map columns by its indexes.

[Fact]
public void Given_value_using_standard_format_should_parse_without_extra_configuration()
{
    var reader = new VariableLengthReaderBuilder<(string Name, DateTime Birthday, decimal Money, Color Color)>()
        .Map(x => x.Name, indexColumn: 0)
        .Map(x => x.Birthday, 1)
        .Map(x => x.Money, 2)
        .Map(x => x.Color, 3)
        .Build(";");
  
    var result = reader.Parse("foo bar baz ; 2020.05.23 ; 0123.45; LightBlue");
  
    result.Should().BeEquivalentTo((Name: "foo bar baz",
                                    Birthday: new DateTime(2020, 05, 23),
                                    Money: 123.45M,
                                    Color: Color.LightBlue));
}

Sequential is useful when you want to map columns by its order.

[Fact]
public void Given_ignored_columns_and_value_using_standard_format_should_parse_without_extra_configuration()
{
    var reader = new VariableLengthReaderSequentialBuilder<(string Name, DateTime Birthday, decimal Money)>()
        .Map(x => x.Name)
        .Skip(1)
        .Map(x => x.Birthday)
        .Skip(2)
        .Map(x => x.Money)
        .Build(";");
  
    var result = reader.Parse("foo bar baz ; IGNORE; 2020.05.23 ; IGNORE ; IGNORE ; 0123.45");
  
    result.Should().BeEquivalentTo((Name: "foo bar baz",
                                    Birthday: new DateTime(2020, 05, 23),
                                    Money: 123.45M));
}

Default Type Convert - Reader

You can define default converters for some type if you has a custom format.
The following example defines all decimals values will be divided by 100 before assigning,
furthermore all dates being parsed on ddMMyyyy format.
This feature is avaible for both fixed and variable length.

[Fact]
public void Given_types_with_custom_format_should_allow_define_default_parser_for_type()
{
    var reader = new FixedLengthReaderBuilder<(decimal Balance, DateTime Date, decimal Debit)>()
        .Map(x => x.Balance, 0, 12)
        .Map(x => x.Date, 13, 8)
        .Map(x => x.Debit, 22, 6)
        .DefaultTypeConvert(value => decimal.Parse(value) / 100)
        .DefaultTypeConvert(value => DateTime.ParseExact(value, "ddMMyyyy", null))
        .Build();

    var result = reader.Parse("012345678901 23052020 012345");

    result.Should().BeEquivalentTo((Balance: 0123456789.01M,
                                    Date: new DateTime(2020, 05, 23),
                                    Debit: 123.45M));
}

Custom Property Convert - Reader

You can define a custom converter for field/property.
Custom converters have priority case a default type convert is defined.
This feature is avaible for both fixed and variable length.

[Fact]
public void Given_members_with_custom_format_should_use_custom_parser()
{
    var reader = new VariableLengthReaderBuilder<(int Age, int MotherAge, int FatherAge)>()
        .Map(x => x.Age, 0)
        .Map(x => x.MotherAge, 1, value => int.Parse(value) + 3)
        .Map(x => x.FatherAge, 2)
        .Build(";");

    var result = reader.Parse(" 15 ; 40 ; 50 ");

    result.Should().BeEquivalentTo((Age: 15,
                                    MotherAge: 43,
                                    FatherAge: 50));
}

Nested Properties Mapping - Reader

Just like a regular property, you can also configure nested properties mapping.
The nested objects are created only if it was mapped, which avoids stack overflow problems.
This feature is avaible for both fixed and variable length.

[Fact]
public void Given_nested_mapped_property_should_create_nested_instance_to_parse()
{
    var reader = new VariableLengthReaderBuilder<Person>()
        .Map(x => x.BirthDay, 0)
        .Map(x => x.Name, 1)
        .Map(x => x.Mother.BirthDay, 2)
        .Map(x => x.Mother.Name, 3)
        .Build(";");

    var result = reader.Parse("2020.05.23 ; son name ; 1980.01.15 ; mother name");

    result.Should().BeEquivalentTo(new Person
    {
        BirthDay = new DateTime(2020, 05, 23),
        Name = "son name",
        Mother = new Person
        {
            BirthDay = new DateTime(1980, 01, 15),
            Name = "mother name",
        }
    });
}

Fixed Length Writer

There are 2 flavors for mapping: indexed or sequential.

