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

• The fastest CSV parser in .NET
• Sylvan Benchmarks
• Fastest CSV parser in C sharp

Currently there are parsers for 2 record formats:

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

Custom Converters

1. Readers
   • Default Type Convert *
   • Custom Property Convert *
2. Writers
   • Default Type Convert
   • Custom Property Convert

*ㅤyou 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 featur