Both indexed and sequential builders accept the following optional parameters in Map methods:

  • format
  • padding direction
  • padding character

Indexed is useful when you want to map columns by its position: start/length.

[Fact]
public void Given_value_using_standard_format_should_parse_without_extra_configuration()
{
    // Arrange

    var writer = new FixedLengthWriterBuilder<(string Name, DateTime Birthday, decimal Money)>()
        .Map(x => x.Name, startIndex: 0, length: 12)
        .Map(x => x.Birthday, 12, 11, "yyyy.MM.dd", paddingChar: ' ')
        .Map(x => x.Money, 23, 7, precision: 2)
        .Build();

    var instance = (Name: "foo bar baz",
                    Birthday: new DateTime(2020, 05, 23),
                    Money: 01234.567M);

    // create buffer with 50 positions, all set to white space by default
    Span<char> destination = Enumerable.Repeat(element: ' ', count: 50).ToArray();

    // Act

    var success = writer.TryFormat(instance, destination, out var charsWritten);

    // Assert

    success.Should().BeTrue();

    var result = destination.Slice(0, charsWritten);

    result.Should().Be("foo bar baz 2020.05.23 0123456");
}

Sequential is useful when you want to map columns by its order, so you just need specify the lengths.

[Fact]
public void Given_value_using_standard_format_should_parse_without_extra_configuration()
{
    // Arrange

    var writer = new FixedLengthWriterSequentialBuilder<(string Name, DateTime Birthday, decimal Money)>()
        .Map(x => x.Name, length: 11)
        .Skip(1)
        .Map(x => x.Birthday, 10, "yyyy.MM.dd")
        .Skip(1)
        .Map(x => x.Money, 7, precision: 2)
        .Build();

    var instance = (Name: "foo bar baz",
                    Birthday: new DateTime(2020, 05, 23),
                    Money: 01234.567M);

    // create buffer with 50 positions, all set to white space by default
    Span<char> destination = Enumerable.Repeat(element: ' ', count: 50).ToArray();

    // Act

    var success = writer.TryFormat(instance, destination, out var charsWritten);

    // Assert

    success.Should().BeTrue();

    var result = destination.Slice(0, charsWritten);

    result.Should().Be("foo bar baz 2020.05.23 0123456");
}

Variable Length Writer

There are 2 flavors for mapping: indexed or sequential.

Both indexed and sequential builders accept the format optional parameter in Map method.

Indexed is useful when you want to map columns by its indexes.

[Fact]
public void Given_value_using_standard_format_should_parse_without_extra_configuration()
{
    // Arrange 

    var writer = new VariableLengthWriterBuilder<(string Name, DateTime Birthday, decimal Money, Color Color)>()
        .Map(x => x.Name, indexColumn: 0)
        .Map(x => x.Birthday, 1, "yyyy.MM.dd")
        .Map(x => x.Money, 2)
        .Map(x => x.Color, 3)
        .Build(" ; ");

    var instance = ("foo bar baz", new DateTime(2020, 05, 23), 0123.45M, Color.LightBlue);

    Span<char> destination = new char[100];

    // Act

    var success = writer.TryFormat(instance, destination, out var charsWritten);

    // Assert

    success.Should().BeTrue();

    var result = destination.Slice(0, charsWritten);

    result.Should().Be("foo bar baz ; 2020.05.23 ; 123.45 ; LightBlue");
}

Sequential is useful when you want to map columns by its order.

[Fact]
public void Given_value_using_standard_format_should_parse_without_extra_configuration()
{
    // Arrange 

    var writer = new VariableLengthWriterSequentialBuilder<(string Name, DateTime Birthday, decimal Money)>()
        .Map(x => x.Name)
        .Skip(1)
        .Map(x => x.Birthday, "yyyy.MM.dd")
        .Map(x => x.Money)
        .Build(" ; ");

    var instance = ("foo bar baz", new DateTime(2020, 05, 23), 0123.45M);

    Span<char> destination = new char[100];

    // Act

    var success = writer.TryFormat(instance, destination, out var charsWritten);

    // Assert

    success.Should().BeTrue();

    var result = destination.Slice(0, charsWritten);

    result.Should().Be("foo bar baz ;  ; 2020.05.23 ; 123.45");
}

Default Type Convert - Writer

You can define default converters for some type if you has a custom format.
The following example defines all decimals values will be multiplied by 100 before writing (precision 2),
furthermore all dates being written on ddMMyyyy format.
This feature is avaible for both fixed and variable length.

[Fact]
public void Given_types_with_custom_format_should_allow_define_default_parser_for_type()
{
    // Arrange

    var writer = new FixedLengthWriterBuilder<(decimal Balance, DateTime Date, decimal Debit)>()
        .Map(x => x.Balance, 0, 12, padding: Padding.Left, paddingChar: '0')
        .Map(x => x.Date, 13, 8)
        .Map(x => x.Debit, 22, 6, padding: Padding.Left, paddingChar: '0')
        .DefaultTypeConvert<decimal>((span, value) => (((long)(value * 100)).TryFormat(span, out var written), written))
        .DefaultTypeConvert<DateTime>((span, value) => (value.TryFormat(span, out var written, "ddMMyyyy"), written))
        .Build();

    var instance = (Balance: 123456789.01M,
                    Date: new DateTime(2020, 05, 23),
                    Debit: 123.45M);

    // create buffer with 50 positions, all set to white space by default
    Span<char> destination = Enumerable.Repeat(element: ' ', count: 50).ToArray();

    // Act

    var success = writer.TryFormat(instance, destination, out var charsWritten);

    // Assert

    success.Should().BeTrue();

    var result = destination.Slice(0, charsWritten);

    result.Should().Be("012345678901 23052020 012345");
}

Custom Property Convert - Writer

You can define a custom converter for field/property.
Custom converters have priority case a default type convert is defined.
This feature is avaible for both fixed and variable length.

[Fact]
public void Given_specified_custom_parser_for_member_should_have_priority_over_custom_parser_for_type()
{
    // Assert

    var writer = new VariableLengthWriterBuilder<(int Age, int MotherAge, int FatherAge)>()
        .Map(x => x.Age, 0)
        .Map(x => x.MotherAge, 1, (span, value) => ((value + 2).TryFormat(span, out var written), written))
        .Map(x => x.FatherAge, 2)
        .Build(" ; ");

    var instance = (Age: 15,
                    MotherAge: 40,
                    FatherAge: 50);

    Span<char> destination = new char[50];

    // Act

    var success = writer.TryFormat(instance, destination, out var charsWritten);

    // Assert

    success.Should().BeTrue();

    var result = destination.Slice(0, charsWritten);

    result.Should().Be("15 ; 42 ; 50");
}

File Processing - Read

While version 1 of the library allows to parse individual records, version 2 introduced features to read/write records directly from/to files. This functionality bridged the gap between the existing readers/writers and TextReader/TextWriter.

One awesome feature that makes RecordParser innovative and special is the hability to process files using the power of the parallel programmimg! You can take advantage of multicore to use parallel processing and speed up reading and writing! You can disable the parallelism just by set the feature flag to false, thus using sequential processing.

Just import the namespace RecordParser.Extensions to see the extension methods to interact in TextReader and TextWriter.

File reading for FixedLengthReader

using System;
using RecordParser.Builders.Reader;
using RecordParser.Extensions;
using System.IO;

var fileContent = 
    """
    01 123-456-789 00033.5
    99 abc-def-ghk 00050.7
    """;

// I am using StringReader because in the example the content is a string
// but could be StreamReader or any TextReader
using TextReader textReader = new StringReader(fileContent);

var reader = new FixedLengthReaderSequentialBuilder<Record>()
    .Map(x => x.Foo, 2)
    .Skip(1)
    .Map(x => x.Bar, 11)
    .Skip(1)
    .Map(x => x.Qux, 7)
    .Build();

var readOptions = new FixedLengthReaderOptions<Record>
{
    Parser = reader.Parse,
    ParallelismOptions = new()
    {
        Enabled = true,
        EnsureOriginalOrdering = true,
        MaxDegreeOfParallelism = 4
    }
};

var records = textReader.ReadRecords(readOptions);

foreach (var r in records)
    Console.WriteLine(r);

public record class Record(int Foo, string Bar, decimal Qux);

File reading for VariableLengthReader

using System;
using RecordParser.Builders.Reader;
using RecordParser.Extensions;
using System.IO;

var fileContent = 
    """
    Foo,Bar,Qux
    1,123456789,33
    12,abc-def-ghk,50.7
    """;

// I am using StringReader because in the example the content is a string
// but could be StreamReader or any TextReader
using TextReader textReader = new StringReader(fileContent);

var reader = new VariableLengthReaderBuilder<Record>()
    .Map(x => x.Foo, 0)
    .Map(x => x.Bar, 1)
    .Map(x => x.Qux, 2)
    .Build(",");

var readOptions = new VariableLengthReaderOptions
{
    HasHeader = true,
    ContainsQuotedFields = false,
    ParallelismOptions = new()
    {
        Enabled = true,
        EnsureOriginalOrdering = true,
        MaxDegreeOfParallelism = 4
    }
};

var records = textReader.ReadRecords(reader, readOptions);

foreach (var r in records)
    Console.WriteLine(r);

public record class Record(int Foo, string Bar, decimal Qux);

File reading for VariableLength Raw

Note: only recommended when user need to receive each field as string. Other methods that does not force string uses ReadOnlySpan<char>, which speed up processing and reduces unnecessary allocations.

using System;
using RecordParser.Extensions;
using System.IO;

var fileContent = """
    A,B,C,D
    1,2,3,X
    5,6,7,Y
    9,10,11,Z
    """;

// I am using StringReader because in the example the content is a string
// but could be StreamReader or any TextReader
using TextReader textReader = new StringReader(fileContent);

var readOptions = new VariableLengthReaderRawOptions
{
    HasHeader = true,
    ContainsQuotedFields = false,
    ColumnCount = 4,
    Separator = ",",
    ParallelismOptions = new()
    {
        Enabled = true,
        EnsureOriginalOrdering = true,
        MaxDegreeOfParallelism = 4
    }
};

// getField is a callback of type Func<int, string>,
// it receives the index column and returns its content as string
var records = textReader.ReadRecordsRaw(readOptions, getField =>
{
    var record = new
    {
        A = getField(0),
        B = getField(1),
        C = getField(2),
        D = getField(3)
    };
    return record;
});

foreach (var r in records)
    Console.WriteLine(r);

File Processing - Write

This feature is agnostic to builder type since it just receives the delegate of TryFormat method.
So you can use it with instances of both FixedLengthWriter and VariableLengthWriter.

using RecordParser.Builders.Writer;
using RecordParser.Extensions;
using System.IO;

var writer = new VariableLengthWriterBuilder<Record>()
    .Map(x => x.Foo, 0)
    .Map(x => x.Bar, 1)
    .Map(x => x.Qux, 2)
    .Build(",");

// I am using StreamWriter + MemoryStream in the example but could be any TextWriter
using var memory = new MemoryStream();
using TextWriter textWriter = new StreamWriter(memory);

var parallelOptions = new ParallelismOptions()
{
    Enabled = true,
    EnsureOriginalOrdering = true,
    MaxDegreeOfParallelism = 4,
};

var records = new []
{
    new Record(12, "123456789", 33),
    new Record(34, "abc-def-ghk", 50.7M)
};

textWriter.WriteRecords(records, writer.TryFormat, parallelOptions);
    
public record class Record(int Foo, string Bar, decimal Qux);